Yet an enduring constraint remains: the tyranny of distance.

The time it takes to traverse features like a mountain range or a body of water has defined where we live, who we see, and what we do. Tackling peak hour traffic in today’s cars, or even a Tesla or Waymo is not meaningfully faster than in a Ford or GM from 100 years ago. In more rural settings, our vehicles remain bound to highways and the contours of the Earth. We have reached the limits of ground-based everyday travel.

We think the answer is to make door-to-door flying as normal as driving.

This will not meaningfully start with air taxis over urban skies. The $1B costs and many years to certify these aircraft have made that clear. Cities offer limited spaces for helicopters (or eVTOLs) to take off and land, at least for now.

Instead, true door-to-door flying experiences will start in more open spaces.

It starts at a rural property, with two people vertically taking off for sunset vistas before returning for dinner. It’s a quick hop from a mountain cabin to a remote trailhead for an unforgettable hike or fishing excursion. It’s island-hopping for a morning adventure or a trip to a winery or golf course.

Flying from your personal property at 120 mph (almost 200 kph) can be up to 5 times faster than a drive constrained by geography. You arrive faster and the flights themselves become as interesting as the destination. Vight‘s first generation eVTOL will enable all of these experiences.

This world is a lot closer than most people think. Enabling technologies now make these experiences possible; while the FAA’s new MOSAIC regulations allow for compelling personal-use electric vertical takeoff and landing vehicles (eVTOLs).

Electric propulsion, software and autonomous technologies are reshaping personal aviation. Multirotor electric propulsion offers safety, redundancy and noise profiles not possible in single-engine airplanes and helicopters. Software-defined controls dramatically lower the workload needed to fly, turning the pilot into a system supervisor. The controls are engineered with autonomy in mind, with minimal pilot inputs needed from takeoff to landing.

A credible regulatory path finally exists. MOSAIC’s world-first framework provides a clear path to simple, safe personal-use eVTOLs. It enables FAA oversight of factory‑built aircraft certification based on industry consensus standards. That means significantly lower cost and higher velocity development than possible before.

Our strategy is built for speed and scaling demand. We will deliver vehicles directly to private customers, starting with high net worth rural landowners. In the US alone, our analysis shows a market approaching a million people that already have the means and the land for a two-seat personal eVTOL. Globally, that number approximately doubles. Our products will delight this vast, underserved market.

This market will grow rapidly as Vight scales. We will relentlessly increase range, seating/payload and other capabilities. We will shrink the takeoff and landing footprint. Scale will drive costs lower, accelerating the decades-long cost curves already in place for electric and compute technologies.

Combined, that means our path ahead is clear:

1. Build a compelling, intelligent vehicle

2. Build vehicles capable of landing on more surfaces

3. Use the proceeds to build more affordable and longer range vehicles

4. Add more capabilities and form factors

5. Overcome the tyranny of distance

Over time, all light aircraft will converge on autonomous personal flight. Ours is the path to get there first, starting with thousands and eventually expanding thousand-fold. Of course, we’ll be partnering with regulators and sharing our safety data at every step along the way.

We know these volumes have never been achieved in aviation. But that’s because it’s never been possible before. Indeed, it’s an insanely difficult set of engineering problems. But now the technology is finally ready, the regulations have changed and the path to market is clear. We believe the world-class team we’re assembling will lead the charge.

This is how we soar above the tyranny of distance. This is how we make door-to-door flying as normal as driving.

I’m building Vight to tackle this. Follow our progress on X and LinkedIn.

I made this 130 mile capable VTOL drone in only 90 days. It can fly for 3 hours on a single charge. That would make it one of the longest range and endurance 3D printed VTOLs in the world.

This is the thing I'm most proud of building to date!

Reid Hoffman called my project “superagency in action” and compared it to the Wright Brothers’s work pioneering powered flight!

Before this project, I was a total CAD, 3D printing and aerodynamic modeling beginner. I had only built and flown one VTOL before. My CAD skills were restricted to extrusions of basic sketches. My Bambu A1 3D printer was only one month old, and I had never tried printing foaming PLA. I had never used any aerodynamic modeling tools to simulate lift-to-drag ratios or stability curves like Cm-alpha.

This video edit ended up shorter than I planned. Being my first Youtube video with significant post production effort, I underestimated the work required to make a longer in-depth video with voiceover, edited footage, etc.

Because of that, I did not discuss dozens of project challenges like how I chose the design parameters, learnt airframe CAD design, sourced components, improved foaming PLA print quality, in-depth power loss troubleshooting, and many more challenges this project faced.

If you want more details on my design and build process, let me know!

The VTOL is wrapped and ready for the 1500 mile trip. Flight testing will have to wait.

Why the South Bay? I'm on a quest to build eVTOLs for personal use. The South Bay has incomparable talent density to help me do this.

Most cracked aero, mechE, electrical and embedded system engineers working on eVTOLs stateside are at companies there. Tesla's engineering HQ is still minutes away in Palo Alto, not in Austin. Apple's consumer hardware engineers are close by in Cupertino. America's best robotics (various South Bay startups) and AI engineers (SF) are in the region.

As I get closer to starting an eVTOL company, it's a no brainer to be near this density of talent.

My wife and I visited the area two weeks ago, and were surprised at how livable it was and family friendly. The South Bay is worlds apart from the Gotham City that SF has become since we left there for Austin in 2021. In Sunnyvale and surrounding areas, we saw people leave personal belongings unsecured outside their front doors and leave food on the table at a restaurant to say hi to friends (would be stolen within seconds in SF). Crime overall is far lower than SF, and even lower than Austin.

We also saw less homeless people than in Austin, and way less than in SF. The only few tents we saw were pitched at Sunnyvale public library, yet there were barely any downtown or in Mountain View.

There's also the nature. Before the trip I had not spent much time in the South Bay. Driving around, up into the Santa Cruz mountains and the redwood forests, it's just beautiful. Having grown up near the ocean and in the southern hills of Adelaide in Australia, it was incredibly refreshing to see the mountains, the ocean and the Bay.

Yes the Bay Area is probably the most expensive metro to live in the world. But for all these reasons, and despite all the negative Bay Area and California chatter we hear online, I think it's actually underrated for building high volume and complexity consumer hardware.

Austin has been a good base for us for the last four years. People are friendly and I had space to design, build and fly my VTOL models. In the last four months in my garage here, I've built two VTOL models from scratch. I did numerous flight tests on the v0.1 foam and the v0.2 printed endurance bird is the one that's built and packed.

It's time to move to the South Bay and the next chapter of my quest!

Three months ago, I had zero experience in building anything that flies, no aerospace engineering degree, and really no business even trying to design and build a VTOL.

What can I say, I like a challenge.

I made a video walkthrough of my build. It’s much easier to show than tell with hardware.

If 30 minutes of a camera pointed at my VTOL build, my voice narrating and fingers point is not your thing, let’s walk through the build here.

For the first few weeks, I modeled, mathing, sketching, sourcing and assembling a foam VTOL.

I also learned how to 3D print parts, none of which have broken (so far).

Here’s most of my build process in 30 seconds.

Followed by dissembling, reassembling, iterating on software parameters and repairing. The first version of the build was complete.

Time for the maiden hover flight.

Many fixes followed to get a stable hover and enough control authority. This is essential before attempting a transition to fixed wing flight, like a plane. Remember that VTOLs (vertical takeoff and landing) transition from hover to fixed wing flight in midair.

When the transition to wingborne flight starts, the pusher motor on my VTOL spins up, which destabilizes the frame and the motor rpms that were working in hover. That’s why a stable hover is needed before attempting a transition, unless I wanted to rebuild the whole VTOL again.

I flew into trees, fixed compass interference issues, tilted motors for better yaw authority and upgraded some structural joins to 3D printed PETG parts.

After the iterative process to fix and tune the software and hardware, it was holding a solid hover, even in winds.

The foamie (foam VTOL) was ready for its first transition flight. There was an outside chance it would gain enough airspeed, becoming wingborne and turning off the vertical motors. Alas, that did not happen.

The VTOL basically did not produce enough thrust for its weight, or so I thought at the time. I did not want to have to rewire everything and upgrade the pusher motor and thrust. So it was going on a diet.

I had been flying with the battery on the left, weighing almost 740g or 1.6 lbs. That had to go, and I swapped it out for the one in the middle, weighing about half that, 370g or 0.8 lbs.

That was the pack I had used in my quadcopter build. It did not work out well, leading to voltage sag. The VTOL could not draw enough instant current to keep hovering and was falling a few feet while hovering. Not good.

So I ordered the pack on the right, which was almost as light, but much better at handling higher current draw.

It worked, kind of. I had thought thrust to weight ratio was the problem. Just fix that and open blue skies of VTOL flying to come! Not quite.

The lighter battery made me realize the root cause. I had mounted the pusher motor too low on the fuselage. Ideally, the thrust line of planes, ie the direction of the vector propelling the plane forward, should be *in line with the vertical center of gravity (Cg)*.

Mine was too low, causing a pitch up moment. In other words, the pusher motor was using a (too) large fraction of its thrust to pitch up during the transition. You can see it in the video.

At this point I thought maybe the problem was not fixable without building a new fuselage and remounting everything inside it - the avionics, the wiring, the ESCs, etc.

I gave myself one more test run to try to minimize the amount of pitching up it would do during transitions. Ardupilot parameter tweaks, tilting down the pusher motor, and manually pitching down on my radio transmitter during the transition. Turns out that did enough.

It transitioned into fixed wing flight and flew!

My VTOL accelerated and hit 44 mph or 71 kph, before banking a full 45 degrees! I managed to keep it in the air for almost two laps of my local airfield.

A sole person building a VTOL is a recent phenomena. Two decades ago, the motors were not power dense enough, the flight controllers not small enough or cheap enough, there were few open source communities and forums, and no YouTube to learn quickly from others. The control systems for multirotor drones required algorithms and enough compute that simply could not fit onto the memory and processor of microcontrollers at the time.

Since then, we’ve had miniaturization breakthroughs in controllers and sensors thanks to the iPhone and later Arduino, and more power dense motors thanks to drones.

There may only be single digit thousands of people who have built a VTOL from scratch by themselves. Ever. O1 and Claude think less than a 1000 people have done it.

That does not include the software, firmware for the control and other systems which Ardupilot handles like a dream, albeit with steep learning curves.

Still, wild to think so few people have done this!

Next Build — 3D printed long range VTOL

I have already started specs and design for the next build, v0.2. The foamie was v0.1.

My CAD skills are basic, but surprisingly good enough to actually design the entire airframe. See the lighter 3D printed foaming PLA nose in the pic, currently on the VTOL, vs the older foamboard one.

Printing the airframe will be the way. I’m learning fast how to make a lighter airframe that my robot employee, I mean 3D printer, can make for me day and night. That should accelerate my iteration speed on this build and future builds.

I want to push it to fly as far as possible, and I think four hours is achievable. One of the most important aspects of full scale eVTOLs will be range. You and I will be able to fly them next decade, and maybe, just maybe, these VTOL experiments will help me build those eVTOLs of the future.

Distributed electric propulsion (DEP) redefines safety and simplicity for aerial mobility. DEP involves using multiple electric motors distributed across the vehicle, enhancing safety through redundancy. This also enables vertical takeoff and landing (VTOL) capabilities similar to helicopters, without autorotation but with said redundancy.

DEP also offloads most of the compute required for human piloting, making flying almost trivial for autopilot multicopters and VTOLs. No human could adjust the speed of multiple electric motors hundreds of times per second or more. Only software could do that. Control systems including PID controllers, maintaining altitude and spatial positioning with GPS, stabilizing orientation and pitch/roll/yaw all happen on flight controllers with less than 500 MHz clock speeds (and some less than 200 MHz). That’s a processor with merely late 90s to early 2000s Intel x86 CPU speeds!

Directing my quadcopter to lift off the ground and hover for the first time was wild. It was an aha moment of how stable and safe (and cheap) the consumer tech is today, and where it is headed.

Flying a well tuned multicopter is *far* easier than driving and as easy as playing a video game. The difficulty is the hardest it will ever be.

But we will not need to do much flying ourselves at all. Autonomous capabilities of multicopters are already incredibly impressive with relatively cheap hardware and a 200 line python script (more later). They can detect obstacles using vision and/or LIDAR, and unlike terrestrial EVs, do not have to contend with road rage, cyclists and busy parking lots. The best way to avoid obstacles is if there are no obstacles, much easier done at 1500ft/500m above traffic than in traffic.

Of course, I’ve not even mentioned other important enablers. Ever improving battery technology, increasing specific power of electric motors and inverters, better power electronics, composite materials and the rise of consumer robots are rapidly expanding the design space for drones and flying EVs.

I don’t think enough people have truly grasped how game changing this all could be. I’m not sure it’s possible to without getting your hands on current generation HW and SW, and seeing the potential firsthand.

As I continue tinkering more in this space, I’ll write more about the tech stack and share more about I see things going. Briefly: not air taxis you can’t own. Instead, a flying EV you own and can land in your own (initially large) yard that is very safe and easy to fly, with surprising range and enough payload for two.

Ok, let’s get into my quadcopter build.

Why build a quadcopter

I wanted to assemble my own quadcopter to see if I would enjoy it and gain intuition quickly on the build process and flying. Both were a resounding yes, and motivated me to continue tinkering.

I was blown away at what quadcopters were capable of. High level learnings:

1. So easy to fly and incredibly stable GPS-assisted hover

2. Planning missions is easy to use and setup

3. With some work and help from GPT, impressive autonomous obstacle avoidance

4. Components are rock solid. Except for props, unsurprisingly. They break. A lot

Contents

Build requirements

Hardware and software

• Frame and dev kit

• US sourced components?

• Battery

• Radio controller

Assembly

Maiden flight

• How a multicopter hovers

GPS navigation and landing

Obstacle avoidance — lidar

• $15 obstacle avoidance sensor

• Upgrading to 100 meter sensing range

• Field testing

• Surface tracking and terrain following

Obstacle avoidance — vision

• OAK-D depth cameras

• Nvidia Jetson Nano companion computer

• Python script and flight test

What’s next

Build requirements

Multicopter. I wanted similar flight characteristics to an eVTOL (electric VTOL) in hover. A quadcopter was the simplest rotor configuration.

Weight and size. Bounded by cost and space in my workshop aka living room at the time. Chose a 20 inch (50 cm) quadcopter frame with 10 inch propellers (props) and good payload. 

Build speed and instructions. Found a quadcopter that I estimated I could assemble in an hour or so as a newbie.

Ease of flight and controls. I wanted the easiest possible controls with software taking as much cognitive load as possible. That meant an autopilot drone, not one for FPV racing. A reason for this is that personal eVTOLs should be as simple to fly as using an iPhone.

Ease and speed of part replacement. Likely that I would crash many times (and I did ofc), and I needed to be able to order parts for repairs as quickly as possible. 

Autonomous capabilities. I wanted to test obstacle avoidance and autonomous flight features — ideally both vision and lidar based systems and onboard compute for vision.

Hardware and software

There were only two firmware/software options that made sense for an autopilot drone. These being PX4 and Ardupilot, both open source.

I chose PX4 for my quadcopter build initially because it was easier and faster to get started with. PX4 is maintained by the non-profit Dronecode Foundation, a part of the Linux Foundation. I later switched to Ardupilot for the quad (and my first radio controlled VTOL build) because of better support for autonomy features. QGroundControl, the recommended ground station software for PX4, had a Mac app and seemed easy to start with. I wanted to avoid needing to set up a Windows environment if possible, but it was not in the end.

Frame and dev kit

As for the hardware, first I chose the X500 v2 dev kit from Holybro. Affordable, fast and easy assembly for my first build, good payload, with space for a companion computer for running computer vision algorithms.

Some specs:

• No soldering required (again for speed)

• 510mm (20 inch) frame size

• carbon fiber CFRP frame with fiber-reinforced nylon connectors

• 610g (1.34 pounds) empty weight

• 1500g (3.31 pounds) payload including battery 

• 35g Pixhawk 6C flight controller with 480 MHz Arm processor, 1 MB SRAM and onboard IMU with accelerometer, gyro, magnetometer and barometer

• 4x 920KV 272W (17A @16V) motors 

• 4x 20A continuous current electronic speed controllers (ESC)

• 4x 10 inch propellers with two spares

• Power distribution board with XT60/30 connectors

• 32g/1.1oz GPS module included (up to 32 simultaneous satellites) with built in compass

• Telemetry radio (for sending live telemetry data to ground station app on my Mac)

All for $575, ordered from US based multicopter retailer NewBee with a 30 day warranty. 

US sourced components?

Sourcing these components from the US would have been ideal, but there was not anything close to the price point that I could find. Modal AI, a Qualcomm spinout, offered this 24 inch dev kit for $3700 though it does include significant onboard compute. They target use by US government agencies with Blue UAS 2.0 certification. Driving down the cost of US drone components is another topic, but Andrew Bunnie Huang’s post is a good place to start.

The same day I posted this, Alex Klimaj from Ark Electronics actually launched a pretty impressive US made flight controller for $165. It features similarly spec’ed and speed controller to the Pixhawk 6C, and is comparable in retail pricing. Some caveats are that the Ark FC is single PCB, made more for FPVs, and is between 1.5 to 2.5x more expensive than other single board FPV FCs from China. Still, it’s promising progress!

Battery

The frame came with recommended specs for a battery and I went with this lithium polymer 4S 3300 mAh 80C pack for only $35. Let’s break down what that means. 

Lithium polymer (LiPo) batteries have a polymer electrolyte, instead of a liquid one found in standard lithium ion batteries (LIBs). LiPo packs have more specific power (power per unit mass) than LIBs, and are usually used for FPV racing quads, as well as autopilot quads. Specific power is expressed as a higher C discharge rating. An 80C pack means the battery can discharge at 80x current of the rated mAh (or Wh) capacity, in this case at 264 amps or 3.9kW! That’s enough power to (briefly) run a central heat pump HVAC or an EV cruising at 30 mph (48 kph) for less than a minute. 

4S means there are 4 3.7V cells wired in series, increasing the voltage of the cells without increasing the current. This allows it to more efficiently run the brushless DC motors and ESCs with less heat loss. More EVs and eVTOLs now run an 800V architecture would require about 216 3.7V cells in series, and enough cells in parallel to meet the current draw for these vehicles. 

To charge my LiPo, I sourced this ISDT charger capable of both AC and DC input for $65. It can balance charge LiPo cells at 200W with DC input, but the 50W AC input was plenty for recharging my LiPo at its recommended 1C charging speed. It can also discharge the LiPo cells to ~3.75 V, essential for safe storage when not in use. 

Radio controller

I chose to pair the Frsky Taranis QX7 transmitter (Tx) with the Frsky Taranis X8R receiver (Rx), which allowed a solder-free connection to the flight controller. The Tx and Rx cost me $178. 

That brings the total cost to less than $800 (sans battery charger). I just had to assemble it and then maiden, aka do a maiden flight.

Assembly

Assembly was fast and I’m glad I picked this beginner friendly dev kit. The process helped me gain a much better intuition for quadcopters including integration of components, distributed electric propulsion and the high strength/weight for carbon fiber parts.

For reference, I used this and this video guide, as well as the build guide in the PX4 docs and took less than an hour for the dev kit. There was more time needed for pairing the Tx and Rx and setting up qGroundControl.

Once assembled, I next had to pair my radio Rx and Tx and finally calibrate sensors and configure parameters in the qGroundControl app. 

With the hardware assembled and software set up, I was ready to maiden!

Maiden

Smooth as butter. That sums up how the maiden flight went. As a first time flier, I was surprised just how stable the quad was in hover.

How a multicopter hovers

This comes down to the PX4 software doing all of the heavy lifting to maintain level flight. PX4 and other autopilot software like Ardupilot have very mature control system architectures.

1. Sensors on the flight controller like the accelerometer, gyroscope, altimeter, etc all input data streams to the attitude controller, which maintains orientation in space. 

2. Position sensors like GPS and rangefinders (more on this later) help keep the quad stable in the horizontal plane and altitude. The GPS input corrects for drift from wind, and helps the altimeter estimate altitude.

3. Inputs from my RC transmitter are mapped to pitch, yaw, roll and climb/descend controls for the attitude controller. The image below shows how these map to position flight mode, the mode in the autopilot software that maintains altitude and positioning if you don’t move the joysticks on your transmitter.

4. The attitude (and speed and attitude) controllers then sends the inputs to the ESCs (electronic speed controllers, also known as inverters) which provides millisecond-speed, 3 phase AC to power the electric motors. This AC current allows the thrust output for every motor to be precisely and rapidly controlled. For example, to pitch and move forward, the flight controller increases thrust to the rear motors, while making smaller adjustments to compensate for altitude, wind, etc. 

Position mode in PX4 and Ardupilot are the easiest way to learn to fly an autopilot drone. It uses GPS to counter wind drift, and allows someone with no training to intuitively grasp the controls.

I also started with Position mode because I want to build flying EVs. One of the key requirements is that it should be extremely easy to fly — as easy as using an iPhone and an order of magnitude less to think about then when driving. Position flight mode most closely mimics how I think hover should work in a flying EV operated by the pilot. 

I also tested manual flight mode briefly, but crashed pretty hard. In manual flight mode, the flight controller does not counteract wind nor try to maintain altitude. It is similar to how FPV racers fly, and requires practice! I tried in a parking lot with many trees. Bad idea. I was lucky to have only broken one prop!

That just reinforced for me that flights should offload as much of the cognitive load of flying from the pilot to the system. To test that, I wanted to test what the quad could do with autonomy. After all, PX4 and Ardupilot are called autopilot systems!

GPS navigation and landing

First I tested return to home functionality. This allows GPS to fly the vehicle back to the takeoff GPS coordinates autonomously. After a basic test to ensure it was working, I wanted to see how accurate the landing could be, using GPS alone.

Armed with a yellow spatula, I did four landings using autonomous return. The accuracy was impressive, ranging from approximately 0.3-1.5 meters or 1 to 5 feet. Significantly more accurate than a smartphone would be because of their multi-constellation support and better antennas.

There are far more accurate ways to precision land a quadcopter, like using an IR-LOCK sensor, RTK positioning, and more. But for my testing, the standalone GPS was good enough.

I also tested autonomous flight missions by setting up waypoints in QGroundControl. It’s super easy to set up missions using the map based UI. You can even use it to do structure scanning and other types of preset mission patterns. I was seriously impressed with the functionality of these free and open source tools.

Obstacle avoidance — lidar

The most fascinating autopilot quadcopter capability is autonomy. Miniaturization, in part driven by smartphones, have improved performance and lowered the cost of cameras, LIDAR and other sensors that enable autonomous drones and robots. 

My goal was to gain a base technical understanding of the abilities and limitations of these sensors.

$15 obstacle avoidance sensor

I started with time of flight (ToF) LIDAR sensors, because they were the cheapest and easiest way to get started in a tiny form factor. ToF sensors emit a point light source to an object, and the time of flight is measured when it bounces back. Unlike LIDAR, they reflect light off a more limited field of view, and are not used for simultaneous localization and mapping (SLAM). 

For only $15 (!), I purchased the memorably named Holybro VL53L1X ToF sensor that was practically plug and play, no soldering or even wire stripping needed. This particular sensor only has 4 meters/13 feet of range at 50 Hz sampling rate. I figured that was enough for initial testing.

I immediately ran into issues with this Holybro unit. It outputted extremely variable distance readings, like when sitting on the ground pointed at a fence about 2.5 meters / 8 feet away. The readings are on the Y axis in centimeters, time is on the X axis, as seen in this screenshot. 

After some troubleshooting (many, many such cases), I realized the sensor has issues in direct sunlight. See how stable the following readings are, once I pointed the sensor inside, around the ~03:05:00 mark.

Clearly the $15 sensor had its limits, which were not clearly stated in the PX4 or Ardupilot docs, nor on Holybro’s product page. Oh well, on to the next one.

Next, I tried the Benewake TFmini Plus, which offers 12 meters of range and up to 1000 Hz sampling rate! The wire connector was not compatible with my Pixhawk flight controller, so I finally had to do some wire stripping and very hacky splicing.

I had more issues with this sensor. The sampling frequency was less than 1Hz, and I decided to switch to Ardupilot software from PX4 to see if that would fix the issue.

Ardupilot had better support for vision based obstacle avoidance and so this was worth the fairly small time investment. Its recommended ground control software, Mission Planner, runs natively on Windows with no Mac support and VMs were not great either. So I set up a 10 year old Surface with Windows 10 to run it, fortunately without hiccups. 

Upgrading to 100 meter sensing range

I could not get the TFmini Plus working at a usable frequency, so I switched to another ToF sensor. This time I bit the bullet and bought a high end ToF sensor for consumer drones — the Lightware LW20.

The LW20 is a huge step up from the ToF sensors I had tried. It has 100 meter range and can be configured to see through transparent materials. It is also a lot more expensive at $280. Surprisingly, it is actually engineered in South Africa, which, fun fact, is where a YouTuber currently holds the world record fastest drone hitting over 500 km/h (312 mph) without a wing. 

After bench testing with a Arduino-based Uno R3 microcontroller, I soldered the wiring and applied heat-shrink tubing to allow the sensor to connect to my flight controller. I had never soldered before so I was a little nervous, but it worked out fine if a bit messy. Testing at home, I was getting readings of up to 30 meters. I configured Ardupilot with a new rangefinder (distance sensor) and was ready for obstacle avoidance testing. 

Field testing

The first test was simple obstacle avoidance. That is, will the quadcopter stop moving when something is in its path, even if I am directing the quad towards it? 

It worked well enough! Pitching forward (right thumb) moves the quadcopter forward only as far as the sensor’s set distance. The flight controller was clearly taking the ToF sensor’s distance input, and adjusting ESC outputs to the motors. 

Next, I wanted to test if the quad could route around an obstacle while navigating on autopilot, using an algorithm called Bendy Ruler. Basically, the functionality allows a “lookahead” distance for the rangefinder to scan head to find obstacles. It also has a “max margin” distance which is how far away it should stay away from obstacles. Ardupilot offers vertical or horizontal avoidance strategies, which are configurable in Mission Planner ground control software.

To test if Bendy Ruler obstacle avoidance worked, I configured a simple autopilot flight plan in Mission Planner for the quad to move back and forth between two waypoints about 10 meters / 30 feet apart and at an altitude of about 2 meters / over 6 feet. It was ready for flight testing. 

Bendy ruler vertical demo with the LW20 sensor:

Bendy ruler horizontal demo with LW20:

As these video demos show, Bendy Ruler was working pretty well. Sometimes the sensor did not detect the obstacle, but after some configuration, I was quite impressed overall with both the sensor and the Ardupilot software to make this possible.

Surface tracking and terrain following

The next feature I wanted to test was surface tracking / terrain following. They do exactly what they sound like, with terrain following for autopilot flight modes and surface tracking being a feature for auto and some manual flight modes. 

First I tried keeping my LW20 as a forward facing rangefinder and pointing the TF Mini downward to track altitude for surface tracking. That did not work. The readings from the TF Mini were too inconsistent. It seems it is best used indoors in consistent lighting conditions. So I removed the TF mini and repositioned the LW20 to be downward facing. 

Again, the LW20 performed well! Both for surface tracking and terrain following.

In this surface tracking test, you can see the quad adjusts altitude dynamically using measurements from LW20 ToF lidar.

Terrain following is a similar feature but for autonomous GPS based mission plans.

I had some minor drone hover stability issues, but not enough to derail the tests. 

Obstacle avoidance — vision

The final obstacle avoidance tests I wanted to do were vision based using AI.

I had all but just over 100 hours of quadcopter tinkering experience at this point, and this was going to be a level up in difficulty. 

Deciding on the hardware to do this was the first step. I would need both a camera system and a companion computer to process the images in real time. Autopilot flight controllers do not have enough compute capacity to do this, topping out not much more than 1B operations per second. Onboard companion computer with at least 2 more magnitudes of compute was needed. 

Oak-D depth cameras

I looked at a couple of options for the camera system first. Luxonis’s OAK-D product range in particular had some interesting features, the most important being stereoscopic vision. 

Stereoscopic vision means perceiving depth using a two camera system like our eyes. It allows for less compute intensive obstacle avoidance, at least up to the range of the system — up to 8 meters in this case. A single camera has to estimate depth using various algorithms, which meant more potential work for me in writing up scripts for image processing. The OAK-D Lite that I decided on also had longer range than competing products like Intel’s Realsense cameras, while being lighter and cheaper. 

Another important benefit was that these cameras had good docs. This would be important later when prompting GPT to help with writing my custom image processing script. 

There were more benefits to OAK-D’s camera systems. They were not only camera sensors, but included enough compute onboard for very light depth AI processing. In addition, Luxonis is based in the US. Also fun fact: OAK-D cameras are also used in many other robotics applications, even Figure AI used them during development. 

Nvidia Jetson Nano companion computer

The OAK-D Lite camera has 4 trillion OPs of compute, which meant I could run various deep learning vision models at 15 to 30 fps at up to 640x640px. Why did I need a companion computer?

Basically 

(1) for recording video output from the camera system for troubleshooting, 

(2) gain intuition about how a flight/companion computer interfaces with the flight controller and sensor stack for aerial autonomy and 

(3) offloading some compute from the OAK-D 

So despite the extra cost ($155 at the time, a little less now), 20-30% increased weight and power draw, I bought the Jetson Nano dev kit. Though it has less compute power than the OAK-D Lite, the Jetson Nano actually outperforms it on ResNet-50 @224x224 (36 fps vs 26.5 fps), the one deep learning model I could find benchmarked by both. The exact compute capability did not matter though, as long as the Nano could do (1), (2) and (3) for me, which it was perfectly capable of. 

With the Jetson Nano in hand, I had to connect the OAK-D to the Nano and the Nano to the flight controller, with the help of this video and guide. I also had to make sure I could write code and push it to the Jetson Nano. I had to work out how to SSH into my Jetson Nano’s linux system from my Mac and resolve a few other software issues, which (last year’s) GPT 4 model was very helpful in doing. 

I also had some space constraints mounting the Jetson Nano onto my quad’s frame. The frame has a dedicated payload platform, but it seemed more intended for a smaller (and far less powerful) companion computer like the Raspberry Pi 4. The Nano was borderline too big, given I had to connect its USB port to the USB C output from the OAK-D. That USB connection causes EMI (electromagnetic interference) issues with the quad’s GPS module in particular. I had to go through several iterations of relocating the GPS module, until I landed on putting it next to the flight controller, which was not ideal, but it worked well enough. 

Power delivery was a problem too. The OAK-D Lite uses up to 5W when running image processing models and it draws power from the Jetson Nano via the USB to USB C port. However, the Jetson Nano itself draws up to 10W at 5V and 2A, which is too much to supply from any ports on my Pixhawk 6C flight controller. The Jetson Nano needed power from the PDB, power distribution board, which itself draws power from the 4S LiPo battery. I sourced a UBEC, universal battery eliminator circuit, which is basically a DC/DC converter to step-down the LiPo’s voltage to 5V. I soldered it to the PDB wiring on one end, and a DC barrel jack wire to the Jetson Nano on the other.

Python script and flight test

Luxonis has great documentation and code samples for how to use their OAK-D depth cameras. I first adapted their spatial location script to get depth distance data. My workflow was to code the script on VS Code on my Mac, and SSH the changes to my Jetson Nano. At runtime during flight testing, I would execute the Python just before takeoff. Hacky, but it worked well enough. 

I moved on to Luxonis’s multi region of interest (ROI) spatial calculator. This allowed depth tracking of a wider field of view than a small bounded point source. I heavily adapted the script to display a 3x3 ROI for the whole image and show a bounding box for the ROI in which the closest obstacle theoretically is located.

The script I got working is shown in this RGB video from the OAK-D taken during a test flight, for which I had set a 3 meter obstacle avoidance distance limit.

OAK-D Lite and Jetson Nano depth camera processing. Note the 3x3 ROIs and the detection of the nearest obstacle from these ROIs, displayed in the top left corner in realtime. At 20 seconds I am pitching the quadcopter forward, but it refuses to move because of the 3 meter distance limit from obstacles.

There was quite a bit of LLM wrangling to get the pipeline working well enough for testing. I wrote only a handful of the 200+ lines of code myself in the script. Instead I prompted GPT4 for code snippets that I could copy pasta into the script. This was mid 2023, soon after GPT4’s release, and it was already good enough to produce useable functioning code to tie together the functionality I needed. I did, however, need to ask the Arudpilot forums to resolve one problem with the script. Even with that said, it’s wild to think how much more effort this would have taken without LLMs. And this long before canvas or Claude 3.5 Sonnet.

Here is the high level breakdown of the script, summarized recently by Sonnet. I’m happy to share the full script if interested. 

## OAK-D Lite:

1. Stereo depth processing

2. Spatial detection network (MobileNet)

3. Spatial location calculations

4. Color camera operations

## Jetson Nano:

1. Main script execution

2. OpenCV operations (visualization)

3. MAVLink communication

4. Video writing

The OAK-D SoC handles the depth processing, the deep learning model execution to process the depth and spatial data, and the color image data from the RGB camera for writing to an mp4 file on the Jetson Nano. The Jetson Nano visualizes the bounding boxes and displays the depth data on the video output, records the video, and sends the depth data to the flight controller using MAVLink, a simple protocol for communicating with small drones.

What's next

I was really surprised how far I was able to push obstacle avoidance with my quadcopter build. I started with no quadcopter experience or really any hardware experience.

In somewhere around 200 hours of effort, I was able to:

• Learn a ton about how to assemble a modern quadcopter 

• Troubleshoot hundreds of hardware, software and vision script fixes

• Integrate the hardware and software for lidar sensors and stereoscopic cameras

• Write a working python script for detecting the nearest obstacle from depth sensing cameras

• Run dozens of flight tests 

• While recovering from long COVID

• And having a blast! 

This project definitely kindled a passion for hardware and aerial mobility.

I’ve since rebuilt a radio controlled tilt-rotor VTOL after destroying the flight controller in a crash. I’m also building a 4ft/1.2m lift and cruise VTOL that is a first subscale prototype of a flying EV design I am working on.

Much more to come!

Share

Twenty milligrams, about the weight of a feather. Spiber's founders had only made this much spider silk protein in three months in 2008. For their first year operating, it was an inconspicuous amount and very humble beginnings.

Since then, spider silk protein startups have managed to almost double production on average every six months since 2008. These startups are closing in on about eight tons of production this year, over eight orders of magnitude more than in 2008. Spiber and others have too much demand and are hastily scaling to keep up.

This article is Part Two in a two part series about materials. Part One outlined a case on why we need a new materials paradigm. This piece dives into amazing new performance materials leading this seismic shift, how many of them have underpinnings in synthetic biology and how startups can use these materials to engineer products we've never had before.

In writing these articles, I have been learning as I go and do not have all the answers. That said, I now have conviction that emerging materials with better properties will create a new normal. As I wrote in Part One, synthetic polymers became a new class of materials that changed the world during the 20th century, enabled by fossil fuels. In the coming decades, I believe new materials will replace synthetic polymers and other materials as well as being used in completely new ways. A ton of opportunities will open up for new startups, which is my interest as a founder.

We will summarize some key takeaways before diving in.

tl;dr

The rapidly evolving synbio stack is allowing startups to be faster, better and cheaper. The technologies underpinning synthetic biology and by extension, performance biomaterials, have rapidly improved over the last two decades. A core enabler driving this is that costs of DNA sequencing and synthesis have fallen orders of magnitude, and productivity has increased similarly.

Companies making chemicals using engineered microbes, like Amyris, Geno and Lanzatech, have leveraged these improvements. Amyris and Geno have drastically reduced development times of their molecules from 2-4 years to 6-10 months or less. Amyris lowered the cost of farnesene, their first commercial molecule, fourfold in four years while Lanzatech's ethanol is cost competitive with incumbents today.

These three companies are all scaling rapidly. Amyris is at thousands of tons of production volume for squalene alone, while seeing blistering revenue growth of their consumer products. Geno is at 30,000 tons of 1,4 butanediol and Lanzatech 60,000 tons of ethanol production. Both are building new plants that aim to increase scale threefold in Geno's case and tenfold for Lanzatech in the next few years.

Biomaterials like spider silk and mycelium are scaling up at historic speeds, yet are not being talking about. Spider silk protein made by Spiber and AMSilk, the industry leaders, has grown production at an incredible ~273% compounded annual growth rate (CAGR) since 2008 (see below). They are currently at about eight tons, but are scaling up to over ten thousand tons in the coming years. At that scale and even as growth slows, it should still close in on the production volume of global 3D printed polymers. How many people would have predicted that could happen by the mid 2020s back when milligrams of spider silk protein were made in 2008?

Mycelium production is also rapidly scaling. Last year, Ecovative produced about 45 tons for a variety of uses. Their spin-off, MyForest Foods, uses AirMycelium, Ecovative's patented tech that grows mycelium in open-air racks. MyForest Foods has just opened a new commercial plant that alone aims to scale to over 13,000 tons by 2024, which will accelerate their 212% CAGR growth over the last two years. Though that plant is being used to produce fungi-based bacon, other plants that make leather and foams would be largely the same except for the strain, mycelium growth substrate (e.g. woodchips and hulls of seeds) and control algorithms.

Performance biomaterials have novel properties that surpass existing materials. Spider silk protein, algal oil polymers and squid-inspired protein are some examples of performance biomaterials we'll discuss.

Spider silk protein (AMSilk and Spiber) is lightweight, has incredible toughness, high tensile strength and is highly elastic. It's being used in cosmetics, medtech, apparel and watch straps, and is being tested by Airbus for replacing carbon fiber in composite parts on airplanes, Mercedes in interior car parts and more.

Checkerspot's polyurethanes made from an algal fermentation process are used in their direct to consumer skis and snowboards. These performance materials offer excellent damping and great low-temperature tensile strength.

Tandem Repeat's squid-inspired protein heals in seconds by applying 50°C/122°F heat and also has thermal properties for helping keep clothing cool. In a recent Nature paper co-authored by one of Tandem Repeat's cofounders, squid-inspired proteins have been demonstrated as a strong artificial muscle with possible soft robotics applications.

Though not a biomaterial, graphene has exceptional properties and is finally hitting its stride. It's being used in a wide array of industries, surging 61% CAGR from three thousand tons in 2018 to twelve thousand tons of global production in 2021. At this scale, like with spider silk protein, it may also overtake the global production volume of 3D printed polymers by 2025.

Startups that source performance biomaterials and integrate them into products are a largely untapped pathway to market.

Almost all performance biomaterials producers today partner with incumbents that make end products and distribute them. Two of the only exceptions I know of are Spiber partnering with sustainable fashion startup Pangaia and Checkerspot's own brand WNDR Alpine making skis and selling them direct to consumer.

Two benefits of biomaterials working with consumer product startups are worth calling out.

1. The potential to offer more differentiated products free of constraints of larger incumbents that customers can be really excited about.

2. These products can help pull demand forward for biomaterials producers.

Incumbents did not build the iPhone, any Tesla EV, Fitbit wearables or even Allbirds shoes. Just like in those and countless other examples, the upstarts will push the boundaries of performance materials more than any established players.

In the years ahead, I believe sustainable materials will be the norm as more producers scale up. They will become table stakes for tomorrow's products and no longer be a differentiator. Ambitious consumer startups will need to push performance boundaries to drive more adoption.

The future potential of these materials is enormous, and hard to fathom this early in their ramp up.

The novel properties of biomaterials and scalability will improve existing products and unlock new products that do not exist today. Using multiple performance biomaterials in one product will become more common and enhance the product's benefits. Engineering them together with other technologies will also increasingly happen.

Possible applications will include self-repairing smart clothing, biocompatible human-computer interfaces, electric propulsion vehicles (especially next-generation aircraft), and in humanoid robots as artificial skin and muscles.

That's a summary of what to expect in this piece. Now let's dive deeper into each of these ideas starting with the synbio tech stack. We'll see how it has enabled companies like Amyris, Geno and Lanzatech to build better, faster and cheaper and how that bodes well for performance biomaterials as they scale up.

The synbio stack is a key enabler

Let's cover some brief definitions to level set.

Biomanufacturing is the use of biological systems that have been engineered, or that are used outside their natural context, to make a product. It is enabled by synthetic biology (synbio), an interdisciplinary domain that involves the application of engineering principles to biology.

From this, biomanufactured materials refers to materials produced using engineered biological systems. Examples of biological systems include submerged, solid state or gas fermentation using engineered microbes, cell-free systems using enzymes as catalysts and finally mammalian cell cultures. Note: when I mention fermentation or microbial fermentation in this piece, it means submerged unless stated otherwise.

Biomanufactured materials is a mouthful, so I use ‘biomaterials’ for short. Just be warned that this term is also used to describe any material that can interact with biology, like prosthetics or other material therapeutics.

So biomaterials it is. Let’s now build towards the case for biomaterials by looking at some important developments in biomanufacturing and synbio in the last few decades. These developments underpin what is happening now with biomaterials and are ongoing tailwinds for the future of the space. I’ll focus more on microbial fermentation, but there’s also rapid progress in improving solid state and gas fermentation, as well as cell-free systems.

Biomanufacturing has been quietly ramping up

The origins of synbio go back just over four decades to Genentech’s synthetic insulin, first produced in 1978. Genentech's breakthrough was the first microbe-produced biopharmaceutical drug, aka biologic.

Synbio has gone on to disrupt pharmaceuticals with biologics having grown to contribute over 32% of total industry revenues in 2018 in the US. Six of the top eight drugs by revenue in 2016 were biologics, though perhaps helped by market distortions. Fun fact: many biologics are so valuable they can be more valuable than gold on a weight basis. This super-high value product shows two interesting points. First, a $300B+ global biologics market is not niche in a monetary sense, but in terms of the quantity of biomanufactured goods, this is still a relatively low volume market. This is a classic example of high value, low volume which often serve as beachheads for nascent technologies. Second, the stratospheric price means biologic developers did not have to worry about optimizing product titers (concentration) in their process.

Companies making lower value products (basically everything else) did have to focus on titers and costs.

As titers improved for fermentation, startups began focusing on lower value products. Around twenty years ago, companies like Amyris, Genomatica and Lanzatech were founded to use fermentation to produce various chemicals. Encouraged by high oil prices, many of these Cleantech 1.0 companies focused on making biofuels and raised a boatload of capital during the late 2000s. Unlike the three companies above, most did not survive. Many that died did not pivot away from trying to make biofuel unit economics work.

Amyris, Genomatica and Lanzatech are amongst the survivors. They have started gaining strong traction in markets where petrochemicals or corn starch have been incumbents for decades.

Amyris survived a costly biodiesel foray and pivoted to becoming a vertically integrated health and cosmetics company. Their fermentation process first produced high-purity squalene at a far lower cost than can be extracted naturally, and has come to dominate the global market with 70% share or over 1500 tons per year. Amyris has also expanded to produce 12 other higher value ingredients using fermentation of engineered yeast. Further, they have found exceptional product market fit for their consumer products with direct to consumer (DTC) being a core driver of growth for new product launches. Aided by DTC success, their consumer business has grown at 133% CAGR since 2019 to $77M in the first half of 2022. Their consumer business now accounts for the majority of their revenues.

Genomatica (now Geno) has been slower growing in recent years than Amyris, but by targeting lower value commodity chemicals, has scaled fermentation capacity even more. Their first target molecule was 1,4 butanediol (BDO), a chemical intermediate used for various polymers like spandex with annual global capacity at 2.5 million tons. More recently, they've expanded to make nylon intermediaries and personal care ingredients. Rebranded as Geno, they are more than 3x'ing their BDO capacity in two years from 30,000 tons to over 100,000 tons per year in 2024. By then, they will have about 4% the global market. It's worth noting they sell their ingredients B2B, and have grown annual revenues at 37% CAGR from 2016 to 2021 to $48M.

Lanzatech produces ethanol and more recently, sustainable aviation fuel (SAF), ethylene and other commodity chemicals. Unlike Amyris and Geno, they use gas fermentation technology, which is less established than using crop-derived sugars. Their process feed syngas, a mix of carbon monoxide and hydrogen, instead of sugars to their microbes. Lanzatech's demo-scale plant in Georgia and first commercial plant produce about 60,000 tons per year combined and their product is already cost competitive, even with first generation plants. They are building seven new plants and another seven are in engineering. These plants will allow them to ramp capacity about 10x to 600,000 tons per year of ethanol, SAF and other chemicals.

These low value chemicals are multiple orders of magnitude cheaper than biologics to make. To viably produce these molecules, the underlying tech stack of synthetic biology had to improve exponentially in the last two decades. Strain and metabolic engineering had to improve. Iteration cycles had to accelerate. All of this happened.

Synbio stack: faster, better, cheaper

We can see just how quickly synbio has improved by thinking of synbio as layers in a stack. As we'll see shortly, these improvements have enabled Amyris, Geno and Lanzatech to accelerate their progress. Biomaterials companies have also benefitted a ton (yes, a pun).

This image above gives a good high level example of a synbio stack. See Amyris featured in the application layer, as well as a couple of biomaterials companies we'll touch on later. A couple of other frameworks are worth pointing out. Drew Endy at Stanford articulates abstraction layers for synbio clearly and simply in many publicly available speeches. Elliot Hershberg has laid out a more recent outline of the synbio stack on his Substack.

The notion of a stack comes from software. For example many teams that built a v1 of a web applications historically (e.g. Wordpress blogs) used a LAMP stack consisting of Linux, Apache, MySQL and PHP software layers. Cloud, containerization and other more recent developments have abstracted away or simplified more layers of the tech stack. This has, among other things, allowed founders to build faster and more cost effectively and software engineers to focus more on development rather than worrying about provisioning and maintaining servers.

The most surprising thing about the synbio stack was just how fast the whole synbio stack was improving.

DNA sequencing and synthesis costs have been falling at a super exponential rate for decades, as shown here and known as the Carlson Curve. Meanwhile the speed of DNA sequencing and synthesis has similarly increased at a super exponential rate, as shown below. Note the vertical axis is log scale for both charts. The cost of sequencing a human genome has fallen from $100 million dollars in 2001, with the first ever human genome taking 13 years to sequence. Today, they are as cheap as $100 in as little as five hours. The workhorse microbes used for most fermentation processes today, S. cerevisiae and E. coli, have four and three orders of magnitude less base pairs in their genomes vs humans.

Engineering of microbial strains have also been rapidly improving. Strain engineering really helps startups that are looking for an optimal strain, usually of e. coli bacteria or s. cerevisiae yeast, to produce a molecule with specific traits. One startup, Inscripta, offers a gene editor in-a-box that allows companies to test hundreds of thousands of genetic variants in days. For example, Ginkgo used Inscripta's product to develop strains 10x faster and halve their Design-Build-Test-Learn cycle time.

There are many more examples of the synbio stack helping make development and early stage biomanufacturing faster, better and/or cheaper. Want to run more experiments but don’t have lab space? Cloud labs have got you. Too much pain using software not designed for bio to run experiments? Benchling can help. How about abstracting away cloud compute infrastructure for bio? Latch will do it. Need fermentation capacity? You can do it in the cloud with Culture, or use various contract manufacturing organizations (CMO). Strain development not core to your startup's value? Ginkgo or a contract research organization (CRO) has you covered. Most of these companies were founded in the last ten years.

That’s just a brief sampling, but the picture is clear. The synbio stack makes biomanufacturing better, faster and cheaper. Amyris, Geno and Lanzatech are amongst the many companies to have taken advantage of these trends. Let's have a look at how they've done it. This will be instructive in how biomaterials startups are starting to do this as well.

Amyris has accelerated time for molecules to go to market by 80%, and are gaining traction in markets faster. Meanwhile, development time has fallen six fold from over 36 months for their squalene molecule in 2012 to about 6 months years for molecules launched a few years ago. Geno's BDO took 27 months to reach a commercializable titer (50 g/L), whereas their second molecule only took 10 months, and is probably faster today as this was back in 2014. To be fair, both Amyris and Geno have developed platform molecules which they can use as building blocks for chemicals and other products. Amyris’s farnesene is a good example, as it can be made into squalene, hemixsqualene and other fragrances which are used in several of their brands.

Costs to biomanufacture have come down too. In 2011, farnesene cost Amyris $7.8/L to manufacture, but they had lowered costs more than 4x to $1.75/kg by 2015. Lanzatech’s ethanol is at cost parity and Geno’s BDO is approaching it as they scale towards 4% share of a $17B market.

Further, better metabolic engineering and other improvements have upped titers, helping these companies and the synbio industry come down the cost curve. Geno’s BDO titer improved from around 10 g/L in 2009 to 80 g/L in 2011, reaching 140 g/L by 2016. See the chart below.

These gains have helped enable these three businesses to be amongst the first synbio companies to be closing in on industrial-scale capacity. Amongst other products, Amyris is producing 1500 tons of squalene and selling DTC in their cosmetics brands, Geno is making 30,000 tons of BDO and Lanzatech is cranking out 60,000 tons of ethanol. All have plans to ramp capacity fast and extend their platforms as discussed above.

Synbio products like these are closing in on over a million of tons of molecules capacity per year. It is ready to scale.

Biomaterials are at an inflection point

Like the companies above, biomaterials startups have employed advances in synbio technology to accelerate their progress. I’ve tracked at least two biomaterial products which have scaled capacity at stunning growth rates.

Spiber and AMSilk are leading companies producing biomaterials processed from spider silk protein. AMSilk initially focused on high value cosmetics and medical applications, whereas Spiber developed fibers for apparel. They are also targeting aerospace, mobility and other industries. In the first paragraph, recall Spiber made only 20 milligrams of spider silk protein in solution in 2008. It took three months to make this amount, as shown in this image from a talk by the founder Kazuhide Sekiyama. At that point, Spiber had not even began honing the fabrication steps needed to turn the protein into a usable material. According to C&EN, Spiber's capacity in 2021 was "several" tons per year. Let's call it two tons per year to be conservative.

Being founded in 2008, one year after Spiber, AMSilk would have also made a negligible amount around that time. They have managed to scale up though, with their protein used as a cosmetics ingredient in twenty products by 2017 to help come down the cost curve. More on that later. This year, AMSilk is on track to contract manufacture six tons of their protein in Europe alone, up sixfold from last year.

Based on Spiber's 20 mg in three months in 2008 together with their current capacity combined with AMSilk's, spider silk protein production volume has approximately grown at a stunning 273% CAGR on average since 2008. This growth rate is even comparable to the very successful early synthetic polymers like Bakelite or nylon when they were at similar production volumes. As I wrote about in Part One, these plastics helped kickstart the last materials paradigm.

The capacity growth is set to continue apace. Spiber and AMSilk are in the middle of aggressive scale-up to commercial production. In the next few years, Spiber aims to ramp production to thousands of tons. Firstly at a demo scale plant in Thailand with hundreds of tons of capacity, and then at a commercial scale plant in the US in Iowa with about 10x capacity of the Thailand plant. To be fair, Spiber has overpromised on milestones in the past, so I am not using their best-case timelines for plant openings. AMSilk is aiming for several thousands of tons of capacity by expanding its range of CMOs beyond Europe.

Mycelium, the “root structures” of certain species of mushrooms and other fungi, is another biomaterial scaling up fast. In particular, the startup Ecovative produced 45 tons per year in 2021 for a growing range of uses and can double capacity. When Ecovative was first founded in 2007, Eben Bayer made a small palm-sized piece of mycelium composite grown under his bed in his dorm room, shown here.

If that piece conservatively weighs 1kg (2.2 pounds), Ecovative has more than doubled their mycelium capacity every year on average. MyForest Foods, a spin-off from Ecovative, has just opened a plant that will scale up to over 13 thousand tons of mycelium per year by about 2024. Bayer is also CEO of MyForest Foods, and this plant is focused on making mycelium bacon and other alternative proteins. Ecovative is also scaling up mycelium production for their own Forager brand, as well as providing technology for products like Bolt Threads' Mylo leather and more.

Of course, even tens of thousands of tons of capacity is still peanuts vs the tens of millions of tons of capacity for the highest volume synthetic polymers today. That said, those materials have had several generations to scale-up and it would be unwise to bet against sustained non-linear growth. Ten or twenty years ago, few people anticipated solar PV at the scale it is today. Just as few people, ten or even five years ago, expected lithium ion battery packs to power 12% of all cars sold globally now as pure EVs. Growth rates matter and most people (especially analysts) underestimate their compounding effect. Inflection points only accelerate this growth.

I think we are close to an inflection point for these biomaterials. It does seem that with scale-up plans and very strong demand, scale-up of spider silk protein, mycelium will continue. Production growth can probably be sustained but there will be other challenges in addition to improving titers, lowering costs and accelerating cycle times discussed above.

The proteins being made in fermentation bioreactors are long and complex carbon molecules. This often makes the downstream processing much more complex. Once the molecules are isolated and purified from the fermentation medium, biomaterials startups face the challenge of fabrication, ie turning these molecules into materials. R&D in this stage of the biomanufacturing process has involved deep technical risk, as much of the materials science and engineering to commercially formulate biomaterials is still new.

We can now see how Spiber, AMSilk and Ecovative have begun scaling up production to commercial scale. Let's look into biomaterials that are aiming to mimic the performance, look and feel of existing materials before diving in to the case for performance biomaterials.

Drop-in biomaterials

Many startups producing biomaterials are focused on largely being drop-in replacements for existing materials in products. Many more are focused on replicating performance characteristics, but, at least for now, not aiming to go beyond.

Companies working on mycelium or collagen-based leather and other materials are making good progress selling to incumbent brands — especially in apparel. These materials are more sustainable, and usually easier to process at end of life vs synthetic or natural options. They also will reduce resource consumption and manufacturing emissions, in part because fermentation takes days to produce materials instead of months. At or close to price parity with legacy fabrics like leather, they could take large shares of these existing markets.

That said, as biomaterials keep ramping, sustainability will become table stakes. I am more excited personally about biomaterials that can deliver better performance than incumbents. Making better products is how technologies we take for granted became so widespread. iPhones > Nokias, LEDs > incandescents, computers > electronic word processors, EVs > gas-powered cars, etc. Early in the s-curve, a new technology is better at replacing existing use cases. For example, nylon stockings was a replacement for silk stockings. As the technology matures, it starts to solve new problems and creates entirely new product categories because of novel properties. Lithium-ion batteries enabled drones, solar PV enabled human-scale power generation and synthetic polymers enabled all kinds of products to be better and cheaper. Better performing materials also matter for speed of adoption. All of these examples of technologies gained adoption very quickly because of their tech-driven differentiation.

It's fundamentally hard and pioneering work to develop to bring novel materials to market, whether they are drop-ins or performance-orientated. Yet mimicking the properties of a synthetic polymer plays to its strengths and not to those of biomaterials. One of the key differentiators of using engineered biology is enabling performance properties not possible with materials derived from fossil fuel feedstocks. Why can't we make synthetic polymers that have these properties? In short, they rely on a limited range of fossil feedstocks and there are only so many ways to economically process them. I wrote about this in Part One of this series. Engineering with biology allows more control over end products with bespoke properties using the nanomachinery of cells and enzymes. Drop-ins do not take advantage of this.

Drop-in materials are good. I understand some startups need brand interest to secure funding and/or recruit talent in their early years and to help move biomaterials forward. Yet as we’ve covered, I think the bigger breakthroughs and step changes in scale will be unlocked by leveraging superior properties.

Performance biomaterials

Let’s look a little closer at examples of performance materials and startups that are making them.

By performance, I mean higher performing along any dimension or property of the material. Performance biomaterials will unlock products that synthetic polymers and other materials simply could not be used in. Plastics themselves disrupted natural fibers, wood, metals and other materials during the 20th century because of their novel properties and production scalability.

Spider silk protein from AMSilk and Spiber

We are talking specifically about spider silk protein made from microbial fermentation. One of the appeals of this protein as a biomaterial is the exceptional properties, even straight from nature. Incredible toughness, high tensile strength, lightweight and compatibility with biology is quite the list. Also, being made from engineered microbial "factories" allows tuneable properties, no matter if the protein is spun into a fiber or made into a gel, coating or composite material.

Of the numerous startups developing spider silk protein into materials, AMSilk and Spiber have the most traction. We've discussed the growth of these companies above, and will focus more on material properties and commercialization here.

AMSilk is perhaps the most under the radar biomaterial startup I’ve come across given its age and relative traction. Founded and based in Munich, Germany, their beachhead market was as high value, low volume gels for cosmetics products. By 2017, their gel was already in 20 speciality cosmetics products. Early on, they also developed Medtech applications including a coating for breast implants, tested in vivo in preclinical trials, where the spider silk gel’s biocompatibility lowered inflammation and other post-procedure problems. Their fiber can be used for biosensors and even in 3D printable structures. By focusing on non-fiber applications, they avoided the challenges of material fabrication we spoke about earlier.

In 2015, AMSilk made their first fiber, branded as Biosteel. In the past few years, they’ve started partnering with several large companies and have multiple products in the market. Airbus is researching how to replace carbon fiber with AMSilk's fiber in composite parts on their airplanes. This would allow for planes to better withstand impacts and help with light-weighting and is a great example of how innovation in performance biomaterials will extend far beyond fabrics. Airbus's Innovation Manager for Emerging Technologies and Concepts, Detlev Konigorski, had this to say about spider silk:

We haven’t even begun to scratch the surface here. Ultimately, this material could enable us to approach design and construction in an entirely new fashion.

It bends without losing strength. So, it could be integrated on parts away from the fuselage that are prone to debris impact or bird strikes. It could help protect space equipment in a similar manner or be applied to defence products.

Omega launched a commercial watch strap line using AMSilk's Biosteel that is 30% lighter and hypoallergenic. Adidas has been developing a shoe upper with them and Mercedes is exploring using Biosteel in interior parts, starting with door pulls. To focus on these and other products in their pipeline, in 2019 AMSilk sold their cosmetics business to Givaudan, the world's largest flavor and fragrance company. Granted, these products are all premium for now, but that's largely due to AMSilk's still small, but rapid growing manufacturing capacity.

Like most other biomaterials startups, AMSilk is supply-constrained. They can’t produce enough protein via their CMOs to meet demand. To address this, they are in aggressive scale up mode, as mentioned in the growth section above. Their capacity was 1 ton last year. They expect a six-fold surge to 6 tons this year and are aiming for several thousand tons of capacity by expanding their CMO network beyond Europe.

I’m impressed with AMSilk’s execution. They've raised $42 million, with only $7 million of that lasting from 2008 through last year, when they secured their series C. This is far less than their peers Spiber and Bolt and at least in part because they outsourced manufacturing, which is a huge cost. Despite their frugality, they have the highest current output of a startup producing fermentation-based spider silk protein and are ramping scale-up. Starting with cosmetics and coatings for medical applications probably enabled high value / low volume production for long enough to move down the cost curve. This was harder for Spiber or Bolt Threads to do given their focus on lower value apparel products.

Spiber is the other pioneer of fermentation-derived spider silk protein, branded Brewed Protein. Founded in 2007 and based in Yamagata, Japan, they have largely been focused on the apparel market as their beachhead. They have also developed concepts for fiber-reinforced materials with Toyota group companies for use in car doors instead of steel that are lighter and better shock absorbing. In that same project, they also developed polyurethane composite foam for car seats with Bridgestone, the tire maker. In apparel, they partnered with Goldwin to make a wool blend sweater, and North Face on a limited edition Moon Parka with the outer shell made from Brewed Protein. The jacket was delayed for several years years because of “supercontraction", which happens when spider silk protein becomes wet or is washed. Despite this setback, their spider silk is now in a (slightly less limited) commercial run of several hundred hoodies from Pangaia Lab.

Spiber has raised a whopping $640 million in equity and securitization financing. A large chunk of that is invested into building their upcoming demonstration plant in Thailand and commercial plant in the US in Iowa. Once they have these plants operating at capacity, they will be producing thousands of tons per year of spider silk protein. As they approach that scale in the next few years, they aim to be cost competitive with cashmere, then silk, and then even wool. Despite setbacks, this is pretty remarkable given the founders produced a mere 20 mg in 2008.

Bolt Threads is another startup that started by developing fermentation-based spider silk proteins. After being founded in 2009, Bolt did some limited run experiments with their fibers in ties and beanies and later launched their B-silk protein for cosmetics creams and other products. However, they've pivoted largely to Mylo, a mycelium-based leather using tech licensed from Ecovative. As CEO Dan Widmaier put recently, "Mylo is where the heat in the industry is right now". Bolt has raised $472 million.

Spider silk protein has naturally exceptional properties, tunability due to synbio and downstream processing, the traction in an early range of applications and finally the rapid and continued scale up of leading companies. It demonstrates promise as a leading platform for performance biomaterials in terms of both scale and potential.

Algal oil polymers from Checkerspot

Founded in 2016, Checkerspot is a Bay Area based startup making performance biomaterials. It is one of the only biomaterials companies vertically integrated from developing microbial strains all the way to directly selling products to consumers.

They use microalgal fermentation to produce an algal oil, which can then be processed into different materials. So far, they have commercialized an algal urethane, polyurethane and a wicking fabric coating. Checkerspot’s founders, Charles Dimmler and Scott Franklin, had both worked at Solazyme, one of the Cleantech 1.0 companies that didn’t make it. Solazyme’s direct to consumer algal oil products inspired Dimmler and Franklin to focus on building consumer products themselves. They both love skiing, and thought the algal oil could enhance the properties of skis. Indeed, Solazyme had earlier tested algal oil polymers in surfboards.

Checkerspot’s materials used in skis replace petroleum-based polyurethane in ski sidewalls. The sidewalls are 138% better at damping vibrations vs petroleum-based ABS sidewalls and 60% better tensile strength at skiing-friendly temperatures. It also does not need any glue to adhere to the ski. Their polyurethane used in ski cores is also lightweight, allowing their skis to be significantly lighter than average skis. The most surprising part is their price competitiveness with performance skis. Better performance effectively at price parity. I don’t think any other biomaterial company can claim that to the extent they can. They're either not using enough algal oil for prices to matter, or their algal oil is not overly expensive vs petroleum-based polymers, or both.

Vertical integration stands out most about Checkerspot’s business model. Most of their skis are produced and distributed by their own brand, WNDR Alpine, which has a manufacturing plant in Salt Lake City, Utah. No other biomaterials company I’m aware of has taken this full stack strain-to-product approach. Checkerspot has also partnered with other companies to use their performance materials. DPS, a premium ski maker has integrated their polyurethane sidewall into a line of skis, ink and chemical company DIC is using algal oil to add more color to their range, Algenist set up a strategic partnership to use their materials in skincare products and Gore is experimenting with the materials too.

Alongside their own DTC distribution through WNDR Alpine, Checkerspot is also leveraging B2B partnerships. They are building out algal oil as a platform molecule that could be processed and integrated into many different consumer products.

Other emerging performance biomaterials

Alongside spider silk protein and algal oil products, there are a growing number of startups focused on making performance biomaterials.

When Amyris sent out samples of farnesene early in its development, they probably didn’t expect a company to commercialize a performance tire. Kuraray, a Japanese plastics manufacturer, developed two novel materials by blending farnesene into their rubber products. The first is liquid farnesene rubber (LFR). When used in tires, it reduces hardening in cold temperatures, improving grip, reducing fuel use and is more energy efficient to manufacture. LFR has been commercialized in some of Dunlop’s winter tires in East Asia. Indeed the demand was so strong that Amyris's CEO John Melo said (emphasis mine): "Using LFR has caused notable interest by leading tire companies and has resulted in substantially stronger demand in 2017 than either Amyris or our collaboration partner had anticipated." LFR can also be used  in adhesives, coatings and sporting goods.

Using farnesene, Kuraray has also developed hydrogenated styrene farnesene block copolymer (HSFC), which has superior grip performance for shoes, grip handles, mats, etc. It also has better damping properties. Kuraray’s farnesene derived materials are still combined with conventional polymers for now, but are produced more sustainably and with less emissions than petrochemical rubbers. Further, recycling tire rubbers could be much better, with LFR and HSFC being more incremental improvements. Still, they point to higher performance biomaterial improvements over commodity uses of plastics like tires.

The wide ranging performance applications for this one molecule developed by one company hint at the potential impact of biomaterials. Also, as drop-in chemicals and biomaterials scale up, expect to see more applications that integrate both drop-in and performance biomaterials together in products.

Tandem Repeat is commercializing Squitex, a composite biomaterial made by spinning squid-inspired protein and cellulose or acrylic together. The squid-inspired protein is cultivated in fermentation tanks using microbes engineered with tandem-repeated genetic sequences from squids. Hence the startup's name.

Melik Demirel is one of Tandem Repeat's cofounders and a professor at Penn State. He co-authored a a 2020 Nature paper demonstrating the rapid self-healing properties of other squid-inspired proteins. In the paper, the biomaterial locally heals in seconds, using only mild heat at 50°C/122°F, as shown in the video below with some tests in the paper showing healed areas being at least as strong as the pristine ones. State of the art comparable materials take at least an hour. The paper also reported experiments using an artificial muscle capable of lifting a dead load 3000 times its own mass, which exceeds biological muscle. This is shown in the image above.

Sorry, your browser doesn't support embedded videos.

Tandem Repeat is seeing strong interest in fashion brands looking to partner with them. Squitex's thermal properties can help keep clothing cooler when active. They are first targeting textile adhesives as a high value and low volume beachhead market, but may also look into PPA equipment and other markets that can best leverage Squitex's properties.

Humble Bee Bio is working on performance biopolymers inspired by the solitary Australian masked bee. The bee produces a polymer that is resistant to heat up to 240°C, is resistant to water, industrial solvents, acids and strong bases. Humble Bee Bio’s team are engineering yeast strains that can produce this polymer using fermentation. They are aiming to develop a proof of concept of their material to be used for fibers and fabric finishes by mid 2023, a 6-12 month timeline that would have been incredulous only 10 years ago. Humble Bee Bio has also just raised an oversubscribed round. Long term, they are aiming at a suite of biomaterials for textiles, medical devices, electronics, aerospace, defence, animal health and even construction. There's a ton of genetic potential in insects that could lead to performance materials that we just do not understand well today. For example, Raspy crickets make silk for shelter.

Bacterial cellulose is another biomaterial that has performance characteristics. Modern Synthesis is using bacterial cellulose to “microbially weave” materials that allows for more design freedom in textiles and films. Bucha Bio’s bacterial cellulose has tensile strength higher than polyurethane, and is targeting textile markets. Polybion has built a first of a kind plant that will produce 1.1 million sq. ft. of bacterial cellulose per year, with production and polymerization only taking 20 days. They are partnering with the world's largest shoe manufacturer and three largest German auto companies. Simplifyber's cellulose is plant-based, but it is poured into a 3D-printed mold to make a one-piece sneaker, shirt or other clothing.

So far, we have not talked about mycelium in this section. That’s because mycelium applications have usually not focused on performance. Mycelium leather in hats and other clothing, meat alternatives, other textiles, foams and packaging are some of the commercialized products so far. Though the focus has been on drop-ins and matching performance, there may be structural materials being explored for mycelium that have performance applications. As mentioned above, MyForest Foods' production capacity is aggressively ramping and will continue driving costs down.

Engineering mycelium and other biomaterials for performance should become easier as costs continue falling, processes improve and as our underlying scientific understanding improves. It bodes well for future applications.

Challenges for performance biomaterials

Like any new and growing industry, progress will not all be smooth sailing.

The biggest question perhaps is scaling and associated cost declines. Can spider silk (and mycelium) keep scaling production at a rapid rate? Can other startups follow and also scale production? I don’t know, and no one does. There are no perfect comparisons, but overall the signs are good. As discussed earlier, Spiber, AMSilk and Ecovative all have massive near-term scale up plans. Synbio chemical companies like Amyris, Geno and Lanzatech have demonstrated the ability to reach tens or hundreds of thousands of tons of capacity. They are still growing fast despite being in mostly commoditized markets. Performance biomaterials should have an easier time scaling in comparison because their tech-differentiated value will accelerate demand.

Related to this question is the bottleneck of fermentation capacity. While I attended the SynBioBeta conference in April, this was one of the most talked about challenges. It's worth remembering that supply constraints is a good problem to have overall and means biomanufacturing is booming. Demand for fermentation is strong at bench scale up to tens of thousands of liter tanks and beyond, with 6 to 12 month or more lead times. Well capitalized companies are building out more capacity (like Amyris, Geno, Spiber) and upstart CMOs are looking to new reactor designs and to embed more automation and data tracking. However, there’s a lag for investments in supply to catch up, especially for larger scale fermenters. Liberation Labs, started by experienced synbio executive Mark Werner, as well at least one funded stealth startup I know of is working on addressing this need. The lithium supply chain is a great analog, where investment has lagged surging EV-driven demand in recent years, but it is starting to catch up. Still, for fermentation capacity, there will be some short term pain before supply catches up.

For startups with lower technological readiness, there’s the technical risk of developing a proof of concept for a new biomaterial and bringing it to market. Titers have to dramatically improve from first results in a lab. However, unlike Spiber’s early days, the synbio stack makes it faster and cheaper than ever to develop new biomaterials. We saw this happen for Amyris and Geno above as they developed more molecules. Development and go-to-market timelines will keep falling and make it easier to commercialize performance materials.

A huge challenge in executing quickly is working with incumbent customers on their usually slow timelines, as discussed in the section above on drop-in materials. Leveraging DTC like Checkerspot or partnering with startups, both of which we’ll cover later, are ways to speed up iteration cycles.

The startups we’ve discussed so far mostly use sugars from crops as feedstocks for fermentation. Land and resource use of crops comes up as a problem as production capacity of fermentation scales. Corn, sugarcane and other feedstocks require land, but sugarcane expansion in Brazil, a key producer, is largely done by converting degraded livestock pastures. Currently 77% of all crop land is actually used for livestock pastures or crops for animal feed. Even today's biomanufacturing technology reduces land intensity. As an example, Impossible Foods's burger patty uses 96% less land to make vs beef. Many biomanufactured products from chemicals (Lanzatech) to foods (Air Protein, Solar Foods, Circe Bioscience) are using feedstocks like CO2 or carbon monoxide and hydrogen. This trend will only continue as scientists better harness types of bacteria which thrive on simple carbon-based molecules. Costs of hydrogen and other feedstocks will also fall with clean electricity prices, making these approaches more competitive over time.

Also, we can’t talk about biomaterials without mentioning Zymergen. From the outside, Zymergen looked like a cautionary tale of how not to execute on a new biomaterial. Their lack of success should not be a deterrent for startups building biomaterials. If anything, having a failure like this, reminiscent of Cleantech 1.0 companies, is probably a good thing for the companies actually building the future. I looked into Zymergen a little more closely in this Twitter thread. Ginkgo’s recent fire sale acquisition seems like a good way to bring on the IP from Zymergen's platform and gain their core engineering talent.

Future biomaterials will keep getting better

The present and near term future for performance biomaterials looks pretty good. Looking forward, we are in the very early stages of the potential of better properties at increasingly economic costs. Let’s briefly survey some of the emerging breakthroughs that may enable next decade’s biomaterials to well exceed what's possible today.

The synbio stack for biomanufacturing is only getting better. I'll recap what was covered above in the synbio stack section. Strains are faster to engineer and more automation and software is improving productivity of biomaterial startups. Fermentation capacity, though bottlenecked today, should ramp in the coming years with more investment to keep up with the growing range and scale of biomanufactured products. Startups, some in stealth, are working on improving the design of bioreactors and making scale up easier. Using microbes that feed on CO2, hydrogen and other small-molecule feedstocks could allow for novel materials. Further, current solid state fermentation and cell-free catalysis will also keep improving as more companies commercialize molecules using these technologies.

Biomanufacturing has relied on e. coli and yeast (s. cerevisiae) as workhorses in microbial fermentation. Almost all companies today use engineering strains of these two rather versatile microbes. Increasingly, new ventures are developing so called non-model organisms, which open up the design space for potential molecules (and materials) that can be made. We’re in the early stages of this, but startups like Wild Microbes, focused research organizations like Cultivarium and no doubt many research labs are pushing forward our understanding of non-model organisms.

Using cell-free enzyme catalysis approaches are also in their early days. A prominent startup is Solugen, a chemicals company, with a 10,000 ton capacity plant built last year in Houston. Startups like Rubi are developing enyzme-based pathways to make materials, while Invizyme and Aether are developing enzyme engineering platforms. Intropic Materials is using enzymes embedded in plastics to accelerate self-degradation.

As promising as biomaterials are, there are other performance-focused materials that could be key parts of a new materials paradigm. Let's take a look at these, before turning to how consumer-facing startups can partner with biomaterials producers.

Other performance materials

Graphene, which was first isolated using scotch tape, seems to have already survived a hype and trough cycle. Some of the remarkable properties of this one-atom thick 2D material include its incredible strength, hardness, elasticity and lightness. Applications are now booming as graphene is now hitting its own s-curve inflection point. Last year twelve thousand tons was produced globally as films, powders, nanoplatelets and in other forms. In the last three years according to IDTechEx, graphene production has surged over 4x, at a scorching 61% CAGR. It is scaling almost twice as fast as nylon or synthetic polymers overall at the same level of scale.

All that graphene is being used for supercapacitors, battery cooling, coatings for wind turbines, Head tennis racquets, and even in composites for Airbus and Fiat-Chrysler. Challenges for graphene include only being used as an additive, and still relatively high costs. But as production ramps, costs will come down and engineering will improve and we’ll likely see a lot more of this material used in the coming decades. Graphene manufacturing has historically relied on chemical vapor deposition, which is also used for semiconductor and solar cell fabrication but produces volatile by-products or CO2. There are alternative manufacturing processes being developed like flash graphene and using palm kernel shells as feedstocks.

Commoditized metals are probably not the first thing one thinks as performance materials. Yet, usage of many metals is constrained by price and supply, and often manufacturing processes are emissions intensive. Electrolysis of aluminum has been around for over a century, but electricity that gets cheaper and cleaner will unlock even more uses for this lightweight metal. Other metals that could see higher adoption as costs fall include magnesium, which is lighter than even aluminum and has superior damping, or titanium using hydrogen assisted reduction. A future where electric planes are made of lower cost lightweight metals and biomaterial composites is no longer science fiction.

Like graphene, additive manufacturing (AM or 3D printing) has been through a hype cycle and is coming out stronger. Though not new materials, AM is a range of processes that can improve structural properties of materials in novel ways. In metal AM, one promising startup is Seurat, named after Georges Seurat's pointillism-style paintings. Seurat's printers work by using a laser to micro-weld powder layers with clean electricity. They aim to aggressively bring costs down and scale up so that they can print even cheaper than commodity kitchen silverware by 2030. They have already partnered with established companies like Porsche, GM and Siemens. Another interesting startup is Relativity Space. They have engineered the world's largest 3D printers to print their rockets, including engines, with 100x less parts.

Meanwhile, AM to make synthetic polymers is also starting to scale quickly. The global polymer AM market is at a similar scale production to graphene, though growing at a slower 22% CAGR from 2016 to 2023 (estimated).

How can startups apply performance bio (and other) materials to create far better products? This question is one I’ve perhaps been thinking about the most as I’ve explored the terrain above. Let’s dive in.

Go-to-market and scaling demand

This decade is the first time since the age of polymers in which new classes of materials will help drive the performance of consumer products. Above, we’ve already surveyed performance (bio and non-bio) materials. Their novel properties open up the design space, much as synthetic fabrics and resins did for textiles, the Space Race and beyond during the mid 20th century.

Let’s explore strategies for biomaterials startups to go to market. The sum of these strategies help inform how demand scales up for the whole industry. We'll first outline two types of strategies used today. Then, we'll hone in on why I think startups sourcing from biomaterial partners will increasingly be a compelling alternative. Finally, we'll also talk about why DTC as a distribution model is important for driving growth.

Keep in mind that while I believe some approaches to be better than others, it’s better to have a plurality of strategies to see which ultimately work. As discussed in the section on drop-ins, there needs to be many shots on goal. I am just sharing what I find most interesting as I explore startup ideas.

1) Partnering with incumbent brands

This is by far the most common strategy. Practically all biomaterial companies partner and sell their products to established companies in apparel, biomedical, automotive, aerospace and/or other industries. Benefits include incumbents having deep strength in making and distributing consumer products to their existing customer bases. Smaller brands are less exacting in the specifications they have, and the number of product units they need to move. Both Pangaia Lab and Goldwin's partnership with Spiber are good examples of this.

There are some drawbacks in my opinion, especially in working with larger companies. Perhaps the most limiting is that these companies have limited freedom to use completely different materials and properties to what they've used historically. Dan Widmaier, co-founder and CEO of Bolt Threads, had this to say about what incumbent buyers said when asked what they wanted from spider silk:

The number one request we used to get in the beginning of microsilk was "I would like merino wool but cheaper".

... that’s not very creative. Humans are good at incrementally innovating on something they know. There’s very few people who are very good at striking out into whitespace and correctly guessing the future on a product.

Also, incumbents want to adopt technologies that work within their manufacturing processes:

Something we really look for in terms of integration is that partner or startup understanding the supply chain and where they fit in. We've often struggled if partners just think that we’re going to reinvent the wheel and create a whole new supply chain.
— Adidas spokesperson. Biofabricate

Further, they move slowly and will likely take several years to bring a product to market, especially if it’s not in apparel. They are not likely to throw much of their marketing weight behind the product(s). Their large, established customer bases are as a whole not interested in products better suited for early adopters.

Early startups often simply don’t make enough material to use viably in their partner’s products. Patagonia needs thousands of units to launch a new product, Adidas aims to sell hundreds of thousands of shoes. Incumbents often ask for exacting specifications that are hard for startups to meet, and that's before talking about (current) price premiums. Partners may also have unrealistic expectations of scale up speed before the producer has even tuned its pilot or demonstration scale process.

The biggest challenge that we have now internally is everyone's kind of tired of doing pilots. We did enough of that and we just really want to be able to start offering products.
— Spokesperson for Stella McCartney (a fashion brand). Biofabricate

2) Vertical integration

In this context, vertical integration means doing the R&D and commercialization of both the molecule and the material, as well as owning direct to consumer distribution channels. One of the only, perhaps the only, example of this today is Checkerspot with their own WNDR Alpine brand.

Checkerspot's R&D expertise in algal fermentation is exceptional, as is their materials science to turn their algal oil into high-performance materials like the polyurethane used in WNDR Alpine’s products. WNDR integrates these biomaterials into skis and snowboards, and sells them at very competitive prices on their site.

Checkerspot’s vertical integration is impressive. I’ve seen few examples of synbio companies generally that vertically integrate, let alone biomaterials. The ones that seem to be doing it well like Amyris, Zbiotics and Cronos are integrating chemicals into cosmetics, food or health products.

However, I don’t think the vertically integrated approach is reproducible for most other founding teams. Checkerspot's founders, Charles Dimmler and Scott Franklin had both worked at Solazyme, also an algal oil company. Franklin has credited working on their consumer product as an inspiration for Checkerspot's DTC strategy. Despite being a relatively young biomaterials company founded in 2016, they were able to leverage the many years (and millions of $) of development that Solazyme had invested into algal oil R&D. As Dimmler said:

(Solazyme) really started to pick up momentum in 2008. So if you were to start the clock it took from 2008 to 2014/ 2015, about six plus years to get to commercial scale manufacturing. Checkerspot is building on a lot of the know how and lessons learned from that experience.

...that's a really important qualifier because if I'd say Checkerspot was founded in the summer of 2016 — we're four years old and we have three products on the market, that would convey that Checkerspot has somehow done something extraordinarily unique in bringing a product to market at a ridiculously fast pace.

Further, I think we'll increasingly see products integrating multiple biomaterials from more than one producer. For Checkerspot, it might be harder to source from another biomaterials manufacturer given it’s tough to be both the platform and the buyer. There are exceptions of course, like Tesla sourcing batteries from BYD, which is both an EV and battery manufacturer. Overall, it's hard to see many performance biomaterials companies going down the vertically integrated path.

3) Partnering with consumer product startups

Spiber and Ecovative are the only two examples I've seen of biomaterials producers partnering with consumer product startups. Both partner with Pangaia, a four year old materials-focused fashion brand, and Ecovative's collaboration is through their Fashion for Good cooperative. Despite the lack of current examples, let's walk through how these partnerships could look in the future and some potential differences in working with larger brands.

The core benefits of this approach are that the consumer product startups can build products that pull demand forward for biomaterials, and scale with the biomaterials producers. The producers of performance biomaterials that can grow faster will move down the cost curve sooner. Partnering with startups all in on making products that leverage the performance properties of biomaterials will, I think, grow demand faster.

Consumer startups can deeply innovate on the integration of one or more performance biomaterials into novel products. They are free of the creative constraints, large volume requirements and exacting requirements of incumbent companies. They have to ramp distribution as part of building the product, and can develop a tight customer feedback loop akin to what companies like Allbirds, Tesla and Fitbit did. These startups would bet their businesses on biomaterials as being core differentiators for not only their product, but their company.

Novel materials could get early adopter customers really excited to buy high performance products. Of all things, nylon stockings triggered this in the 1940s and I’ve written that few materials-driven innovations have done that since. I believe sustainable materials are becoming table stakes, and will not offer real benefit when everyone else uses them. The challenge will be in making products that people would love to use using world-class engineering.

Startups focused on engineering the materials into products can also cultivate strong relationships with many producers of performance biomaterials. Over time, approaches like cell-free, small molecule feedstocks and superior solid-state derived biomaterials probably become more available. No problem, as these consumer-facing startups can develop the expertise to integrate these as well and do it fast.

These consumer-facing startups could form in a range of industries, though the immediate pull might be towards apparel. However, areas like next generation mobility form-factors, aerospace and even robotics and other hardware products could be improved with performance materials as we’ll talk about shortly. The likely early products will be higher value goods made in reasonably small volumes, perhaps with proof of concepts engineered with biomaterial samples. As biomaterials prices will continue to fall as production ramps in the coming years, consumer startups will not need to auction jellyfish collagen book covers for over $12000, or charge $1400 for limited edition North Face jackets to gain traction. AMSilk and Spiber are already eyeing the aerospace and mobility industries as they scale, with AMSilk having partnerships with Airbus and Mercedes in place. This hints at where growth might be beyond apparel and textiles, though I think an eVTOL/eCTOL startup or e-bike/EV startup could build more compelling products that gain traction faster than incumbents.

There are companies in adjacent spaces emerging that are useful analogs as consumer facing products. Arcaea is leveraging Ginkgo’s strain development platform and is focusing on formulating fermentation-derived chemicals into cosmetics products. Startups like Vollebak and Graphene-X have been building brands around graphene in textiles and other novel materials, having kickstarted and preordered their way to early traction. Allbirds and On Running became billion dollar public companies (On expects to do over $1B in revenues this year) despite using existing materials in their shoes developed with novel approaches.

In the final section below, we'll take a longer view at what the potential of performance biomaterials could be.

Future applications

The early examples of consumer products using performance biomaterials are in apparel or sporting goods, e.g. Checkerspot. But what other types of products might biomaterials begin to be used in? This section outlines my far-from-comprehensive thoughts on this question, and how biomaterials might contribute to a more exciting future. The obvious narrative would be that they solve valuable problems that people will pay to be solved, for a growing variety of problems and larger groups of people over time. I think it’s worth exploring in a little more detail. By the way these are not predictions!

We’re still very early in the rise of performance biomaterials and other next-generation materials. Given that, it makes sense that the products available today are very much trying to do existing things better instead of brand new things. Over time, we’ll see a self-reinforcing feedback loop: more startups creating more products, increasing demand for more biomaterials, which will drive more product adoption. That will, like how other technologies have diffused historically, allow for increasingly complex products to be engineered that integrate performance biomaterials.

Two things about how this could play out. Firstly, 1+1 > 2. Integrating multiple performance biomaterials in one product could enhance benefits over and above applying them separately in different products. Of course, that’s with the caveat that the characteristics of the biomaterials combined makes sense for such a product.

Second, engineering biomaterials together with other technologies that make products better. Skeleton Technologies, which has raised over €200M making graphene-enabled supercapacitors, is an example of leveraging performance materials with supercap technology. Technologies that might play well with performance materials include battery-powered propulsion, miniaturized hardware, exponential compute, biosensors, robotics and more.

Smart clothing, in which textiles are integrated with electronic components, could continue to integrate performance materials. Self-thermally regulating clothing brands like Ministry of Supply's Mercury jacket have integrated batteries, heating elements and use a thermostat to heat to a comfortable temperature. Oros Apparel uses a super thin NASA-inspired aerogel layer to trap heat instead of thick down in their jackets. In some use cases, it just makes more sense to heat (or cool) one’s body directly instead of a heat pump for an entire room/home/building.

Performance biomaterials could also allow smart clothing to do much more. They could one day repair itself as discussed above for Tandem Repeat's squid-inspired protein. Graphene electrodes in a biosensor monitoring amino acids in sweat can give nutrition-tracking and recipe recommendations, as reported in a recent Nature paper by Caltech researchers. Biomaterials could even help brain-computer interfaces be more biocompatible in a similar way to what AMSilk’s spider silk does today for breast implants.

Mobility is another promising area, especially electric vertical/conventional take off and landing (eVTOL/eCTOL) aircraft. Lightweight biomaterial composites, like what Airbus is exploring with AMSilk, could be used as coverings, propeller blades or interior parts for eVTOLs and eCTOLs. Given how weight-constrained these vehicles are during takeoff and for flight range, the likely cost premium (for now) in using these composites could well be offset by range and weight benefits in the coming years. Improving battery performance will help. Airbus's partnership is to explore spider silk alternatives for carbon fiber reinforced polymer (CFRP) composites. Aerospace grade carbon fiber is about $90/kg today, less than cashmere and a little more than silk. As AMSilk, Spiber and others start to scale to tens of thousands of tons per year scale of spider silk protein in the coming years as planned, it should become cost-competitive and perhaps even more affordable than carbon fiber sooner than we think. This is great for commercial planes, but it should be even better for startups working on eVTOL/eCTOL or other aircraft with performance constraints. Other vehicles that could benefit from biomaterials include e-bikes, EVs, and even yet-undeveloped form factors.

Another application for novel biomaterials could be humanoid robots. They are closer to market and usefulness than we probably think, partly because of rapidly advancing AI algorithms and exponential compute. Biomaterials could play a key role in soft robotics. In the Nature paper mentioned above, the squid-inspired protein was tested as a pneumatic soft actuator that, when self-healed, still performed just as well as when it was pristine. These bioactuators allowed for experimental success gripping objects with non-flat shapes such as a cherry tomato. Further, the lightweight but high toughness of spider silk protein and other biomaterials could apply well to nimble and durable robots.

Over time, performance biomaterials will be better understood. As this happens, they will become integrated in increasingly complex products with other emerging technologies.

Conclusion

A future filled with performance biomaterials is coming fast. These materials are the tip of a new materials paradigm spear. Startups will integrate them into compelling new products, industries will be shaken up by them, and I am optimistic that people be more inspired to build the future.

Personally, I am in the early stages of exploring ideas for a consumer startup integrating performance biomaterials. Let’s talk more about this if interesting to you.

Reach out on Twitter or email.

Thanks to James Giammona, Eben Bayer, Catherine Tubb, Henry Lee, Yang Fan and Anna Delas for feedback on earlier drafts.

Over eighty years ago, the rise of synthetic polymers led to one of the most successful original products ever. This product used nothing that we would consider to be technology today. That's how long it’s been since there have been any material breakthroughs at scale.

Few people seem to question why we've not seen much progress in materials. It's easy today to think new materials are not that important. But as I've looked more into their history and potential startup opportunities, I increasingly think they will be.

This article is the first of a two part mini-series on a new and emerging materials paradigm. If someone had written similar articles, I would have simply shared that and not written them. But I did not come across anything satisfying enough. If you know of any, let me know!

Here in Part One, we’ll steelman the case for a new materials paradigm and why it matters. Part Two looks mostly at emerging materials, especially biomanufactured ones, and what they could enable.

I don't have all the answers here and I'm learning myself. For a long time, I did not really question why materials science and technology had or had not progressed. It just did not seem that interesting, and I had drunk the kool aid on the primacy of progress in the world of bits. But, what I have realized recently is that better performing materials will actually be a pretty big deal. As I have gone down this rabbit hole, I have become fired up about the opportunities I've uncovered by exploring ideas in this space. As has been true in previous energy transitions, new materials will become essential inputs to manufacture products in the coming decades.

I want this blog mini-series to clarify my thoughts and to also invite you to explore the possibilities of better, sustainable and high performance materials.

Contents

• How, the rise of synthetic polymers led to the commercialization of one of the most successful original products ever.

• How synthetic polymers were the most recent materials age and the problems that arose from this.

• Why progress in new materials has stagnated, including the limitations of fossil fuels as feedstocks.

• How companies like Allbirds and On Running had impressive traction (no pun intended) by using existing materials in new ways and using direct to consumer distribution.

• How products integrating software, hardware and other engineering might inform the future of products with next generation materials.

• Introduction to Part Two. How biomanufactured and other next-gen materials will create better, higher performance and eventually cheaper products.

An outsized product success

What comes to mind when you think of generation-defining first of a kind products? Maybe the original iPhone. I definitely would have thought the 2007 iPhone or perhaps a consumer electronics product.

Apple sold about 700,000 units of the original iPhone in the first weekend. Nintendo sold 300,000 units of the original Gameboy in the first two weeks. Tesla had 276,000 preorders for the Model 3 in the first three days. Candy Crush Saga had 4 million users within the first few weeks.

Yet this product I have in mind, developed and commercialized during the rise of synthetic polymers, held its own against any of those above. In the first few days, this product sold five million units; comparable to sales of the iPhone, Gameboy, the Model 3 preorders and number of early Candy Crush users... combined. It retailed for about $1.15 in 1940, or $24 in today's dollars. By the third full year of availability, 360 million units were flying off the shelves per year, good for over $6 billion in today's dollars. Original iPhone aside, I think only a handful of first of a kind products have had steep enough growth rates for revenue numbers like that in their third year. What makes the success of this product even more remarkable was that this happened before 1950.  

That product was nylon stockings. It's probably one of the least well known product successes in history.

During the 1930s, silk was the incumbent material used in stockings. Silk was largely imported from Japan. Americans became less and less enthused about this as war tensions rose in the late 1930s, helping nylon's rise. Also, silk didn’t last very long and it tended to “run” once torn or ripped. Nylon in contrast, lasted longer, was strong enough to move a small car and didn’t run as easily. It had superior material properties for this market. Also, being synthetic with fairly price-stable inputs, it didn't face the same price volatility as silk supply even though it was more expensive than silk. Dupont saw the potential of their new material and executed fast.

Nylon stockings became available in retailers in 1940, a mere five years after being invented. Five million pairs flew off the shelves in the first few days. The following year, production was diverted for the war to make parachutes, ropes, mosquito netting and even flak jackets. After the war ended, Dupont again turned to producing nylon stockings in late 1945 but could not ramp supply fast enough. Nylon Riots happened in late 1945 as women fought over limited supplies. For example, On one day, 40,000 people in Pittsburgh waited in a line over a mile long just to have the chance to buy a pair.

Dupont's success with nylon stockings is perhaps shown in terms of how quickly a majority of women started wearing them. Nylon stockings sold for $1.15 a pair in 1940 or about $24 today adjusted for inflation. Sales in 1946 were over $400M after Dupont ramped production to 360 million pairs per year. Similar volumes were sold until the early 1950s. There were 42 million American women 21 and older in 1940 and on average women were buying 15 pairs of stockings per year. So in the late 1940s, there were enough sold for 24 million or 57% of US women to exclusively use nylon stockings.

This pace means nylon stockings were adopted faster than all of the following products, even smartphones and tablets:

Nylon’s early success shows how much people could value a product made better with superior material properties. Outside of a few high value low volume industries that rely on advanced manufacturing today, startups largely no longer think of new materials as a key factor in building better products. We just have not had many new materials that have seen enough demand to come down the cost curve to produce at substantial volume.

Subscribe now

Synthetic polymers: the most recent materials paradigm

During the early and mid 20th century, an increasing variety of synthetic polymers were invented and found many uses. It was a golden age of new materials with better performing properties.

Bakelite, the first synthetic polymer, was used in everything from radios to telephones and many other products from the 1910s thanks to its non-conductive properties and ease of production. At the peak of polymer discoveries during 1930s to 1950s, scientists at companies began to commercialize polymers like polyethylene, polypropylene, polyurethane, teflon, polyester and nylon as mentioned earlier. Low density polyethylene was a military secret used in Allied radar insulation during WW2 and later in kitchenware, toys, prosthetics and of course, packaging. Teflon, which was also used secretly as a coating for uranium parts in the Manhattan Project, before becoming synonymous with Tefal "non-stick" pans in the mid 1950s.

The early 20th century saw a veritable Cambrian Explosion of materials. Synthetic polymer production doubled every 2.5 years (at 31% CAGR) between 1911 and 1950 from 45 tons to 1.5 million tons. As their usage grew, costs came down and quality improved in a flywheel effect that opened up even larger markets. Note that since 1950, production has slowed down almost every decade. The 1950s saw a CAGR of 18.2%, whereas in the 2010s, it was 4.4% off a base that is two orders of magnitudes higher.

One way to capture the incredible success of plastics is to consider the speed and scale that production is at today alongside other foundational materials.

Today, the world produces more plastics per year than almost every other material. Only steel, iron and materials used for steel or concrete inputs exceed plastics production by mass. Plastics output is about 3x that of aluminum, which is the metal with the second highest production volume after steel. The value of the plastics industry is over half that of the global steel industry.

Perhaps the best indication of success is how invisible a technology becomes and the extent to which most people take it for granted.

We rarely stop and wonder at how steel and concrete came to be all around us. Or how the internet has enabled far more and higher bandwidth communication across time and space. The same is true of synthetic polymers.

If we didn’t have them today, think about all the products that would be more expensive, would be worse products or even would not economically exist. On average, 330 pounds (150 kg) or 8% of a car's weight is plastics for fuel efficiency and light-weighting. This share will rise with soaring EV adoption. Without them, most cars would be significantly heavier, more expensive to fuel up and run, and require more robust structural components, tires and suspension. Goods like appliances, many electronic devices and even shoes would need to use more expensive natural polymers. These natural alternatives are also resource intensive and prone to bad crop harvests. Before Bakelite, insulating resins were sourced from shellac from the lac beetle. Ivory from elephants were used for billiard balls and natural rubber from trees. To provide insulating materials for wires and power boards instead, we'd need natural alternatives at low costs that are likely unprofitable. Polyurethane-made memory foam mattresses would not exist, and we’d have to find an alternate source for furniture foam cushions. These examples are just the tip of the iceberg for the vast array of synthetic polymers used in practically every industry today.

In addition to better properties, plastics have become a part of so many products because of cost. They've become so inexpensive that commoditized plastics are synonymous with being cheap. However, cheap materials does not have to imply poor quality. The success of a new paradigm of materials would hinge upon the ability of many of those materials to become cheap, ie come down the cost curve, and to offer better properties than plastics and other materials today.

Problems arising from world-scale plastics use

Globally, we’re now using an estimated 438 million tons of plastic per year. Barring better alternatives, we're on track to quadruple that by 2050. Much of that tonnage ends up in waterways killing animals, accumulating in microplastics with unknown health risks or landfills where most do not degrade. End-of-life processing has become a serious problem. Plastics production uses about 8% of global oil demand, and helped continue the rise of fossil fuels in the mid to late 20th century.

In 2018, packaging generated 46% of global plastic waste but was only 36% of global production. This gap arises from a short-used lifespan of most plastics for packaging and almost no recyclability. This needs to improve, but it's hard to see much progress today where a mere 10% of all plastics produced since 1950 have estimated to have been recycled, let alone contribute to a more circular economy.

We can acknowledge the negative impact on climate and environment whilst also appreciating the progress and innovation that plastics has enabled. New, better and cheaper materials would resolve the glaring problems with plastics whilst unlocking even more benefits. We’ll talk more about that in Part Two of this series.

Along with steel, concrete, aluminum and silicon, plastics are foundational and essential materials for how we live today. But there’s one pressing question. If new materials are so important, why have we not seen new and better materials in the many decades since the rise of synthetic polymers?

Progress in materials has stagnated

Why? From my research, it seems largely from a lack of new raw inputs and technologies to process them. Yes, fossil fuel feedstocks have been vast and relatively cheap, disincentivizing most alternatives. Significant R&D in polymers, especially prior to the 1980s, helped commercialize the plethora of plastics used at scale today. But fossil fuels can only produce so many types of useful polymers economically. A wider diversity of feedstocks are needed.

As we learn to use new inputs, we become better at finding new material applications and integrating them into useful products. This has happened since our early ancestors chiseled stone into tools and spears, cut timber for shelter or fire, and weaved fibers for clothing.

Innovation in materials, like other technologies, follows an s-curve of adoption. We showed the s-curves above when comparing nylon stockings with other products, but it also applies when zoomed out to consider energy and materials technologies. The materials that scale to millions of tons of production tend to be coupled to the cheapest energy inputs available at their time. Coked coal heated steel furnaces, electricity powered aluminum smelting and oil and gas powered polymers. Like with new energy sources, new generations of materials have displaced uses of earlier ones and created completely new uses. As we discussed above, the novel properties of synthetic polymers at low enough cost allowed it to replace uses for timber, metal and other materials in some cases. These benefits also enabled them to create new applications that simply would not exist without them. Today, biomanufactured materials are so promising because they can potentially replace usage of plastics with molecularly identical alternatives and create entirely new applications not yet possible.

With plastics, the inflection point of the s-curve was early in the 20th century. This was when Bakelite and later nylon and other synthetic polymers quickly found uses because of novel and better properties. The problem was that the key inputs, fossil fuels eventually become constraints. They could only be wrangled and processed into a limited number of different molecules and hence materials. Despite all of the ways plastics are used today, this is still a key limitation.

High and/or volatile prices of oil and gas for material inputs and process heat may also limit the growth of the plastics industry. We’ve seen this play out in the oil shocks during the inflationary 1970s, and in a more limited way during the late 2000s. Unlike those periods, there’s good reason to believe the current high oil and gas prices will persist. To be fair, humanity has managed to scale oil and gas production for over 100 years, no simple feat.

However, demand is falling slowly today and will accelerate largely because of the rapid electrification of mobility. EVs have already lowered oil demand by 3% globally, with 2/3 wheelers having the most impact and cars only beginning to contribute. This sounds like it would put lower pressure on prices but will likely have the opposite effect. New supply in fossil infrastructure is very constrained as companies avoid investing in assets with multi-decade payoffs that rely on that shrinking demand instead boosting shareholder returns.

In addition, drillers have been bitten by several boom bust cycles in the last decade, and their fuel, chemical, steel and labor costs are rising. Gas shortages, especially in Europe, can’t be replaced with other energy sources overnight, especially for Germany’s industrial reliance on gas. These factors could keep gas prices elevated for years. It's more likely than not that high prices of fossil inputs for input and process heat for polymers are here to stay.

This puts polymers in a weaker competitive position against any new emerging materials. We’ve seen many false alarms for high prices of incumbents accelerating a shift to replacements and none more recent than cleantech 1.0’s excesses. Yet when the new technology has been ready, a glance at history shows many examples of transitions away from incumbent products accelerated by supply constraints and high costs. Coal’s rise in 17th century England, synthetic rubbers and a vital enzyme in cheesemaking are all examples where upstart technologies have benefitted from supply constraints.

By the 1980s, companies developing polymers saw growth slowing and they had begun cutting R&D. At that point, engineers and scientists had already picked most of the low hanging material fruits off of petrochemical tree of materials. As William Nugent, an organic chemist at Dupont for 25 years explained:

R&D wasn’t as sure a winner as it had been in the 1950s and 1960s.

During that golden age, “there were things you could do with cheap building blocks that were abundantly available to the industry...”. But in more recent decades, “... you can’t afford to go down and mine exploratory chemistry anymore.”

Since fossil fuels, we have had no new inputs to produce better materials at scale. There are some exceptions. Silicon has achieved widespread adoption as the material that practically all semiconductor chips and 95% of solar PVs are built on. Lithium ion battery packs seem like they’ll increasingly do double duty as structural materials in cars and in the future in eVTOL and eCTOL vehicles. They can simplify engineering, reduce weight and save on material costs. Performance composites and materials like carbon fiber have found mostly niche though very useful applications. However, these materials largely do not replace the need for the expanse of material properties that plastics offer for a plethora of products today.

For the first time since the age of synthetic polymers, there are now a rising number of new inputs to produce new materials at scale. Synthetic biology will likely be one of the most important enabling new technologies. It opens up the potential for a huge range of cost-effective processes for producing new materials that we simply could not have had access to with fossil fuels. I'll be talking more about these new materials in Part Two of this article series.

How have startups created value in industries where materials are essential to product value? As we'll see next in a couple of startup case studies, some have been able to harness better ways to use existing materials and using direct to consumer distribution that have only become possible in the last dozen years or so. These startups have been able to make high performance and high growth products even with the lack of innovation in materials.

Subscribe now

Current paradigm: using existing materials in better ways

Unlike Dupont's nylon stockings, startups from the last decade or so have not enjoyed the benefit of being founded during an emerging new materials age. Instead, they've had to focus on other ways to differentiate themselves. In industries in which materials drive the value of a product, startups have focused on novel uses of existing materials and tapping newer distribution channels like going direct to consumers (DTC).

We'll now look at Allbirds and On Running, two examples of recently public companies applying materials science and technology to shoes. The product vision, execution and distribution that have made these startups standouts in new consumer products will be table stakes for the startups of tomorrow using new materials. Being a first mover to utilize novel materials would just amplify potential value creation. Because of that, I think new startups can learn a lot by looking at what made these two companies successful. Let's, er, step in to it.

Allbirds

Tim Brown, the Kiwi founder of Allbirds, had wanted to make a more comfortable all day shoe for years. Years before starting the company, he had realized using an engineered merino wool fabric was a great start, improving the feel of the shoe’s upper. This also took advantage of merino’s naturally useful properties like wicking, odor resistance (no socks!) and a soft, plush feel. After scoring almost 1000 preorders in a 2014 Kickstarter, he reached out to Joey Zwillinger, his later cofounder and an industrial engineer. Joey had worked at Solazyme, a now-defunct synthetic biology startup, and has said just how important performance materials were to Allbird’s product development:

But of course, Allbirds was only able to draw from the same set of materials that other shoe companies had access to before. Their key product insight was in redesigning an existing material (merino wool) to be suitable for shoe uppers, which was a world first.

On Running

In the late 2000s Olivier Bernhard, the Swiss brainchild of On Running, was searching for a better running shoe. An engineer had finally persuaded him to try a radical design: a sole made from cut garden hose tubes. Bernhard was skeptical but felt the surprisingly value of cushioning and bounce from running in these. On Running was born from this idea as a new way to cushion running shoes. They designed novel geometries using existing synthetic materials like EVA (ethylene vinyl acetate) in their soles.

They're continuing to experiment with combining existing materials, and even castor bean-derived bioplastic in pursuit of a better running experience. It's likely that younger companies like On will be far faster than incumbents to adopt the high performance materials of the new paradigm.

Direct to consumer

DTC have helped both Allbirds and On and many other startups in the last decade go to market and scale. DTC distribution has exploded because of the internet, enabling startups with strong product differentiation to do several things they couldn’t before. DTC made discoverability and organic growth far easier and faster. It also lowered the cost, time and complexity to having a storefront. Finally, DTC gave startups control over crafting their story, product perception and brand in a way that was never possible before. The key is that the product has to be better. Then, DTC and brand and marketing overall can be more effective.

Richard Branson's Virgin Records and later also Netflix (their DVD offering of course) initially used mail order distribution to seed their companies. More recently, computers in every pocket and on every desk has enabled DTC to be a huge competitive advantage against a variety of incumbents. Startups making products from shoes (Allbirds) to wearables (Fitbit) to cars (Tesla) have had DTC-first distribution strategies.

Allbirds used DTC online superbly to demonstrate product market fit and scale. The early kickstarter campaign had to be stopped after only 5 days. It showed over one thousand people were willing to buy the product. They launched in early 2016 and it was an instant hit with millennials and SF techies. Demand exploded, with their initial 15000 shoe inventory sold out fast and $1M of sales in the first month. I first heard about Allbirds after moving to SF in late 2017, where many of my then tech colleagues at Uber would rave about them. Their product differentiation unlocked strong word of mouth organic growth, avoiding sinking too much into online paid ads and keeping customer acquisition costs low. They also later set up their own brick and mortar stores. One product concern has been the durability of the shoe's upper and the outsole, which is indicative of the performance/sustainability tradeoff in many products today.

On Running didn’t find as much traction early on with online DTC. I think this comes down to two factors. It was founded earlier and runners in Europe in particular perhaps were less used to buying shoes online at the time. They scored their first two thousand orders at a conference in Munich during their first year operating in 2010. Despite having their online store live in 2011 shipping product to Europe and US/Canada, they leaned more on retail distributors early on. Hesitant retailers became convinced when they tried running with the shoes. Later, they invested in their online store and recently established a flagship store in Manhattan. DTC share of sales have grown to 37% in 2021 and is now On's fastest growing channel. Without DTC, their retail growth would have been limited post COVID and their 2021 IPO would not have been as compelling. On also has netted a small profit as of Q1 2022.

Strong growth despite being constrained by the current materials paradigm

The growth of Allbirds and On Running have both been exceptional. In 2021, Allbirds had grown in seven short years (counting from the cofounder's 2014 Kickstarter) to $277M in revenues and On in eleven years had climbed to $724M. Allbirds is not yet profitable, though On has been marginally profitable for a quarter. However, some of the largest public SaaS companies like Atlassian, Twilio and Hubspot amongst others are still not profitable either despite all being more than 12 years old. This is not dunking on them or SaaS startups in general, these are all companies with valuable category-leading products. What this comparison shows is just how well Allbirds and On Running have done whilst being constrained by the materials available today.

Because they could not innovate on the material level, the product and distribution had to be better for these companies to have had this success in recent years. How much faster would they have grown if they had been started during a time when novel materials were emerging? How much better would their products have been and how much more value would customers have gained from using their products?

The role of hardware, software and other technologies

As I said up front, I have been exploring these ideas and don't have all the answers. Perhaps this point most applies to how novel materials will integrate with hardware, software, AI and/or batteries in future compelling products. Unlike materials, these have been more important and more familiar technological drivers of innovation in recent decades.

I do think a new materials paradigm will be an enabler and complementary to these technologies. As discussed above, polymers with better properties enabled us to make products that were just not possible previously. The same happened with steel unlocking skyscrapers and stronger car body parts, aluminum enabling lighter engines and planes, and finally Gorilla Glass being the capacitive touch interface on over half of all smartphones today.

But I am less sure about the shape of new materials x other tech. Of course, novel material properties are vital to what this might look like. We'll talk about some of the properties of emerging new materials in Part Two of this blog series. I'll also talk more about automation and AI then.

For now, I want to touch briefly on Tesla and Fitbit as case studies. These two companies have been world-class at integrating hardware, software and batteries into standout consumer products. This might give some hints at the future of the interplay between these technologies and new materials.

Tesla

Perhaps no company has been more of a hardware, software and batteries integration pioneer than Tesla. For their first model, the OG Roadster, gluing together thousands of cylindrical battery cells from laptop batteries was good enough. This allowed them to pioneer the integration of electric propulsion into a vehicle form factor. Tesla pushed the boundaries, reengineering a car around the integration of battery pack, motor and power electronics.

You can see in this video @1:15, in which Tesla's VP of Vehicle Engineering, Peter Rawlinson, discusses the original Model S's motor, gearbox and power electronics all integrated in the rear suspension system. It was recorded in 2011 before the Model S launch the next year.

This world-leading powertrain engineering was coupled with DTC distribution since their beginning. Their DTC sales has been exclusively online-only since 2019. A DTC strategy was a risk in the late 2000s but has driven incredible success with preorders, especially for the Model 3 and Y. DTC has lowered Tesla’s customer acquisition costs by $2000 per car vs legacy automakers like Ford.

Materials science and engineering is quite important for Tesla. In late 2015, Elon poached ex Apple product design director Charles Kuehmann to work on both Tesla and SpaceX new alloy materials and aluminum in particular. Kuehmann's team helped develop new aluminum alloys for casting larger body parts in Teslas and stainless steel alloys for the Cybertruck and SpaceX's Starship. Last year, Tesla started a new materials application team tasked with accelerating the adoption of new materials and processes. I think startups will increasingly see more integration between engineering teams working on materials and other disciplines and product parts.

Fitbit

Fitbit was was both a tremendous success in its own right, and also a cautionary tale. It was the first hardware company to successfully integrate sensors into something you could wear, most notably an accelerometer for step counting.

In 2008 the founders James Park and Eric Friedman wanted a portable device that could help gamify exercise the way his Nintendo Wii had. They started Fitibit, despite not having any background in hardware nor experience with supply chains. They raised a $400K seed round in 2008, which was only enough to make a prototype of their circuit board in a wooden box. With that, they pitched their Fit tracker at TechCrunch Disrupt that year, aiming for 20 preorders for their then-unbuilt device. Like with On Running in Munich, conference attendees loved the product and they scored 2000 preorders in one day. Before Christmas 2009, they had racked up 25,000 preorders of $99 each. They knew they were on to something. Being clips, men wore them clipped to their pant pockets and many women preferred to strap them to their bras. After the Fit, they shipped three more clip style wearables before their first wristband released in 2013.

Their sales tripled on average every year from 2009 through 2014 as their clip form-factor devices were clearly a hit with health-conscious tech-savvy customers. Fitbit went public in 2015 seemingly brimming with potential. Then the Apple Watch launched in April 2015 and just blitzed the market, blowing a $800 million hole in Fitbit’s band sales. Fitbit’s sales fell nearly every year from 2015 until their acquisition by Google for $2.1B. This made the founders and many early employees very rich, but it was half of their valuation at IPO.

Fitbit was the first hardware company to successfully integrate sensors into something you could wear. But unlike Tesla’s EVs, their products were adjacent to incumbents' core products, in this case Apple. I don't have inside information on Fitbit's product roadmap nor execution at the time, and hindsight is 20/20. But it did seem like they could have been faster to launch both their first wristband (two years later than Jawbone) and their first smartwatch (lacking the features and design of Apple's Smartwatch). In some ways, Fitbit ended up like Nokia. Both built category-defining products but missed the boat on new product insights that led to being crushing by Apple.

The leading performance product startups of the 2020s will likely integrate software and hardware as well as Tesla or Fitbit did. They will probably leverage automation and AI too. I think they will utilize some of the emerging materials, biomanufactured materials in particular, and include highly cross-disciplinary teams of engineers to build amazing things.

Towards a new materials paradigm

Materials matter. I increasingly believe new and better materials redefine what is possible for startups to build, industries to use, and societies to imagine.

As we’ll see in the second part of this series, we are starting to see a transition to biomanufactured and other promising materials that are better performing and default-sustainable. As a quick preview, we'll cover materials like spider silk protein fibers (super strong, tough, flexible and lightweight) and algal oil based urethane (tensile strength and vibration damping), as well as self healing biomaterials and materials like graphene (12 thousand tons produced last year, growing at 62% CAGR over three years).

For the first time in several generations, we may be on the cusp of another material paradigm that changes everything. This would be a multiplier for the accelerating energy transition from fossils to clean electrons. It would also unlock another dimension in tomorrow's products that startups today have not had access to.

This decade will be an incredible time to build.

Reach out on Twitter or email.

Update: Part Two is here.

Thanks to James Giammona, Sid Athreya, Tony Maiorana (The Polymerist), Caleb Meredith and Anna Delas for their feedback on earlier versions.

Powered by solar, wind and lithium-ion batteries, the clean energy transition is reshaping the world.

Just over a year ago, I would not have believed that statement.

Back then, I did not appreciate just how fast these changes were multiplying. I have since learned a ton about energy transitions and climate tech, crunched the data and triangulated insights.

Today, I have no doubt about the clean energy transition and the impact it will have on our lives and societies. The phrase energy transition is also an understatement and few people alive have lived through one before to know any different. Looking at history overcomes our collective amnesia to see how important energy transitions have been, and hints at the path we are on now.

This post explains what is driving the transition and explores emerging opportunities. Fairly new to the topic? This could be an overview that helps orient as you learn more. Already playing your role in the transition? You might find a perspective that challenges your thinking or surprises you. You could even be nudged to build or join a company to tap into the energy transition's huge opportunities ahead.

Let's go.

Introduction

The following headline insights are short summaries. In the Key Takeaways section below, we revisit each insight in more detail. In the main sections, we then discuss each point in the most depth.

Insight 1: Plummeting Solar and Battery Costs Are Now At Tipping Points

Solar and lithium-ion batteries costs keep falling, and are now at tipping points of sustained and rapid growth, as shown in these charts.

In generating electricity, solar and wind are rapidly rising to replace fossil fuels. In mobility, lithium-ion battery prices have allowed electric vehicle (EVs) costs to be comparable to gas-guzzling cars.

These tipping points are accelerants propelling us towards a clean energy future.

Insight 2: As Production Scales, Solar, Wind and Batteries Become Cheaper and Better

We can manufacture solar panels and wind turbines to generate energy, and batteries to store it. This allows more of these resources to be deployed whilst costs fall and improving performance, shown below.

In previous energy transitions, this self-reinforcing cycle had never been sustained over multiple decades. As a result, predictions about clean energy deployments and costs have been persistently underwhelming.

Insight 3: Cheap, Clean Electricity Enables New Industries

Already historically cheap, the price of electricity from solar and wind will keep falling. Solar will likely drive most of the cost reductions.

Electricity prices are becoming cheap enough for new electrochemical industries to thrive, as shown above. Emerging opportunities include electricity-driven direct air capture, and production of new materials.

We'll see later that entrepreneurs who see the speed of the energy transition are the ones building the future of it.

Insight 4: Emerging Energy Sources Scale Up Through Similar Phases

The coal, oil and electricity transitions have scaled up through similar phases, which you can see in the chart below. Today solar, wind and batteries are also scaling up through these phases.

Energy sources have mutually reinforcing feedback loops with other technologies, which help to drive further adoption. For example, the internal combustion engine and petrochemicals drove more demand for oil, allowing it to gain an increasing share of global energy use in the early to mid 20th century.

These feedback loops and phases can help guide how far along we are in the transition and what opportunities are emerging.

Subscribe now

How to read this guide

Make no mistake that this is a long read. If you are feeling motivated, just keep reading. There is an overall narrative connecting the sections, but each section is largely self-contained.

If you just want the tl;dr, that would be the key takeaways. Then, you could choose a section to dive into further using the table of contents.

Table of Contents

Motivations

Key Takeaways

1) Plummeting Solar and Battery Costs Are Now At Tipping Points

2) As Production Scales, Solar, Wind and Batteries Become Cheaper and Better

3) Cheap, Clean Electricity Enables New Industries

4) Emerging Energy Sources Scale Up Through Similar Phases

1 Patterns in Previous Energy Transitions

1.1 Coal: Steam, Locomotives and Iron

1.2 Oil, Internal Combustion Engines and Polymers

1.3 Electricity, Aluminum and Transmission Lines

1.4 Clean Energy Now On The Rise

2 Generation: Solar and Wind

2.1 Solar Adoption: Slow, Then All At Once

2.2 Falling Costs Have Opened Up Larger Markets

2.3 Solar’s Growth Keeps Being Underestimated

2.4 Intermittency

2.5 Transmission upgrades

2.6 Strong Growth Ahead Is Likely

2.7 What About Nuclear Fission and Fusion?

3 Storage: Batteries, EVs and More

3.1 LIB Adoption Is Even Faster Than Solar

3.2 EVs Are Rapidly Replacing ICE Cars

3.3 LIB and EV Producers Outpacing Tech Giants’ Growth

3.4 Storage For Grids Is Booming

3.5 Home Energy Storage: Three Approaches

3.6 LIBs Continue Getting Cheaper and Better

3.7 Some Emerging Use-cases

4 As Energy Becomes Even Cheaper…

4.1 Electricity Prices Had Been Stagnant, Until Solar and Wind

4.2 Lowest Prices In Sunbathed and Windy Regions

4.3 Buying Ultra-cheap Electricity, Direct

4.4 Falling Electricity Costs Will Transform Industries

5 Opportunity: Removing & Using CO2

5.1 Negative Emissions Derisk Climate Scenarios

5.2 Direct Air Capture Is A Promising Approach

5.3 Clean CO2 to “X” May Outcompete Fossil Fuels

5.4 Existing Markets: CO2 As End Product

5.5 Emerging Markets: CO2 As Input Material or Additive

5.6 Circular Is Nearly As Good As Permanent

6 Opportunity: Emerging Materials

6.1 Low-cost Electricity Allowed Aluminum To Scale

6.2 Graphene: Potential In Many Markets

6.3 Batteries As A Structure

7 Transition Enablers

7.1 Software and AI Are Key Enablers

7.2 The Boom in Climate Tech Funding

8 A Vision for 2045

8.1 Insanely cheap solar, wind and storage power almost everything

8.2 EVs of all types and are taking over transportation

8.3 Battery storage is everywhere, but in EVs above all

8.4 Most existing industries are electrified, all are low-carbon

8.5 CO2 from the air is a key commodity

8.6 Ubiquitous graphene, fusion and more

Conclusion

Motivations

Energy (and climate) is an incredibly exciting and impactful problem space to be working on.

This space tackles two things in my mind. One is the risks of climate change, where we all should try to avoid the worst outcomes. Two is the continuing progress in energy technology, without which human progress could stagnate.

For over a year, I have been learning about the clean energy transition. I spoke to dozens of clean tech founders and engineers and absorbed books, research reports, papers, podcasts and countless articles.

Two things became clear.

First, solar and lithium-ion battery technologies were the core of the energy transition. Their rapid pace of adoption and dramatic cost reductions was unprecedented for energy technologies.

Second, very few insiders I talked to or learned from seemed to think solar and batteries would be as impactful as I was starting to think.

Was I wrong? I get that there are a plethora of technology, implementation, policy and other problems. Would these put the brakes on the path to a clean, cheap energy future?

As I continued my research, I gathered more data points and conviction that clean energies would be incredibly impactful.

When I was researching, I did not come across an article like this, but it would have really helped me orient myself. I hope this piece can help as a guide for at least a few people who are curious about the transition and where it could be going.

In writing this, I aim to find fellow tinkerers and builders who see the opportunities and may want to discuss and test ideas together.

Why The Clean Energy Transition Is Underestimated

It’s hard for us to fathom what we’ve never lived through. Few people alive today witnessed previous energy transitions happen. This makes it easy to underestimate how impactful they were, and how quickly they tend to happen once in motion. After all, the rise of coal, oil and electricity each happened over one hundred years ago in developed countries.

As almost no one alive had lived through the Spanish Flu one hundred years ago, COVID showed us how short our collective memory usually is for rare events of planetary scale.

It’s easy to think that the rapid scaling of solar, wind and battery technology cannot be sustained. Each transition goes through growing pains. But when there’s insatiable demand, industries find ways to meet it. That was the case for coal, oil and electricity as we’ll see, and it is the case today for cheap, clean energies.

There's the related concern that the decades-long cost reductions for solar and batteries will flatten out. As we'll cover, this is very unlikely. Existing technology and processes will keep improving and new technologies will be commercialized.

It’s hard to think about technologies and change as being non-linear. Let's show this with an example. Suppose you start with one grain of sand. Say that this one grain duplicates every year. How many years would it take for this one grain to replicate into the total number of sand grains on Earth? Assume there's 7.5 billion billion (7.5 x 10^18) grains of sand in the world.

What's your guess? 10,000 years? 1,000? It takes less than 63 years or doublings. If the lone sand grain duplicates every two years (ie grows at 41.4% per year), it only takes twice as long to hit that mark. These examples seem extremely counterintuitive, but it’s hard for people to think in this way. Those with startup experience tend to better grok non-linear thinking, as initial traction and growth are everything.

Solar, wind and batteries have grown and will continue to grow for some time like this. The problem is that over a few-year short time horizon, accelerating growth looks deceptively linear, as shown here.

Solar has grown at an average of almost forty percent every year since 1976, when good data began. From the sand grains example, you might recognize that this is a near doubling every two years. How much do you think installed solar capacity has scaled up since then? By over two million times. Of course, these growth rates will slow as we move up the S-curve. But already, we have installed enough solar to be only about a factor of 15 away from it being the primary generating source of electricity globally.

In our time, there’s a perception that we cannot quickly scale products made of atoms as fast as those of bits. Conventional wisdom holds that software scales better than hard tech. As we’ll see, this perception is not true, as some physical products in clean tech industries are scaling very, very fast.

This view probably stops many bright and driven people in developed countries working on deep and hard technologies. However, there are promising signs though that more people are looking to work in clean energy and climate.1

Key Takeaways

Let’s take a closer look at the four headline insights from the introduction.

1) Plummeting Solar and Battery Costs Are Now At Tipping Points

Solar has already become the cheapest energy source in history, alongside wind. Since 1976, solar costs (in dollars per watt) have fallen an average of over 12% per year.

Since first being commercialized in 1991, lithium-ion batteries costs (in dollars per kWh) have fallen a similar 12.9% per year on average.

These falling solar, wind and battery prices have recently hit tipping points that will drive continued growth.

The cost of solar or wind-sourced electrons are now cheaper than running existing fossil plants in many parts of the world. This is a tipping point that substantially grows the market size for solar and wind power.

As shown above, this has led to solar and wind surging to over 75% of new global capacity to generate electricity in 2020, up from low single digit percentages in the early 2000s. As costs keep falling, solar and wind will continue to outcompete fossil fuels in more regions.

Lithium-ion batteries are now crossing cost thresholds that are powering booms in both EVs and stationary storage.

The Tesla Model 3 is as cheap as competing gasoline models from Audi, Mercedes and BMW, but outsold them all in 2020. In the next few years, a growing number of pure electric models in more countries will become as cheap as gasoline competitors. Low cost, high performance batteries are making electrified vehicles increasingly cost competitive from two wheelers to electric boats and trucks.

Storing electricity in large batteries is driving down the cost of more consistent clean energy. Battery deployments on grids in the US, China and other areas are soaring, and increasingly being paired with utility-scale solar plants. When paired, solar and batteries are mutually beneficial, being more valuable than batteries alone.

2) As Production Scales, Solar, Wind and Batteries Become Cheaper and Better

We are seeing sustained, incredible growth from these clean energy technologies.

Solar is the fastest growing energy source in history.

Electricity generated from solar has grown at nearly 40% compounded annual rate (CAGR)2 for the last twenty years. Wind is no slouch either, having generated almost 22% more electricity every year, on average, during the same timeframe.

Meanwhile, electric vehicles (EVs) and large-scale storage are driving booming demand for lithium-ion batteries. EV sales have grown at around 57% CAGR in the last nine years, as shown in this chart. Storage for the grid is keeping pace, with global battery deployments growing at 58% CAGR in the last eight years.

This emerging boom in lithium-ion battery demand has created some of the world’s fastest revenue-growing companies like Tesla, CATL and others.

Solar, wind and batteries are manufactured energy technologies, not mined and extracted ones. As a result, the rapid growth in their deployment enables cost declines and better performance more akin to computer chips or electronics than fossil fuels. Coal and gas simply cannot continue to stay competitive. Whilst coal and natural gas plants have become cheaper to build as more have been deployed (capex), the cost of the fuels themselves have actually gone up over decades of deployment, driving operating costs higher.

One note is that the production of solar, wind and batteries do require the mining and processing of minerals. But once produced, unlike fossil fuels, no further extraction is needed.  

This self-reinforcing feedback loop is shown in the diagram below.

Whilst costs have plummeted, solar and battery characteristics continue improving. Solar photovoltaics (PV) technology has developed with many improvements across the supply chain. Battery characteristics like energy density, cycle count and safety have also rapidly improved.

3) Cheap, Clean Electricity Enables New Industries

Solar and wind now generate the cheapest electrons in history, and are only continuing to get cheaper. This will keep driving electricity costs lower, especially for commercial and industrial uses.

3As shown in the chart above, US heavy power users have strong incentives to purchase electricity from solar and wind power plants directly.

The pace of the cost reductions for solar-generated electrons is also historic. Falling electricity costs will create industries that are emerging now, as well as those yet to emerge.

For example, startups are betting on cheap electrons to power either CO2 capture and/or conversion. Direct air capture removes CO2 from the air. The CO2 can then be electrochemically converted into carbon-based materials, fuels and other products. Other industries are also set to take advantage of the lower cost of electrification.

Graphene and single-layer materials had not been isolated and invented just twenty years ago. Today, these materials are growing in use in an increasing number of niche markets. Clean and cheap electrons could play a key role in accelerating adoption of emerging materials like graphene.

4) Emerging Energy Sources Scale Up Through Similar Phases

Before we dive into the clean energy transition, it’s very useful to gain a mental model of how analogous moments in history played out. Whilst there are many differences, there are also parallels between past energy transitions and this one.

We will explore the early rise of coal, oil and electricity and where they stand today. It’s easy to think that progress happens much faster today than it did in the past. This is true in some ways, but we’ll see that some energy-fueled technologies scale up surprisingly quickly, even by today’s standards.

Unless linked, stats and references in this section are from Richard Rhode’s Energy, A Human History.

1 Patterns in Previous Energy Transitions

When I first looked into the rise of coal, oil and electricity,4 I was surprised to find some patterns.

These key energy transitions in history all scale up through similar phases, as shown in the figure below.

After being discovered, the energy source (or carrier in the case of electricity) finds a foothold in niche markets where it outperforms the incumbents. Over time, as the technology matures, it can enter larger markets. Then, in the most transformative phase, new infrastructure emerges for the energy source, new materials are enabled, and new industries emerge.

Early on in an energy transition, there is often more attention on how well they can do existing things. Examples include how well coal could heat and oil and electricity could light. But the larger impact for any energy transition, like other new technologies, is to enable us to do brand new things.

Take a steam engine tinkerer in Britain in 1800, when steam engines were mostly limited to heating, improving coal mining and industrializing iron. Could they have conceived that within fifty years, these coal-powered machines would be driving people and supplies across tracks laid all over the country? Or that fifty years after that, skyscrapers made from steel (using coke — heated coal — furnaces) would start to pop up in major cities?

Or take a car hobbyist in New York, in 1900, when there were only a handful of gasoline-powered cars on horse-dominated roads. Would they have believed that only 13 years later, Manhattan’s streets would be filled with cars and few horses would be found? Or that thirty years after that, oil would start to be refined into polymers and plastics?

Finally take an electricity innovator in the US in 1890, when the first city-wide electricity grids had been in place for less than ten years, mostly used to power lights. Would they have thought that aluminum, which was more expensive than gold not long before, would soon be commercially produced using electricity? Or that by the end of the roaring 20s, more than two in three US homes would have access to electricity?

I found something else surprising about previous energy transitions. Each one later enabled new modes of transport and new materials which were very hard to conceive of during its early phases. In each of the past transitions, the energy source unlocked new distribution technologies and materials that helped drive more adoption for the energy source itself. As we’ll see later, the same is true for solar, wind and batteries.

Let’s take a closer look at each of these past transitions as a guide to what might be in store through the door of the clean energy transition.

1.1 Coal: Steam, Locomotives and Iron

Initial niche

In the 17th century, wood was scarce in England. Prices were soaring as London’s population 8x’d from 1500 to 1650, driving more demand for firewood. As more forests were cleared, wood had to be carted in on muddy roads further away from population centers. This created the opportunity for an alternative to serve the booming country.

Despite its somewhat toxic smoke, coal was cheaper and easier to get to market than wood. Back in the 17th century, coal mines were located near rivers. This way, the fuel could be sent on boats to London and other places for smelting iron or ship making. As London boomed, coal shipments to the city rose more than an order of magnitude during the century.

But early coal mines were dangerous, often flooding and killing miners. The first coal-powered steam engines, invented by Thomas Newcomen in 1712, mechanized the pumping of water out of mines. They were stationary, and as large as a small house, but good enough to prevent flooding. Their use saved the lives of many miners and enabled the early industrial scaling of coal mining.

Cast iron commercialized: larger markets

By the 1740s coal was already useful for more than space heating. Coalbrookdale Company was the first to use coked coal to smelt iron, enabling viable mass production of cast iron for the first time in history. The company also used Newcomen’s steam engines to replace horse driven pumps, reducing costs in the process. Less than 60 years later, iron had replaced wood in most manufacturing and construction uses in Britain.

Before the 19th century, coal mines were limited to being near waterways. Roads and horses were unsuitable, and so coal would be loaded onto boats. This meant coal still had to be carted from mine to the water, using wooden tracks at first.

Locomotives: distribution

Railways were developed to solve this problem. Higher-pressure steam engines were developed in the late 18th century, and by the early 19th century improved designs were being used to power steam locomotives. Railway tracks changed from being made using wood, to cast iron and then wrought iron in this period, allowing railways to start transporting coal, other cargo, and people. Coal mines could be increasingly built further and further away from rivers.

In other words, industrial iron production and steam engines, both reliant on coal, helped unshackle the fuel’s initial limitations in supply. These technologies became mutually reinforcing, and their impact would have been hard to fathom during the 17th century. They enabled brand new things to be done.

Once coal supply was unconstrained, production grew rapidly in the 19th century and spread from England to Europe and the US. It enabled iron production to double every 10 years from 1788 for decades. If instead of coal, charcoal had been used to make the iron railway tracks, there simply would not have been enough wood in England.

Steelmaking commercialized

A breakthrough in steelmaking in the late 1850s, the Bessemer process, would help the US to catapult steel output over 400x from 1867 to 1900. By then, the US was producing one third of the world’s output. Just as back then, to this day, steelmaking largely still uses coke made from coal.5

The transition from wood to coal for industrializing countries had been the fastest and most dramatic energy shift to that point in human history. Until then, we had been using wood-based biomass for heat and raw materials for millions of years.

By 1900, coal was dominant, as shown in this chart. It supplied over 90% of the UK’s and 70% of US energy.

But only ten years later, coal’s energy use would peak, largely because oil was rapidly on the rise.

1.2 Oil, Internal Combustion Engines and Polymers

Initial niche

The initial product-market fit for oil was lighting. The search for oil was driven by finding a cheaper feedstock for kerosene, an oil for lighting that had been previously refined from coal. The first well in the US, drilled using a steam engine, started producing in 1859. Oil could be refined into kerosene and only 6 years the oil market for lighting had already 10x'd production.

The oil rush was on, initially in the US in Titusville, Pennsylvania. But, like coal, there were transport and storage problems to be overcome. It took 3 years for a rail connection for the town, and another year for the first rail-to-well oil pipeline.

John D. Rockefeller's Standard Oil was the emerging dominant oil refiner. He consolidated supply and made deals to reduce costs for kerosene by 55% from 1865 to 1870. Standard Oil drove efficiencies in refining, including using gasoline to help power refineries. Before gasoline found a market in internal combustion engines, other refiners were dumping it, often into rivers. Sounds incredible today, but then, they just wanted to make kerosene.

By 1870, cheaper oil-refined kerosene had cornered the US lighting market. But despite supplying almost 5 million barrels produced in Pennsylvania that year, lighting was still just an early wedge into bigger markets. There would be no bigger market for oil, and perhaps any other energy source until our time, than the internal combustion engine.

Engines and automobiles: distribution

The success of the internal combustion engine (ICE) would become the primary driver of oil demand for the 20th century.

Around 1900, steam engine powered cars and even electric battery cars were more popular than ICE cars in the US. In Europe, Porsche’s P1 (standing for Porsche number 1), was built in 1898 with a lead acid battery. However, the early lead of steam and electric car adoption would be short lived. Despite being a mature technology, steam engines were complicated to operate and appealed mostly to early-adopter hobbyists. Electric vehicles lacked charging infrastructure outside of cities, during a time when far more people lived in rural areas.

In the 1900s, ICE vehicles didn’t have the tech maturity of steam engines, nor the simplicity of electric cars. But thanks to the oil boom, ICE cars had refueling infrastructure distributed across the US. In the decades prior, gasoline derived from oil had found many uses outside of cities, used as a cleaning agent and solvent. Farmers had adopted stationary gasoline engines for “everything from washing machines to grain mills”.6 This meant general stores in rural and urban areas were stocked with gasoline.

Gasoline had superior distribution, and automobiles were increasingly cost effective. By 1914, there were nearly 1.7 million registered US motor vehicles, a market that grew at a CAGR of 46% over 14 years. See the photo taken in 1913 on 5th Avenue above. Only 13 years earlier, in 1900, there were almost no cars on the road. This was a remarkably quick transition, which, as we’ll discuss later, bodes well for electric vehicles in the 2020s.

Almost all vehicles sold by then were ICE powered. The Ford Model T had helped rapidly scale adoption of ICE vehicles by lowering costs dramatically. Henry Ford designed the car to have fewer parts (less than 100) which were interchangeable between cars, greatly increasing factory worker efficiency and streamlining the manufacturing process.

Polymers: a new class of material

Oil-derived polymers would become an entirely new class of materials. Bakelite, invented in 1907, was the first synthetic polymer and was actually made from coal. It was used as an electrical insulator for the rapidly electrifying US and in everything from radios, irons, combs as Time magazine praised in a 1924 cover. The success of Bakelite kicked off a race between chemical companies to discover new synthetic polymers, even if they did not have uses at first.

The R&D investment into synthetic polymers by many chemical companies started to pay off in the 1930s. Nylon was invented and used in WW2 (and to this day) for parachutes, ropes and stockings. Plexiglass was used in aircraft windows and polyethylene7 became used for packaging.

Just as with coal, the widespread use and distribution of oil enabled new materials to be developed and used at scale, as shown here. Entire new industries were created.

As remarkable as it was, one key thing limited oil’s rise. Cost. Oil has never been viably cheap enough to produce electricity, unlike coal, or later natural gas and now solar and wind.

Parallel to the rise of oil, electricity was also gaining traction. Despite its already huge impact in the 20th century, electrification is really just getting started.

1.3 Electricity, Aluminum and Transmission Lines

Initial niches

After more than a century of research progress, electricity was still to be commercialized in the mid 19th century. Some early uses like the telegraph and telephone had already proved useful. However, at the time, there were no use cases that required more abundant sources of electrons.

Thomas Edison’s commercialization of the incandescent light bulb in 1879 was the early killer app. Light bulbs were brighter and lasted much longer than kerosene lamps. Ironically, oil had only started replacing coal in supplying kerosene twenty years before Edison’s invention.

Early electric light adoption was rapid. In 1882, only three years later, Edison had set up the US’s first commercial power plant in lower Manhattan, initially for 82 customers on a few blocks. Two years after that, Edison was supplying over 500 people with over 10,000 electric lights.

Early distribution wars

Electricity could only be transmitted very limited distances and used direct current (DC) infrastructure at the time. Edison backed DC but copper wires melted if transmitted over long distances. Plus, there were only so many polluting coal power plants that people could tolerate near cities.

It took a shift from DC to alternating current (AC) to distribute electricity much more widely than a few city blocks. Once developed and tested in the 1880s-90s, it could transmit electrons dozens of miles initially, with transmission range increasing with higher voltages.

Aluminum commercialized

The electricity generated was also cheap enough to enable materials like aluminum to be produced cost-effectively for the first time.

When opened in 1895, Niagara Falls hydropower was first used to power Alcoa’s third ever smelting plant. Before electrolysis of its oxide was possible with cheaper electrons, the metal was difficult and expensive to refine from ore. It had been more expensive than gold, and rare as a result.

Global aluminum production would 15x from 1900 to 1916 as electrolysis costs fell and the metal found more uses. The Wright Brothers used an aluminum alloy plane engine in 1903, and aluminum began being widely used in airplanes in WW2 and satellites during the space race.

Aluminum also replaced copper in transmission lines, allowing longer distance lines to be more economically feasible. It was cheaper, lighter and allowed for lighter (and also cheaper) transmission towers to be built, despite not being as conductive as copper.

Electrified appliances: growing markets

Homes and buildings became increasingly electrified, with about five in six households connected to the grid by the end of the 1920s. Electricity usage per household kept growing as more appliances became widespread, especially after WW2. Washers, fridges, TVs, air conditioners, dishwashers all made their way into homes in developed countries by the 70s.

The increasing electrification of energy usage meant that a larger share of fossil fuels (coal and gas in particular) were producing electricity, and not for other end uses. As shown above, in 1900, less than 2% of the world’s fossil fuel production was converted into electrons, but by 2020, almost 40% was.

Fossil fuels are becoming less competitive with solar, wind and batteries, the new kids on the energy block. Now, another transition is well underway.

1.4 Clean Energy Now On The Rise

Despite all their vices and limitations, coal and oil fueled (no pun intended) the rapid progress in technological progress, and improved standards of living for billions. But the age of using prehistoric dead plants to power our economies is coming to an end.

Let’s look at energy transitions through the lens of innovation cycles, thinking back to how hard it is to understand non-linear growth.

Early on in the figure above, the incumbent has a high relative share of the market. they seem to be maintaining the share well. However, the new technology is rising. It is growing at a non-linear rate to take market share every year from the incumbent. Before the incumbent can react, the new has already surpassed it in market share.8

This has happened over and over throughout history. Horses disappeared off roads as ICE cars took over; offline share of advertising dollars were replaced by online ads; Nokia did not respond to the rise of smartphones; Taxis have been mostly usurped by ridesharing.

Incumbent disruption also applies to the energy industry and energy transitions. Few people alive have lived long enough to be a part of one, unlike for many of the examples I just gave.

How the transitions are playing out

In the chart below, we see a high-level view of how energy transitions from coal to oil and electrification played out. I’ve only included these three as they are the most globally important and dominant energy sources (energy carrier in electricity’s case).9

10Coal’s share of global energy consumption had already peaked earlier in around 1910, when the growing adoption of ICE vehicles allowed oil to rapidly start to eat into coal's share.

By the 1970s, oil was the most important energy source, peaking at above 40% of global energy consumption during the oil crises of the 1970s. Natural gas use has only increased somewhat from 14% in 1985 to 24% in 2019 despite US production rising dramatically since 2005.

Electrification of energy has been playing an increasing role in the last 120 years. Around 40% of global energy usage is electrified, using the substitution method. Another lens on the rise of electricity is that over 30% of (mostly) coal and gas is converted into electricity globally, because that’s a very productive use of those fossil fuels today.

Clean energy is rising and accelerating

In the last 10 years, solar and wind have been eating into coal’s dominance in generating electricity. They generated 1% of global electricity in 2007 and were closing in on 10% in 2020.

Fossil fuels are increasingly uncompetitive against solar and wind in generating electricity. New solar and wind power plants are cheaper to build than running existing coal and gas plants in almost half of the world. That has a lot to do with why over 80% of new electricity capacity in 2020 was from renewables, and the vast majority of that was from solar and wind.

It’s only a matter of years, not decades, before the world is building almost exclusively clean low-carbon power plants. Soon, there will be more coal and gas retirements than new solar and wind plants being added.

This trend is clearly visible in the chart above. The share of new electricity capacity from solar and wind topped 75% in 2020, having surged from around less than 6% in the mid 2000s. In that time, fossil fuels have shrunk from contributing over 80% of new capacity to less than 20%. This is a remarkable inversion in twenty years.

Despite rapid progress, solar and wind face growing pains. I’ll come back to the big challenges like intermittency, i.e. power only when the sun shines or wind blows, and the lack and slow speed of more transmission buildout.

Clearly, the clean energy transition is eating into the share of fossil fuels and at an accelerating rate.

A quick note about nuclear fission plant deployments, which had boomed during the 1960s and 70s. Meltdowns including at Three Mile Island, Chernobyl and Fukushima drove anti-nuclear sentiment, despite fission power being far safer than burning fossil fuels. That led to permitting and regulatory delays in fission plants being approved, with many projects being canceled and facing massive cost overruns.

Let’s next dive into the core of the clean energy transition to see how clean electrons (especially solar) are growing so quickly, why most pundits and industry insiders underestimated the growth and their foundational role in the clean energy transition.

2 Generation: Solar and Wind

We have never seen energy technologies like solar and wind before.

For the first time in history, we can manufacture the generation of energy. Once built, solar and wind equipment cost very little to operate. In other words, they are basically zero marginal cost electron sources.

Their fuel, either sunlight or wind, is free, and renewable. Solar in particular is modular and fractionalizable, meaning one panel can be installed on a rooftop or literally millions of panels at one site.

They have also gained traction fast. Historically fast so far, as the following chart shows. Of course, as deployments keep rising, the pace will likely slow down a little in the coming years.

11Solar and wind exhibit steep and consistent learning curves (also known as experience curves). That is, they become cheaper to build as more units are built and used. This trait of rapidly scaling technologies has never applied to fossil fuel energy sources.

These clean energies show no signs of slowing down. The 2010’s were one of solar’s best decades, as shown in the chart below. Growth is strong, cost reductions drastic, and the industry is innovating and learning to further lower costs at speed.

Installed capacity grew at over 33% CAGR during the decade. This is a blistering pace considering hundreds of millions of modules are being produced every year at this scale. Cost reductions were strong too, with the yearly fall in cost per watt was over 16% during the 2010s vs 12% over the last 44 years.

Finally, the learning rate, which is the rate of cost reduction for every doubling of capacity, was also much better than overall. Solar’s long run learning rate is about 23.6% since 1976, but this soared to almost 35% in the 2010s.

As we’ll discuss, even for optimist industry insiders this growth was unexpected.

There are positive feedback loops from combining technologies, as we saw in past transitions. We’re starting to see more solar plants being paired with lithium-ion storage, helping resolve many of the limitations of variable renewables. These two technologies are mutually beneficial and help each other scale. Analogs include the past transitions above: coal, iron and the steam engine; oil, steel and the internal combustion engine; electricity, aluminum and transmission lines.

Lets first look at how solar in particular has taken off to set the table for diving into the clean energy transition.

2.1 Solar Adoption: Slow, Then All At Once

Unless otherwise stated, stats and references in this section are from Greg Nemet’s How Solar Energy Became So Cheap.

Solar photovoltaics (PV) modules started off solving problems in small niche markets. As production ramped, costs came down and performance of modules kept improving. Solar kept entering larger markets, like the rooftop and later utility-scale power, which has continued driving strong growth.

Visualizing 40 years of non-linear growth

An important point to re-emphasize is how hard it is to grasp non-linear trends. Take solar since 1976. On average, module deployed capacity grew almost 40% per year while costs have fallen over 12% per year.12

What does this look like?

Solar’s growth is an excellent demonstration of the positive feedback loop we’ve discussed for manufactured technologies. Because fossil fuels are not manufactured technologies, costs do not consistently fall over a long period of time. Coal and natural gas plants do exhibit learning rates but the cost of the fuels themselves have trended up over the decades as easy-to-extract resources become depleted, driving operating costs higher.

Now if someone had just glanced at this chart, they might have assumed that the capacity of solar installed was basically zero until 2000. They might also think that solar module prices have been close to zero since the early 2010s.

They’d be wrong on both counts. The compounding growth and cost reductions drive massive changes in scale that are hard to read on linear axes.

There’s an initially harder to read, but much more useful way to express such large changes, as shown in this chart. Use log axes, make the x-axis cumulative capacity and mark time on the data series itself.

As we can see, solar’s growth and cost reductions have been very consistent over multiple decades, with some bumps (lasting a few years) along the way. The last ten years have seen the largest relative cost reduction ever. As we’ll discuss later, it does not look like solar’s growth is slowing down.

A short history of solar’s growth

In 1954, Bell Labs developed the first silicon solar cells. By the late 50s, early solar PV modules were used to power radio transmitters on US Navy satellites.

During the 1960s, the US government procured $150M for 10 million solar cells, as well as glass and solar PV components. They would also invest into solar R&D during the 70s as a response to the oil crises.13 This helped to drive down solar costs an impressive 5x from 1974 to 1981, despite still being a small nascent market with US industry sales only at $100M in 1982.

Through the 70s and 80s, Japanese companies like Sharp continued to find more niche markets for solar, such as lighthouses and consumer electronics — remember solar-powered calculators? They overtook US producers by the early 80s but had stagnated by the 90s when there were no larger markets to sell into.

Meanwhile California was gearing up to play a catalytic role in both solar and wind scale-up. There was newly introduced clean energy legislation and expectations of continuing oil price rises after the 1979 energy crisis. These allowed around a billion dollars of both wind and solar thermal projects to be financed and built before oil prices reverted.

In the 90s and 2000s, Germany would grow to become the biggest market for solar PVs at the time by far. Policy played a huge role, subsidizing over €200B of rooftop solar and accounting for over half of global PV installations during 2004 to 2010. German manufacturers like Q-Cells scaled up to serve the market, but by the mid 2000s, Chinese startups were rapidly taking market share.

In the early 2000s, China’s emerging solar startups executed faster and benefited from timing to scale up with the German market. Most importantly, they had a relentless focus on good-enough quality and low cost, which their German competitors did not. German manufacturers also had to develop more bespoke manufacturing equipment (often adapted from the semiconductor industry). But when later purchased by Chinese companies, the equipment had become more standardized and cost less.

From 2000-2007, in the first 7 years of some of the first startups launching, more solar PVs were installed globally by Chinese companies than by any other country. They raised $7B in IPOs during 2005 to 2007. Beginning in 2009, the Chinese government became very supportive of the industry, giving tens of billions in tax credits to solar manufacturers to accelerate scale-up.

Other regions with plentiful sunlight started to offer policy incentives. Australia and California offered feed-in tariffs introduced in the late 2000s, helping accelerate global adoption.

In 2008, a shortage and price spike in polysilicon, the main material for PVs, bankrupted many solar manufacturers. It also motivated overbuilding of capacity in the solar supply chain which then helped fuel the 2010s boom.

Since 2005, global installation capacity has been growing at a blistering 44% CAGR. Over 60% of the solar modules are made in China, a point that policy makers and manufacturers in the US and EU are now looking to change.

2.2 Falling Costs Have Opened Up Larger Markets

Electricity from solar and wind power are very cheap today.

Cost over the lifetime of an electricity-generating asset is calculated with the levelized cost of electricity or LCOE. Recall at the start of the post I mentioned tipping points for the cost of electricity from solar and wind.

We can show this new tipping point in the chart below. New project LCOE is cheaper than new-build natural gas, and is competitive with existing gas plants in the US. These clean energies have been cheaper than newly built gas plants for years — a previous tipping point.

14In the chart above, the solar and wind LCOEs are unsubsidized. They are even more competitive with existing gas plants when taking current US incentives into account. With incentives, solar is five cents per kilowatt hour cheaper in 2021, at $0.023/kWh. Wind is incredibly cheap, falling fifteen cents per kWh to $0.009/kWh.

The LCOE of new solar and/or wind plants are competitive with, or even cheaper than, running existing coal or gas plants in many other countries. For example Australia, most of Western Europe, China, India and others. The cheapest fossil fuel plant is location dependent, and can be either coal or gas.

Reaching this cost tipping point has started to heavily discourage new coal or gas power plants from being built. In 2020 globally, 82% of new electricity capacity was from renewables.

In the US, no coal plants have been added since 2014 as shown in the chart below. Gas plants added only 10% of new electricity capacity in the first three quarters of 2021, compared to 57% only three years ago.

Learning curves for solar and wind are global, so that more deployments in early-adopting countries help bring down costs for others. This is great news for developing countries too, where solar’s growth is also accelerating.

India increased its solar capacity 600x between 2010 and 2020, and is currently only behind China, the US and Japan in total installed capacity. Vietnam added more solar capacity in 2020 than every country except the US and China, as foreign banks shun coal there (as they’ve done in many countries).

2.3 Solar’s Growth Keeps Being Underestimated

Solar PV has sustained high learning rates. Since 1976, every doubling of solar capacity has led to 20% reduction in cost, driving 300x price fall as capacity scaled 2 million times.

This is an example of Wright’s Law that describes a consistent fall in cost for every doubling of production. It’s been observed for making everything from airplanes, the Ford Model T, semiconductor RAM, and lithium ion batteries. Wright’s Law may even be better suited for predicting semiconductor production than Moore’s Law.

Almost no one has been optimistic enough about the speed of solar’s growth and cost reductions. This is despite Wright’s Law holding well for solar over more than four decades.

An Oxford team recently analyzed historical forecasts for solar PV, wind and other energy transition technologies. Not surprisingly, models consistently underestimated the growth of solar as well as the speed of continuing cost reductions. In their working paper, they wrote:

“Such models have consistently failed to produce results in line with past trends.”

They concluded, not surprisingly given solar’s rather consistent learning rate that:

“In contrast, forecasts based on trend extrapolation consistently performed much better.”

What about the predictions of industry founders, insiders, and experts?

Gregory Nemet, author of How Solar Energy Became Cheap, wanted to see. Between 2008 and 2011, he asked 65 leading solar business pioneers, policy makers and academics to predict the cost of solar in 2030.

In 2018, reality had already exceeded the expert forecasts twelve years early. The cost of solar-powered electricity was already lower than the median predictions for 2030. Remember, these were not pundits, but people who had built the industry.

The IEA, the world’s leading energy body, and leading energy experts have consistently underestimated solar’s growth for the last 20 years. What has happened to solar’s cost in the last 10 years was over a decade ahead of expert predictions.

Why have models and experts been so wrong?

I think in short, solar has expanded into larger markets faster than anyone thought possible. Alongside that, it has continued to drive cost reductions in the process, even accelerating in the last decade.

As the Oxford team discussed, forecasts used arbitrary limits to growth rates not consistent with historical patterns, and just as arbitrary “floor costs” that were baked into models.

The learning rates for solar have been faster since 2010 than in previous decades. As discussed earlier, learning rates in the last decade were almost 35%, above the long-term average of 24% since 1976.

There are several reasons why the LCOE of solar has been falling even faster than the historical average. Solar PV technology keeps improving as production ramps.

One group who clearly did not get the forecasters’ slow-growth memos were the solar manufacturers. They increasingly benefit from economies of scale and standardization in manufacturing. Solar projects are increasingly being built at utility scale, which are now up to multiple gigawatts in size. This has enabled even more cost reductions with lower non-module costs for utility scale solar projects. Finally, solar PVs are becoming more durable with longer warranties being offered over time, with one large manufacturer now offering 40 years.

2.4 Intermittency

Intermittency is the problem of the sun not always shining, and the wind not always blowing.

As more solar has been added to grids like California’s CAISO and South Australia, there can be sunny days with too much solar for the grid to handle (the duck curve phenomena). In grids with high solar penetration like the two mentioned, it creates a need for resolving daily intermittency.

There is also the problem of longer-duration intermittency which can be for several days or seasonal mismatches. In an area where one winter is particularly cloudy, solar capacity can fall by half or more over a month vs the yearly average.

Solutions are at hand though. For shorter durations, big lithium ion storage projects provide up to four hours of viable storage for renewables with current unit economics. Eight hour storage projects are starting to appear. The first grid-scale deployment was by Tesla in South Australia in 2017, with the state having not endured a blackout nor needing to “shed” any clean energy since then.

Storage paired with solar can add more value to high-renewable grids than storage alone, especially during peak evening hours. Also, as battery prices continue to fall, longer duration storage, perhaps 12 hours will be possible. We will talk more about batteries and the storage market in the next section.

Longer duration storage has many clean energy solutions like pumped hydro, mechanical and various electrochemical solutions like iron-air batteries. There’s no shortage of funding for these emerging technologies. It’s still too early to tell which technologies will scale, but likely those that use low cost materials and have high learning rates.

Finally, the grid might not need as much storage as you’d think, though we will need a lot.

A recent paper published in Nature models that if solar and wind provide 1.5x of all electricity capacity, only 3 hours of paired battery storage would perform well. It would result in electricity shortages during less than 200 hours per year. Likely even less shortages would have been modeled if distributed batteries (say in electric vehicles) were included in the analysis.

2.5 Transmission upgrades

Transmitting utility-scale clean energy from solar or wind-rich areas requires more high voltage transmission lines to be built. These lines are often expensive, and get bogged down in permitting red-tape that can take 10 years for approval and deployment.

Of course, we only need transmission lines in cases where clean electrons are generated far away from where they are used, like urban centers and manufacturing hubs.

To move faster and get access to cheap, clean electrons directly, electricity-hungry industries are increasingly co-locating next to solar and wind plants. This way, renewables developers avoid paying interconnection fees whilst new industries powered on ultra-cheap electrons can get the cheapest juice possible without a utility acting as middleman. We’ll talk more about this later.

The other side of this is of course, installing smaller-scale solar projects in areas where energy is already used. Rooftop solar, microgrids and energy storage for homes and industry are growing. These distributed energy resources (DERs in energy-speak) can keep the lights on during extreme weather events even if all power lines into a city go down.

In many scenarios, more transmission does need to be built. To enable more utility-scale solar and wind power requires reform and infrastructure upgrades. This space has seen promising activity by US regulators recently and bills that allow for faster transmission buildout along roads and railways. There are also startups working on increasing the capacity of existing lines and optimizing them.

In reality, most countries are scaling both distributed and utility-scale clean energy as both types offer value in addition to being low carbon.

2.6 Strong Growth Ahead Is Likely

Solar and wind power are likely still very early in their adoption and will continue to see strong growth. The following points make my case.

Solar and wind keep driving down costs as  deployment scales

The cost of solar-powered electricity has declined 35% per industry doubling in the last decade. The learning rate is faster than the 24% rate seen since 1976, but makes sense given the rapid growth and competition of Chinese manufacturers since the mid-2000s discussed earlier.

No physical cost constraints for at least the next decade

There is plenty of land and materials like silicon for scaling up solar PV production. Newer panels are built to last longer. Solar PV technologies are still maturing, with room to drive down costs further. Examples include increasingly standardized manufacturing, better materials that improve cell efficiency and designs such as bifacial or double layer with higher efficiencies, and even anchoring panels directly to the ground to save on system and labor costs.

Looking at the last decade, the CAGR of solar-generated electricity was almost 40%, which outpaced the growth in annual capacity additions of just over 33%.15 This likely was helped by improving solar capacity factor from an increasing number of utility-scale projects using sun trackers as well as more efficient solar cells and otherwise improved manufacturing processes.

Solar and wind outcompete fossil fuels in many places

New utility-scale solar or wind projects are already cheaper than running existing coal and natural gas plants in countries making up 46% of the world’s population. As shown in the chart below, solar is cheapest for China, India and most of Western Europe; wind in Denmark, Brazil, UK, Morocco and others. In the US, 61% of existing coal plants are more expensive than renewables. Solar and wind are competitive with running existing natural gas plants.

As mentioned before, 82% of new electricity generation globally was from renewables in 2020. Coal is in near terminal decline, with proposed plants being canceled at a historic rate. Natural gas use for all uses including electricity, fuel and heat may have peaked in the US in 2019. The decline of gas for electricity generation likely will be accelerated by the late 2021 global spike in prices. As solar and wind become the cheapest source of electrons in even more countries, expect an increasing amount of stranded existing fossil assets.

The case to shut down and replace existing fossil plants becomes more and more compelling every year.

Distributed solar and offshore wind reduce need for transmission lines

As mentioned in 2.4, transmission upgrades are costly and take many years to even approve new utility-scale solar plants.

Distributed solar can complement utility-scale renewables. Types include residential rooftops, commercial and industrial, microgrids and offsite community solar installations.

Distributed solar, batteries and other resources can actually reduce grid electricity costs and even enable more utility-scale renewables to be built.

Solar PVs, by design, are small and modular enough to be deployed near or where it is used. Rooftop solar has boomed in countries where solar resources are great, system costs are cheap, net metering policies means high ROI, and permitting processes are fast. For example, 30% of Australian homes have already installed rooftop solar, and by 2025, solar will supply all the needed electrons at times.16

The US is a country in which rooftop solar is unnecessarily expensive. A combination of permitting costs and city-level red tape, arcane code restrictions and very high customer acquisition costs have made rooftop solar in the US far less affordable. Which of course means far less adoption. In contrast, Australia has streamlined permitting, simplified codes and lowered customer acquisition and labor installation costs. The result? Rooftop solar is almost three times more expensive in the US. The US does seem to be making inroads to reduce costs and friction though, with the SolarAPP public/private initiative to automate permitting, and California looking to pass its own bill to build on SolarAPP.

There are more rooftops being added just to enable more solar-generated electrons. EV charging stations are starting to add covered carports with solar PV rooftops, and stationary batteries. This makes EV charging sites more economical by allowing businesses to avoid demand charges, which can make up half of an energy-intensive business’s utility bill. It also allows restaurants and other commercial spaces to adopt microgrids, saving on electricity costs and having backup energy during outages.

Community solar is another promising type of distributed energy. It covers a broad umbrella of solar installations that allow the financial benefits to be distributed amongst many people.  These projects could allow two third of Americans and billions globally who can’t install rooftop solar to access local clean energy. A huge benefit is community solar does not usually require transmission upgrades. US community solar capacity grew at 130% CAGR during 2010-2020, far outpacing even the overall stellar rise in solar.

Offshore wind turbines can also help alleviate transmission challenges, being located near energy-hungry coastal population hubs. Capacity of global offshore wind farms rose 11x from 2010-2020 (27% CAGR). Even more growth is ahead, with nine times the amount of current capacity in the pipeline, as shown below. Wind speeds at sea are higher, more consistent and often peak when most needed during afternoons and evenings.

Wind power will continue getting cheaper. Electricity costs from offshore wind globally have fallen by nearly half in 10 years, with a 9% fall in 2020 alone and auction results suggest another 40% fall by 2023. In the US, the booming queues of offshore wind projects are very promising, but require some transmission issues to be resolved. Finally, offshore wind resources are especially abundant closer to Earth’s poles, making them very complementary for equator-favoring solar power.

Not to be outdone, onshore wind turbines at end-of-life can be repowered. This means much of the structure can be reused and new turbines be installed, saving on costs and eliminating lengthly interconnection delays and costs. This has already been happening in Denmark for example.

New uses for solar PV will continue to be found. We tend to think solar PVs will only be used to produce electricity. But we will likely see more novel uses and integrations as solar keeps getting cheaper.

One example is the growing momentum for building-integrated PV (BIPV) systems. These systems can replace existing facades, like for the Swiss building in the photo below, or replace rooftop tiles/shingles (think Tesla’s solar roof).

Another example is solar PV that can produce water. Source Global deploys “hydropanels” that make clean drinking water using solar PVs to condense water vapor. It is being used for remote communities and for water conservation in drought-stricken areas like California’s Central Valley.

2.7 What About Nuclear Fission and Fusion?

Nuclear power has tremendous potential, and the promise of abundant fusion power is worth accelerating R&D into.

There’s a lot of chatter amongst some Silicon Valley engineers and Twitterati about the promise of nuclear energy, which I in part agree with. Yet they have, like most everyone else, vastly underestimated the growth of solar and wind and some have acknowledged this more recently.

Around 10% of the world’s electricity is generated by large (gigawatt scale) nuclear fission plants. That sounds like a lot, but it has fallen from a high of over 17% in the mid 90s, mostly because of increasing public safety concerns over meltdowns like Fukushima, Chernobyl and Three Mile Island. Cost blowouts and cancellations have made large nuclear plants uncompetitive with solar or wind energy.

Despite these accidents, current generation nuclear power is very safe. In fact a hundred times more so than fossil fuels. Small modular reactors (SMRs) and next generation designs are likely to be even safer. However, these emerging technologies still face long project delays and cost concerns for buyers especially in developed countries. Other countries show some more promise, with NuScale, a leading SMR company, having plans to build an SMR plant in Romania by the late 2020s.

A promising strategy may be to carve out a compelling niche, avoiding competing on price with low cost renewables. Micro-reactors like Oklo are targeting remote communities that rely on diesel whilst NDB is developing tiny nuclear waste-powered batteries for consumer electronics that never need charging.

As for fusion, more R&D is a no-brainer. Rather than using a tiny fraction of sunlight emitted by the sun’s continuous fusion reaction, we should be building our own safe mini-star fusion reactors. A key step to progress fusion is a demonstration plant producing more energy than it uses. As dozens of well funded companies and public consortia resolve the early technical hurdles, progress has been promising.

Despite the R&D progress, the technology is still likely still at least two decades away from commercialization. By then, as we’ve shown, it is likely solar, wind and batteries will be so cheap as to make it very hard for early commercial fusion designs to be cost competitive in generating electrons. A possible path for fusion in the 2040s or 2050s would also be to scale up in niche markets, perhaps where heat can be directly harnessed.

It is inevitable that fusion, next-gen fission and/or other clean energy technologies will largely replace solar, wind and batteries one day. That’s the nature of energy transitions as we’ve seen.

3 Storage: Batteries, EVs and More

Lithium-ion batteries (LIBs) may be the most underrated technology today.

For years, LIBs have powered our phones, electronics, power tools and even pacemakers and hearing aids.

They run quietly and reliably in the background but it’s hard to imagine living in a world without them. Since they were commercialized only thirty years ago, these batteries have become something we take for granted,17 but already power so many tools we depend on.

LIBs are still in the early days of growth. Prices have tumbled over 97% since 1991, and continue to fall. Battery capacity has risen by over 6 orders of magnitude in that time, with the learning rate being about 18.9%. I.e. prices falling that much for every doubling of production. Sounds like solar’s growth right?

Many developments in technology and processes have continued to improve energy density and cycle count. Gravimetric energy density is up over 3x from ~80 Wh/kg in Sony’s original 1991 design to now almost 300 Wh/kg. This allows battery makers to offer 6 year and 600,000 km (370,000 mile) warranties on their batteries, and Tesla has said their Model S/X batteries still average 90% capacity after 200,000 miles.

Enabled by these improvements, growth is accelerating for these large, linchpin LIB markets. In 2021, EV sales more than doubled in just one year. US storage deployment is set to more than quadruple in 2021, after tripling in 2020!

In both battery markets, the rapidly falling cost of batteries have reached tipping points. They are increasingly allowing ICE car cost-competitiveness for EVs. Also, the cost reductions enable more economically viable deployment of grid storage either as standalone sites or (increasingly) paired with solar.

3.1 LIB Adoption Is Even Faster Than Solar

The rise of LIBs has been simply extraordinary so far.

John Goodenough, then at Oxford, developed a proof-of-concept in his 1980 research with cobalt oxide as a stable cathode. Five years later, Akira Yoshino developed a commercially viable battery based on the work of Goodenough and others. His employer and Sony would commercialize in 1991, initially in a bestselling camcorder, which its competitors like Panasonic quickly copied.

In the 90s and early 2000s, LIBs cornered the market for portable electronics, expanding into laptops, mobile phones, iPods and more. Costs plummeted, as you can see in the chart below. This market continued to grow as battery energy density improved. Not long after, lighter and more power-hungry devices like drones started hitting shelves and battlefields.

By the mid-2000s, cost reductions slowed as saturation in the consumer electronics market slowed growth. People were not buying new tech gadgets as quickly as they used to. Also, the small size of each device’s battery was contributing less and less to LIB deployment over time.

This changed with the rise of electric vehicles (EVs). The first LIB powered EV was Tesla’s Roadster. It launched in 2008 using existing reliable existing 18650 sized battery cells. These cells, assembled into custom battery packs, were actually designed for laptops — no batteries were built for EVs specifically at the time.18 Portable electronics literally provided the first batteries for commercializing EVs. Now, a Tesla Model 3 has the capacity of about 4000 iPhones.

The equally interesting, but less often discussed side of LIBs is how production scaled. Chinese battery cell manufacturers achieved 73% of global production in 2019. By then, the Chinese had rapidly surpassed Japanese (remember Sony?) and South Korean manufacturers who had previously led. They did this by focusing on the booming EVs and storage markets, and not electronics. Generous government incentives and policies dating back to 2009 also helped tremendously. Unlike the scale-up of solar manufacturing there, the public sector took more initiative in developing the battery supply chain before winners emerged.

No surprise: this growth has been underestimated

Much like solar, LIB prices have fallen twice as quickly as “optimistic” forecasters like BNEF and Navigant predicted. This has driven underwhelming estimates in the adoption of EVs and energy storage.

Let’s first look at EV forecasts. In 2015, the US Energy Information Administration (EIA) predicted that only 20,000 EVs with 200+ miles of range will be sold per year by 2040. In September 2021, the US sold almost 30,000 EVs in one month. In the last 5 years, BNEF and others just keep raising their projections for EV sales. Not surprising.

For energy storage, Wood Mackenzie, a leading energy consultancy, made a forecast in 2018 for US sales in 2023 to be $4.6B. Only three years later in H2 2021, they had to raise it by two thirds to $7.6B.19

It’s really hard for people to grasp how quickly high learning rate technologies like solar and LIBs can grow. Even Tesla has been surprised by the recent growth in demand for EVs.

3.2 EVs Are Rapidly Replacing ICE Cars

EV market growth is exceeding most expectations.

In 2021, people buying EVs are no longer just eco-warriors or Tesla fanboys. There are more and more reasons to buy an EV car like price, performance, cheaper cost of ownership, range, improving charging infrastructure and more EV models every year.

As you can see from the chart below, global EV sales keep growing, with 2021 being an inflection point.

Light electric vehicle sales globally have grown 57% every year from 2013 through 2021. There’s around 12 million purely electric passenger cars on the road in 2021. Around 4.3% of global car sales in the first half of 2021 were fully electric.

Growth is so rapid that every month, EVs are taking share from ICE cars. Leading the way in December 2021 were Western Europe with 20% and China with 17%. These markets led with far more vehicle selection and policy support vs the US other markets. The US is a few years behind at current growth rates, with EV’s making up less than 3% of 2021 passenger vehicle sales.

We’ve already talked about the dramatic fall in LIB prices. However, they do still account for about 30 percent of the cost of an EV, and have plenty of room to come down as learning rates continue. Upfront purchase cost is already competitive with ICE vehicles, with government incentives, particularly in China where EV market competition is fiercest.

Within a few years, EVs will reach price parity without incentives as battery prices keep falling. Even now, plenty of research has shown that lifetime cost including maintenance is lower for EVs.

EVs design affords less constraints and more interesting models. Removing a bulking engine means storage in the frunk. The best selling EV in China weighs only 1466 lbs (665 kg) and is less than 3 meters long. The first electric pickup truck, the Rivian R1T, has very high ground clearance for better offroading. The Rivian model even has an optional camp kitchen set built into their “gear tunnel” under the back seats. This is an early example of how EVs free from engine and exhaust requirements can be designed with novel features.

The fastest production cars in a straight line are already EVs. The 5 seater Tesla Model S Plaid is faster than the $3M Bugatti Chiron and every other million dollar hypercar.20 Its battery pack weighs about the same as the engine of the Chiron, and that’s not factoring in a full tank of gasoline.

The Model 3 is the best-selling vehicle in its class in the world, despite being cheaper21 than competing Audi, BMW and Mercedes models.

We are at or close to tipping points for price, in which EVs outperform an increasing range of comparable gasoline vehicles.

As battery energy density keeps improving, EV makers can improve range or reduce battery size and cost. Range is getting closer to parity with ICE vehicles, rising on average globally from 211 km in 2015 to 338 km in 2020. Most of the best selling models sold outside of China have much more range — the 2021 Tesla Model 3 Long Range has 334 miles (537 kms), and the Mustang Mach-E RWD has 305 miles (491 kms).

Charging infrastructure is getting better all the time, though it tends to be better in areas with higher EV adoption like California, Norway or Germany. For people who can, charging at home is like plugging in any device and a much better experience than waiting at a gas station.

For apartment dwellers, charging is much harder for now. Public charging should become much easier as more charging infrastructure gets built or opened up. For example, Tesla’s global network of 25,000 charging locations is being opened up for use by cars by other EV makers.

Perhaps most importantly, there is an ever expanding selection of EVs, especially in China and Europe. There were 235 EV models in 2020, mostly in China and Europe, up from only 55 in 2015. In 2021-2022 in the US, selection is also ramping fast.

3.3 LIB and EV Producers Outpacing Tech Giants’ Growth

Internet companies have set the standard for the rapid speed of growth. Since the 90s, we’ve seen the meteoric rise of Amazon, Google, Facebook and many more since.

That made the following even more of a surprise when I ran the numbers.

I realized that an LIB manufacturer and two EV companies have outgrown the internet giants at the same age. Growth rates are far beyond the tech companies, whilst revenues are comparable. Let’s look at these companies in turn — CATL, Tesla and XPeng.

CATL: Battery Powerhouse

So far, LIB manufacturer CATL is the largest winner from the boom in LIB production. It is ten years old, having been founded in 2011. In 2021, it was the top producer of LIBs globally with over 30% market share and growing, and has been for several years. Sales had grown to over $14B USD22 during the latest 12 months reported.

The pace of CATL’s growth is historic, as shown above. Its trailing 12 month revenue and net income skyrocketed more than 120% vs the previous 12 months. After ten years of operation, none of the tech giants were growing anywhere near as fast. Google was the only tech company that had more revenues after ten years but was growing at a relatively sluggish 31% year on year.

CATL is likely the fastest growing ten year old company ever, as measured by revenue growth. Oh, and it is already highly profitable.

Tesla: Pioneer and Leader

At eighteen years old, the pioneering EV company is also no slouch.

Tesla’s revenue growth also hold up very well vs tech giants, as shown below. In 2021, revenues were growing at 71% year-on-year, far faster than the tech titans at the same point in their histories, as you can see below. Facebook had much higher revenues in it's (latest) 17th year, but was not growing as fast.

Let's put Tesla's astounding growth in perspective. Neither Google nor Amazon have grown this fast for at least 15 years, going back to their earlier days at much smaller scale. Facebook has not grown this fast since 2011, again when they were much smaller.

Tesla is growing this fast, when it is already this big.

XPeng: EV Challenger

Finally, there’s the next generation of EV startups with their sights set on Tesla.

The furthest along are the booming Chinese startups. Amongst those, none have arguably have gained more traction than seven year old XPeng. They’re the first EV startup since Tesla to sell more than 15,000 vehicles in a month, and almost 100,000 in a year.

XPeng’s traction is remarkable for a seven year old manufacturer of complex hardware, software and machinery, as shown below. In the twelve months ending September 2021, revenues soared over 350% from the previous twelve months, an industry standout during China’s record. This is more than double the growth rate of Uber during 2016. Google and Facebook in their seventh years both grew revenue at around 90% CAGR, whereas Amazon had a tough seventh year with low growth.

What About Profitability?

You might think that this looks great for Tesla, CATL and XPeng when considering top line revenue, but what about profitability?

Well, even here, their numbers are quite favorable. A good metric to compare profitability is EBITDA margin, which measures operating profit as a share of revenue.

Tesla's was 21.6% in 2021, up from 5.5% in 2017 during Model 3 "production hell". Auto OEMs averaged far less, at 6.8%. CATL's EBITDA margin was 22% in 2020, and fell a little to 19.5% in 2021 when it made massive investments in new capacity to meet surging LIB demand. XPeng, investing heavily to keep up with extreme growth, had a respectable -7.9% for the most recent quarter.

Amazon had less than 10% EBITDA margin until 2018, and is still "only" at around 13%.  For Q4 2021, Uber's for rides and deliveries combined was 0.3%. With less capital expenses, pure software companies like Google and Facebook do have better EBIDTA margins vs CATL. Google's/Alphabet's has been between 25-40% whilst Facebook/Meta has usually been between 42-57%.

In short, Tesla, CATL and XPeng have profitability margins that looks more like tech companies with real-world operations like Amazon than like automakers.

Unpacking this narrative violation

Ten years ago, if I had said Tesla would be growing faster than tech companies in 2021, I would have been laughed at. People would have dismissed the possibility of a lithium-ion battery maker doing the same.

Yet, today we can no longer think of internet startups as the preeminent hyper-growth companies of the world.

The incredible growth of clean energy companies is most visible today in LIB and EV makers. Solar manufacturers for panels and inverters are also large and still fast growing companies in their own right.

It’s hard to see today’s use-cases that will become tomorrow’s leading industries, but many are very likely to be riding the clean energy transition.

With such incredible recent growth in LIBs and EVs, it’s not surprising US and European policymakers have big ambitions to rapidly scale up LIB manufacturing and mining.

Redwood Materials in the US and Northvolt in Europe have raised billions to build out battery gigafactories — founders from both companies are ex-Tesla executives. Diversifying global manufacturing could not only reduce reliance on a single country, it could also accelerate battery breakthroughs and progress down the learning curve.

The rapid growth of a manufacturing and mining-intensive industry has also raised questions about sustainability and reuse. Both battery module and EV manufacturers are increasingly recycling their batteries. Tesla works with third parties and recycles batteries themselves at their Nevada cell factory. BYD, one of the world's largest EV and LIB manufacturers, designs their cells to be easily removed for reuse. Redwood Materials works with EV makers to reuse depleted cells whilst Northvolt also plans to recycle 95% of the batteries they will produce.

3.4 Storage For Grids Is Booming

The energy storage market is booming.

Stationary battery storage has historically mostly been deployed standalone. These earn revenue by helping the grid in a growing number of ways. LIBs have become cheaper, battery storage has excelled in high-renewable grids like California, Texas and Australia. They can now economically offer 4 or more hours of electron storage (and growing).

Global storage deployments grew at about 58% annually from 2013-2019, and has accelerated since then. The US market has been surging even faster, more than doubling gigawatt-hour capacity every year since 2013 on average. This is shown in the chart below. In 2021, it will nearly quadruple deployments made in 2020, which in turn were triple what 2019's deployments were.

Storage is even more valuable when paired with solar

Part of that recent astronomical growth is the mutually beneficial nature of pairing utility-scale storage with solar. This allows batteries to freely charge during sunny enough days and discharge in the evenings when electricity demand is usually high. Colocating them adds a larger load-balancing benefit to the grid than having separate sites for standalone projects. Another benefit is that a colocated storage and solar install shaves project costs by around 6%.

With paired storage, grid operators and market participants can continue deploying more solar onto grids everywhere with confidence.

This appeal has driven over a third of interconnection queued US solar projects to include onsite storage, as shown below. Even as solar deployments keep rising, storage is catching up fast. This is shaping up to be a flywheel effect for solar and storage to mutually drive more growth in the decade ahead.

In mid 2021, with so much demand for stationary storage, Tesla had an 18 month backlog for Megapack, their utility-scale storage product. Massive battery manufacturing ramp up is coming online to serve the booming EV and storage markets, but even manufacturers did not anticipate this much demand. New battery gigafactories are announced seemingly every week in China , Europe and the US.

It's likely we will increasingly see more integrated approaches to solar plus storage. For example, Yotta Energy has developed lithium iron phosphate battery modules that slot under solar panels on commercial rooftops. The battery acts as a ballast for each panel, increases installation speed and lowers the cost.

We will not have to rely solely on utility-scale batteries to store renewable electrons. Long duration storage can help the grid store days worth of electrons. They promise to be more economically viable than LIBs and add redundancy. Batteries as distributed energy resources can backup large buildings or an RV. Recent research shows that allowing electrons to be stored anywhere can actually make grid investments cheaper and enable more utility-scale renewables. Second life EV batteries are already and will increasingly be repurposed as useful and profitable storage for grid electricity.

3.5 Home Energy Storage: Three Approaches

There is a Cambrian explosion happening for home storage backup solutions.

Home battery storage (called wallbox units) like Tesla’s Powerwall are often paired with rooftop solar for resilience during power outages, such as during the Texas winter storms.

In addition, alternatives like EV batteries and more portable batteries offer other ways to store and use electrons in a pinch.

A mid-sized EV’s battery pack has the energy capacity of about five Powerwalls. Given this, using EV batteries for home storage is very promising for home energy backup. There are problems to solve though. Some drivers and EV makers are concerned about battery warranties if their batteries are also used for powering homes.

This has not stopped EV makers like Rivian from offering power for optional onboard induction cooktops, and Ford’s F150 EV enabling 90 kWh of vehicle to home charging. The Ford’s battery would power everything in the average US home for at least 3 days, and has the capacity of almost 7 Powerwalls.

VW, Hyundai and Fisker are also planning to offer vehicle-to-home (or even to building/grid) charging while Nissan has used Leaf batteries to power lights in buildings during blackouts since 2011.

In a state with almost 15 million cars, having just one in five dispatching their batteries replaces the need for evening peak fossil fuel generation.

People living in most apartments likely can not set up a Powerwall or an EV battery to backup their homes.

Instead, many are starting to use portable batteries. These are modular products offering multiple AC, and DC outlets that are smaller and more portable than heavy wall mounted units. The newer models like the Ecoflow Delta Pro use lithium iron phosphate battery chemistry, allowing for thousands of charging cycles. These in particular can be stacked with more batteries to offer up to 25kWh of capacity (almost two Powerwalls) and even connect to an electrical panel to power anything in a home.

In the last year or so, interest in these power stations has exploded, with Ecoflow Kickstarting over $12M for their latest Delta Pro and Delta Max products, the most ever for a Kickstarter tech product. They also raised a $100M series B recently. A competitor, Bluetti has also raised over $10M across two home battery crowdfunding campaigns.

There’s likely a large role for each type of storage described above in the clean energy transition. More energy backup in homes and EVs improves not only resilience, but also complement and adds redundancy to utility-scale storage.

3.6 LIBs Continue Getting Cheaper and Better

Until LIBs, we had never been able to store clean electrons anywhere or at any scale.

It’s interesting how, unlike solar, wind or any other energy technology, the humble lithium-ion battery has never really been hyped. Batteries have just slowly found its way into billions of electronic devices.

Very likely, battery tech will keep moving down the learning curve and the speed of deployment will not slow down. As we’ve seen, the EV and storage markets alone should be plenty to continue the rapid scale up of battery use, whilst lowering prices at the same time.

There are many companies working on numerous ways to improve batteries — higher energy density, lower cost, increasing cycle count and more. These changes are happening at the cell level (like developing new cathodes/anodes with better characteristics) and pack level (maximize actual cell space used, integrating batteries into product structures, etc).

Battery energy density can keep improving for a long time. In batteries using lithium-sulfur chemistries for example, the gravimetric energy density achievable is up to triple that of commercialized LIB chemistries.

The battery supply chain continues to improve. New lithium extraction and refining methods like from Lilac and EnergyX save water, land and recover more of the mineral. EnergyX's approach expects to lower the refining time from 18 months to a few days. In addition, onshoring lithium mining reduces supply chain risks. Cobalt, an essential mineral in higher-performance batteries, has supply risks and ethical mining concerns. Companies are already reliant on cobalt by either lowering its concentration in cathodes or using cobalt-free alternatives.

Increasingly, EV makers like Tesla, Volkswagen and Fisker are using cobalt-free lithium iron phosphate (LFP) cathodes for their base models. LFP’s share of the lithium ion market is surging, up from 15% in 2020 to 24% in the first half of 2021. The top selling EVs globally like Model Y and 3, China’s best-selling Hongguang MINI EV all use this chemistry. LFP batteries are about 20% cheaper to produce and less fire-prone vs nickel-based ones. They do, however, trade off slightly lower energy density.23

Batteries that replace lithium altogether are also in development. For example, sodium-ion batteries for EVs and storage that promise even lower cost and operation in more extreme weather conditions. They trade off lower energy density for now.

There’s a lot of work ahead for the development of high performance batteries, but it’s very likely batteries will keep getting cheaper, more energy dense and last more cycles. Just like improvements unlocked batteries able to fly drones, power cars and store electrons at scale, future improvements will power new uses and markets.

3.7 Some Emerging Use-cases

There are LIBs already used in vehicles of all sizes, though it is early days for most markets. As batteries get cheaper and better, watch for accelerating EV adoption in the following vehicle types. Which of course, is great news for the clean energy transition.

Let’s start with electric two wheelers, which include motorbikes, scooters and mopeds. In 2020, already over one third of all global two wheeler sales are electrified, driven by China. Notice in the chart below that India is lagging with their share of two wheeler sales. Ola Electric, spun out of the largest ride-sharing company in India, is changing that. They recently sold ~$150M USD worth of their S1 scooter in only two days. This was the highest sales ever in India for a vehicle (or any product online) in such a short time.

In heavier vehicle categories, buses and vans are already starting the transition to EVs. For example, Shenzhen in China made their whole bus fleet electric by 2018, thanks to government subsidies to pull forward the tech. With trucks, industry perceptions have changed from just a few years ago as batteries keep getting better. Research has shown they can already drive short-haul routes, which make up half of those in the US.

Electric trucks will still take a few years to reach cost competitiveness, but large European manufacturers are all in, having pledged in 2020 to sell only electric and hydrogen trucks by 2040. It’s likely they’ll move faster than that. Even electric tractor startups (some with founders who, of course, were ex-Tesla) are growing fast and dealing with long order backlogs.

Electric vertical takeoff and landing (eVTOL) aircraft are a harder problem than terrestrial transport. Battery power density and electric motors are not yet good enough for flight, but this is changing.

The first eVTOL flying short commercial routes could happen as early as next year, connecting Boston and Cape Cod. This specific route is currently unprofitable and subsidized for jet aircraft, as are other short routes in areas around the world because of high fuel and operational costs. EVTOLs offer cheaper flights with faster door-to-door times.

These use cases can be niches that small eVTOL aircraft tap into before expanding into other markets. Airlines and logistics companies are interested, with American Airlines, Virgin, United Airlines, UPS and others all placing preorders for thousands of EVAs from various startups that are aiming to deliver aircraft in the mid 2020s.

4 As Energy Becomes Even Cheaper…

Cheap electricity matters a lot, especially for larger (literal) power users. Aluminum smelters, data centers, crypto miners and more are all heavily electricity-intensive industries which seek out low cost sources of power today.

Solar and wind are now the cheapest electricity sources in history and solar has the fastest falling costs of any energy source in history. In the last one hundred years, electricity prices for power-hungry industries have never fallen as fast as solar PV’s costs plummeting in the last ten years. Not even close.

We will look at how electrical prices have actually been flat since the mid 20th century. Falling electricity prices are not something energy entrepreneurs or policy makers have seen nor expected within our lifetimes.

This section then revisits how electricity prices from solar and wind will continue to get cheaper. It will cover how this will mostly benefit energy-intensive industries in sun and wind rich, policy friendly regions like Texas. We will see how companies are starting to buy electricity directly from producers. Finally, this section ends with examples of how cheap, clean energy is accelerating electrification in existing industries.

4.1 Electricity Prices Had Been Stagnant, Until Solar and Wind

In the US, the real price of electricity for industrial users has been fairly flat since the 1950s.

In other words, industrial electricity users have not seen lower costs since WW2 and the Great Depression when adjusted for inflation. Adjusting for price levels allows us to compare prices across time as the cost of doing business rises with prices.

As seen in the chart below, there was a spike during the 70s triggered by the oil crises but prices reverted starting in the early 80s. Since the late 90s, prices have been pretty stagnant in real terms.

This makes sense from the lens of energy transitions. As discussed, solar and wind are the first manufactured electricity sources in history with substantial learning rates. No other source of electricity has pushed costs down as quickly as solar when scaling. Power-hungry companies around today have never existed in a world where energy costs have fallen, or could have fallen rapidly.

While electricity prices are more or less flat, the cost of solar-generated electrons has plummeted.

As we will discuss, there’s an increasing number of companies taking advantage of and betting on this cheap electricity. When energy costs are the majority of their overall cost structure, this makes economic sense.

You might notice that the LCOE of wind has been much lower than the cost of industrial electricity for over ten years. Why have industries not taken advantage of these low prices earlier?

One reason is that until the last few years, very few companies actually bought power (physically or virtually) direct from solar or wind developers. These contracts are a recent innovation that clean energy developers are becoming increasingly comfortable with only recently.

Another is geographical mismatch. Solar and wind sites that offer cheap electrons tend not to be near population centers. The costs are getting cheap enough that some industries, crypto mining in particular, have been setting up shop next to solar and wind plants to reap the benefits.

Finally, the intermittency of solar and wind cause challenges for existing industries. Mature industries like aluminum have needed "firm" sources of electricity to maximize the capacity of capital-intensive (i.e. expensive) plants. Alcoa and other aluminum smelters had already sought low-cost electricity, for example, from hydropower in Iceland.

As we’ll see, there are industries starting to build for intermittency as a feature. Green hydrogen production, CO2 utilization and crypto mining are all looking to leverage low cost electricity when the sun shines or wind blows. Battery storage will also increasingly enable less intermittent cheap clean electricity as well.

4.2 Lowest Prices In Sunbathed and Windy Regions

According to Lazard as we showed earlier, LCOEs for solar and wind in the US in 2021 are $0.028 and $0.026 per kilowatt-hour respectively.

These LCOEs are already very low but are only aggregated values. Many solar and wind projects offer cheaper electricity than these numbers.

For example, in Saudi Arabia in 2021, a 600 MW solar project was priced at $0.0104/kWh. This price for sun-powered electrons was unthinkable even in 2017, and is the lowest cost electricity in history. But it’s not just Saudi Arabia. Many countries from Mexico to Chile and Ethiopia are now seeing solar LCOE come in below $0.02/kWh.

This is also happening in the US. El Paso electric utility, which serves West Texas and part of New Mexico, agreed to electricity supply from a 100MW solar plant costing just $0.015/kWh, and a solar and 100MW / 50MWh storage plant costing $0.030/kWh. The local coal fired power plant costs much more, $0.045/kWh, just to run.

Texas has historically been very fortunate with its vast wealth of oil and gas resources. It also has always had tremendous solar and wind resources in its west, treasured assets for the clean energy transition. The Lone Star State had the most wind capacity installed of any US state by far, with China and Germany being the only countries that produced more electrons from wind power in 2020. Texas should overtake California by 2023 to have the most solar capacity installed of any US state.

As renewables ramp up faster in parts of the US, Europe and Australia, some areas are increasingly seeing negative pricing for wholesale electricity. That is, solar and wind plants and other energy producers have to pay utilities or companies for electricity they produce. High-wind West/North Texas and the Central Great Plains, some European countries, high-solar South Australia and California to a lesser extent all see this happen. As shown on the map below, some localized areas in wind-rich Texas, Kansas and Oklahoma see negative electricity pricing over 20% of the time.

One point about intermittency is that battery storage does not add much to solar costs, and as we discussed earlier, storage prices will come down even faster. NREL costing has found four hour battery storage with current tax credits adds only around $0.028/kWh to utility-scale US solar prices in 2020. That’s roughly doubling the average cost, which still makes it cheaper than industrial electricity prices today. LIBs for utility scale storage are currently more expensive than for EVs, but expect the costs to come down more as the storage market keeps rapidly growing.

4.3 Buying Ultra-cheap Electricity, Direct

Some companies are already taking advantage of the cheapest energy around. To do this right now, they usually need to buy electrons directly from renewable project developers.

The US is leading the world in corporates buying clean power directly. This trend in corporate buying of so-called power purchasing agreements (PPAs) has been led by tech giants like Google, Amazon and other tech giants with aggressive net-zero emissions targets. In 2020, US companies signed 11.9 GW worth of corporate PPAs for clean energy, which represents about one third of all US solar and wind capacity for the year.

The catch is the majority of these purchases are “virtual PPAs” meaning that the companies do not use the electricity directly, but act more like a direct offset for their electricity used in a location that could be hundreds of miles away from the solar/wind plant. These contracts can help companies to reduce “scope 2” CO2 emissions from electricity use, hedge against rising utility energy costs, or receive renewable energy credits which can be sold.

The chart above shows that virtual PPA prices are quite low for companies, especially in ERCOT (which serves most of Texas) for solar and wind and SPP for wind.24 These virtual PPAs have to be connected to the respective grids, whereas colocated industrial electricity users can bypass these slow and costly grid connections.

Because of the success of corporate PPAs, renewable project developers are now familiar with selling electricity directly to companies. Increasingly, motivated energy-hungry businesses are locating operations right next to cheap solar and wind power. This is economically viable for many companies, especially in manufacturing and crypto mining in which electricity is a significant share (or even majority) of their operating costs. It also saves developers costs and delays in grid connections.

There is the problem of intermittency when co-locating (known as "behind the meter") with a solar or wind plant. If the electricity-hungry process is capable of operating almost all the time, it faces higher capital costs to idle at night or on a still day. Storage is an option, but viability depends on the (additional) cost of the electrons that the storage can provide. Right now, there is definitely a tradeoff.

Driven to reduce costs, crypto mining is an early example

For example, many crypto miners have set up in West Texas for cheap solar and wind power and in New York and elsewhere for cheap hydro and nuclear. They often co-locate with the cheapest clean electricity sources they can find. Electricity is the vast majority of their operational costs, and crypto mining is only profitable below a certain electricity price which depends on the mining equipment used.

Crypto mining can be turned on or off in seconds. This allows miners (and grids) the flexibility to turn off when electricity prices spike. During the Texas winter storms, some crypto miners turned off for 2 days because surging electricity prices incentivized shutdown.

There is a significant divide in the climate tech industry about whether crypto mining should be able to use renewable electrons. Mining can act as flexible electricity demand but often does not. Still, colocating mining (or any other power-hungry operation) next to solar or wind farms means bypassing the long interconnection queues and transmission line limits we discussed earlier.

Other crypto miners are capturing the heat used from mining, and selling it to utilities. As perhaps the first project of its kind, MintGreen almost completely pays for its electricity cost by capturing and selling district heat to 100 buildings in North Vancouver. This enables utilities to avoid GHG emissions from natural gas furnaces and could signal the early days of a new way for reusing distributed heat sources.

More so than crypto mining helping the grid, there will be more and more industries with high-electricity operating costs looking to use the cheapest, cleanest energy available. Colocating these high power users with the clean power plants helps create more demand for clean energy, without making interconnection queues worse or requiring more transmission line upgrades.

4.4 Falling Electricity Costs Will Transform Industries

Solar and wind are far from done in getting cheaper. They and LIBs will continue scaling quickly, and progressing along their learning curves. In the process, they will accelerate the rate of electrification that we covered earlier.

Today, solar and wind only generate about 10% of electricity globally. Only about 40% of all energy is electrified. There’s clearly far more upside as we electrify fuel-burning industries from EVs to space heating and much more.

Space heating and cooling for homes and buildings will increasingly be electrified, with annual shipments of US heat pumps doubling to 3M from 2010-2019. Startups like Gradient are trying to be the Nest of home heating, and are attracting funding.

Manufacturing steel and concrete account for around 20% of CO2 emissions, but there are promising pathways including electrochemical processes for steel that become cheaper as electricity does. Low carbon concrete is already pouring into construction projects while Sublime (electrochemical) and Brimstone (carbon negative) are producing cement in new ways.  

Instead of shutting down, a Colorado steel mill will continue to run by sourcing very cheap electrons from a new 300MW solar plant built nearby. They will get access to solar that’s cheaper than $0.03/kWh, powering 90% of their electricity needs.

Rondo is a startup generating industrial heat (previously derived from fossil fuels), using renewable electrons stored in insulated rocks. Antora is using a thermal photovoltaic process to do something similar. Green hydrogen (made from clean electrons) could be used here too. Even the petrochemicals industry has emerging carbon-negative processes like Solugen’s chemienzymatic approach.

Nitricity creates nitrogen fertilizer from air, solar electricity and water, mimicking how lightning storms break down nitrogen in the air for fertilizer. They use solar PV to generate “containerized lightning” which they combine with air alone to create nitrogen in solution. This is then fed into farms’ irrigation systems.

These are all existing industries that will continue to be electrified. Let’s now hone in on two new use cases that could be accelerated by the clean energy transition.

1. Electrochemical capture and use of CO2 from the air and

2. The production of graphene and 2D materials.

5 Opportunity: Removing & Using CO2

Since the industrial revolution began, humanity has emitted around 2.4 trillion tons (teratons) of additional CO2 into the air and oceans.

As shown in this chart, this mass of atmospheric CO2 is likely even more than all the post-1900 human-made stuff and current living biomass combined!

Whilst this much excess carbon in the air and oceans can sound like climate doom, it is actually a massive opportunity. An increasingly vibrant community of startups, investors, policy makers, researchers and more are working on solving this problem.

That’s an unfathomable mass, so let’s look at yearly cement production and fossil fuel emissions. Cement is the most used human raw material after water, and globally about 4.2 gigatons are produced every year. That pales compared to CO2 emissions, which are over 35 gigatons per year. Every year we produce close to an order of magnitude more CO2 than cement.

CO2 is already being used in beer and carbonated beverages, prolonging shelf life of foods as well as medical and industrial uses. These 2.4 teratons of human-emitted CO2 in the atmosphere could be used to enable a clean carbon economy or be permanently sequestered underground, both of which we will come back to.

5.1 Negative Emissions Derisk Climate Scenarios

There’s still a lot of work ahead to keep rising CO2 levels from having large societal costs.

Warming temperatures, more flooding, drought and rising sea levels can not be avoided because of already emitted CO2. But the worst impacts can only be avoided with carbon removal, as made clear by the IPCC in their 2021 and 2018 reports.

This gif chart by climate scientist Glen Peters helps visualize the need for gigaton-scale negative emissions technologies. These technologies will be needed just to maintain temperatures 1.5-2.2 degrees C higher than pre-industrial times.

This is because of several factors. There will be hard to abate sectors like aviation, shipping, industrial heat and agricultural methane. Emissions can be underreported and unverified, especially for developing countries. Nationalized entities control three quarters of fossil fuel assets but lack incentives to report when their resources are burned. Negative emissions technologies at scale would also help reverse ocean acidification, per the IPCC 2021 report.

There are some who think that negative emissions tech will create a moral hazard for companies to keep burning fossil fuel and slow down the energy transition.

But this might not be true. We’ve seen how solar, wind and batteries keep getting cheaper and better. That has driven continuing coal and gas plant shutdowns and will begin reducing demand for oil.

Incentives will likely play a bigger role than negative emissions tech in fossil industry decision-making. Coal mines are being decommissioned at a historic pace as over 80% of new installed electricity capacity is from renewables and largely solar and wind. Financing for fossil fuels is becoming increasingly difficult, even in developing countries like Vietnam.

Overall, there is a strong economic and climate mitigation case to make progress on negative emissions technologies and to start scaling now.

5.2 Direct Air Capture Is A Promising Approach

We need to try a thousand approaches to removing CO2. Soil carbon, regenerative agriculture, macro algae, biochar, geologic weathering, biomass conversion and obviously reforesting and planting more trees. But technology could well be the fastest way to scale carbon removal to multi-gigatons per year.

Direct air capture (DAC) broadly is a group of chemical technologies that remove CO2 from the atmosphere directly. Even 12 years ago, filtering CO2 molecules that comprise over 400ppm (0.04%) in the atmosphere sounded expensive and impractical to all but a few scientists like Klaus Lackner, a pioneer in this space.

Yet 25 years ago, very few would have believed that in 2021 there would be roughly enough electrons globally harnessed from solar and wind plants to power the EU. Or that EVs would already outsell ICE cars in Norway, and within a few years, China and most of Europe. Just as solar and batteries, some types of DAC technologies are well suited for high learning rates as capacity scales.

Another benefit of DACs is the relatively low land requirements they have. A World Resources Institute report in 2021, coauthored by Noah McQueen, estimated that capturing the same amount of CO2 per year would require 40-2000 times less land than reforestation. Of course, there are other valuable reasons to reforest as well. Further, capturing 1 GT per year would require a parcel of land between 20-157 km on all sides (12.5-98 miles) but this would be distributed across capture sites.

The following simplified diagram shows a DAC capture process and sequestration or utilization. Most current DAC startups capture CO2 using a contactor (basically a filter) made from a solid sorbent or liquid solvent. The contactor allows other molecules in the air, primarily nitrogen and oxygen but binds CO2. Then using heat, pressure, moisture, electrochemical or other methods, the contactor releases the CO2. The result is a gas, usually of high purity, which can be compressed and transported.

Climeworks launched the DAC industry in 2009, having been started by two ETH researchers in Switzerland. Twelve years later, the industry has not yet seen any substantial traction.

Climeworks just launched Orca in Iceland, their DAC plant that can remove 4000 ton of CO2 per year, which is then mineralized into rocks underground by Carbfix. Carbon Engineering, or SeaChange's process to produce limestone in the ocean. the leading liquid solvent startup, plans to start operating their megaton-scale plant in 2024.

One of the problems with existing approaches for leading DAC startups like Climeworks is the heat required. This can drive up operating costs and requires a source of heat nearby, even if it is low grade meaning less than 100 degrees C. Carbon Engineering requires a significant amount of natural gas as an input, which is also problematic.

Electrochemical DAC Could Be Cheap and Scalable

Taking an electrochemical approach to release CO2 from the contactor would not require any energy other than electricity. This would lower operating costs as solar and wind generated electricity prices keep falling.

DAC startups applying electrochemistry include UK based Mission Zero. They claim the lowest DAC energy consumption in producing CO2 at $275/ton today, the cheapest quoted price amongst any player. Mission Zero also claims they can produce CO2 for less than $100 or even $50/ton within a decade. To hit $100/ton, they may require scaling to only 500 tons of CO2 removal per year, which is one to three orders of magnitude less scale compared to what their non-electrochemical competitors are aiming for. They are one of the startups Stripe has committed to purchase CO2 from.

Another startup leveraging electrons to capture carbon is Bay Area based Prometheus Fuels. They’re the first carbon capture company to reach unicorn status, and within only 2.5 years of launch. They use a different electrochemical process to separate the CO2 to make cost effective low carbon fuels that have led to BMW and Maersk, the shipping giant, investing in them. Net-zero carbon-based fuels are especially compelling for shipping vs say hydrogen, which would need to be converted to ammonia to compete on energy density vs so called “electrofuels”.

It is very early days for electrochemical DAC — Mission Zero’s technology is at lab scale for now. Despite that, it is a very promising approach for carbon removal. I think today, many are underestimating the rapid scalability and cost reductions possible with electrochemical processes.

Passive Air Flow DAC

Using passive air flow instead of fans is another way to minimize energy requirements and simplify designs.

One way to do this is to integrate DAC into HVAC units or ducts inside buildings. The first startup in the world doing this is Soletair Power. They have a product that integrates with a building’s HVAC unit (pictured), and another that is a smaller unit for use in meeting rooms. A benefit of this method is reducing “stuffiness” inside buildings due to high CO2 concentrations, which can improve cognitive function. Soletair is selling their capture units themselves as valuable products, in addition to combining the CO2 produced with green hydrogen to make clean fuels.

As with most other emerging industries, distribution will be key to company success. As we’ve talked about throughout this post, energy startups (and entire transitions) generally scale up by getting traction in niche markets, then expanding into increasingly larger markets.

5.3 Clean CO2 to “X” May Outcompete Fossil Fuels

Right now, the cost of CO2 from DAC is prohibitive for most markets.

Scaling electrochemical DAC units that are modular, small, cheap to make and powered by clean electrons is likely the easiest and fastest way to bring down carbon costs.

If this can be sustained, atmospheric CO2 will become a cost-effective input into making fuels, plastics and many other carbon-based industries. Just as with the rapid learning rates for solar and batteries, falling costs and higher deployments can expand addressable markets as shown in the figure below.

Which markets have the most potential to scale up DAC and electrochemical DAC in particular? There are two broad types of markets that startups are currently trying to sell into.

CO2 sequestration or mineralization. One is to earn carbon removal credits to sequester the CO2 underground or mineralize in rocks for thousands of years. Heirloom and Parallel Carbon among others are focused on this group of approaches, whilst 44.01 and Carbfix store the captured CO2 underground.

Right now only a handful of companies like Stripe, Microsoft, Shopify and Swiss Re are purchasing credits. The cost of CO2 removed from DAC is still high for now compared to nature-based carbon removal. Startups pursuing this path see permanent carbon removal (hundreds or thousands of years) as the largest market. They are targeting these early-adopter corporate buyers and seek to leverage government policies like the US-based 45Q tax credit.

CO2 utilization. The other type of market for CO2 is making things that are economic to sell. We will dive more into these markets. There already are existing so-called CO2 merchant markets, and promising growth markets for turning CO2 into other chemicals, fuels and much more.

To get traction and move down the learning curve, CO2 utilization startups are selling into a variety of markets. As we’ve discussed with solar and LIBs, scaling up in an increasing number of markets will drive the cost reductions and performance improvements over time.

5.4 Existing Markets: CO2 As End Product

Food and beverages. In the US, there’s over $450M annually in merchant CO2 sales for food and beverages. CO2 purity requirements (and prices) are higher for food and beverages, which makes it more appealing as an early niche to go after. For larger breweries, prices of $200/ton are possible when there’s no shortage. Purchasing smaller volumes means paying up to ~$3000-$4000/ton. This high cost is largely due to transporting CO2 hundreds of miles. Roughly two thirds of US CO2 made today is a byproduct of ethanol or ammonia plants located far away from the point of use.

Startup Noya had been targeting this market for their distributed DAC units. They integrate air capture into cooling towers on building rooftops, selling to local breweries and other food and bev customers. This allows them to save on transport and handling costs. They have only deployed one DAC unit at time of writing, but have hundreds of potential customers lined up.

CO2 refrigerants. Refrigerants are the substances used in refrigerators of all sizes, and the market was worth $2.7B in 2021 in the US alone. A problem with existing refrigerants like HFCs is their high climate warming potential, which is three orders of magnitude more than CO2. Recent legislation changes in California and other regions around the world will provide tailwinds, requiring using refrigerants with less warming potential. CO2 (aka refrigerant R744) is increasingly being used in some applications like grocery stores. Some startups are even selling more efficient solid-state chip cooling systems using CO2.

5.5 Emerging Markets: CO2 As Input Material or Additive

More and more startups are using electrochemical processes, like the one in this diagram, to turn CO2 into useful chemicals. Twelve turns CO2 into CO for conversion into various chemicals, fuels and plastics. Mars Materials is using CO to produce an intermediary chemical that is turned into carbon fiber. As mentioned earlier, a startup has secured hundreds of millions in funding to turn CO2 to fuels using an electrochemical process.

CO2 is already being used to make high grade diamonds. Startups do this by turning the molecules into hydrocarbons, then into the gems. Of course, given the prices of diamonds, the cost and effectiveness of the process is not as important. It is also increasingly being added to concrete or building aggregates, often improving material attributes.

Other emerging carbon utilization pathways use biochemical processes to convert CO2 (or biomass like corn) into valuable goods. Solugen, a Houston-based startup that became a unicorn in five years, synthesizes chemicals using a carbon-negative process from corn, air and water. They are already producing petro-free chemicals at, or below,  incumbent costs. Lanzatech uses a microbial fermentation process to turn a source of carbon monoxide into fuels, plastic bottles and more. Other startups are exploring pathways to using CO2-derived formic acid to produce chemicals using synthetic biology.

Air Protein feeds a specific type of microbe CO2, that can be converted into protein-rich meat alternatives.25 The microbes are fed nutrients and renewable electricity, producing a carbon-negative protein-rich flour that is then refined into protein.

Falling renewable and CO2 removal costs will help make these carbon utilization pathways increasingly viable for larger markets.

I think we are in the very early days of CO2 utilization, with huge potential to turn the 2.4 trillion tons of prehistoric plant matter in the atmosphere26 into things we use everyday.

5.6 Circular Is Nearly As Good As Permanent

We need permanent storage of CO2, and policy incentives like the 45Q tax credit, California’s LCFS and others will help. There may be enough corporations willing to purchase CO2 removal credits in addition to early adopters like Stripe, Microsoft, Shopify and Swiss Re to create a sustainable long-term market for permanent removal.

But I also think CO2 which is less permanently stored is a huge opportunity. Turning CO2 into fuels, chemicals, plastic bottles and more will also help bring down net emissions. Polymers last for years or decades and many can be recycled. Building materials like concrete can last for hundreds of years and are very unlikely to be re-released into the air.

As the CO2 utilization market grows, the amount re-emitted from “waste” CO2-made goods should become insignificant compared to how much we can capture and use. Electrochemical DAC can be a driving force for a low-cost and viable circular economy.

Also, as DAC deployments rise, CO2 costs keep falling and enable use in more applications. CO2 removed at less than $100/ton becomes a commodity, comparable to the price of bulk concrete ($125/ton average in US), and iron ore ($75-$200/ton). These are obviously not directly comparable, but are useful anchors for thinking about the usefulness of billions of tons of captured CO2 per year.

Throughout history, using carbon has been essential. From using wood for heat to fossil fuels and polymers, we have gained more control over harnessing carbon. In the 20th century we learnt how to manipulate carbon-based molecules to create synthetic materials. It seems likely that CO2 from the air could be the next phase of a more balanced relationship with carbon.

It has (tera)tons of potential.

6 Opportunity: Emerging Materials

It would be very surprising if there were no new commercialized materials resulting from the clean energy transition.

As discussed in the Past Transitions section, each energy transition has played a crucial role in unlocking new materials. Each transition is like walking into another room with a closed door. You know the room you are walking into will look and be very different, and in many ways, better. You just don’t know exactly what will change.

Each transition has enabled massive improvements to existing materials (wrought iron and later steel made using coal), created entirely new materials (polymers from oil-refined petrochemicals) and/or made low-cost manufacturing viable at scale (aluminum using electricity).

I think graphene and other 2D materials are strong candidates for emerging materials. Biomanufactured materials are also interesting. We’ll come back to the case for graphene and the current and potential applications.

But to first present a case by analog, let’s briefly look at the rise of low-cost electricity that helped aluminum production scale up. Having a reference point should help put emerging materials into perspective.

6.1 Low-cost Electricity Allowed Aluminum To Scale

Aluminum was only isolated for the first time in the 1820s, as it is only found as a very stable oxide (bonded with oxygen) in nature.

Before electrolysis, the lightweight metal was very expensive to produce at any scale from aluminum oxide. In 1852, it cost $545 per pound, or $1200 per kg to produce. This was almost double what gold cost at the time.

Despite this it attracted a lot of hype. Charles Dickens, Jules Verne and others envisioned future uses. After much R&D by many chemists and tinkerers as well as  funding, including by Napoleon III, a new process reduced the price to $8 per pound or $17.6/kg in 1885. This made the price comparable to silver. Everything accelerated from then.

The discovery of the Hall-Heroult process only a year later allowed for electrochemical manufacture of aluminum. By the early 1890s, the process’s commercialization had made aluminum even cheaper. It was $2 per pound ($4.4/kg) in 1890 and a mere $0.5 per pound ($1.1/kg) by 1894.

In forty years, the cost of aluminum had come down three orders of magnitude.27

Importantly, non-electrolytic processes had produced higher purity aluminum up until then. But within 5 years of the patents of both Hall and Heroult, electrolysis trumped them in purity too. Alcoa’s Niagara Falls plant opened in 1895 leveraging cheap electrons from the largest hydropower plant at the time. Many other aluminum plants using electrolysis also started production around then.

Driven by the cost declines, global production skyrocketed over 34x during the 1890s — from 175 tons in 1890 to over 6000 tons in 1899. The fall in cost as production ramped is similar to what we've been discussing for solar and wind, but in this case for a material, as illustrated here.

The Wright Brothers used an aluminum/copper alloy engine in their first flight in 1903, at a time when aluminum foil and cookware became available. Frames for WW1 airplanes helped drive demand, and later freight trains in the 1930s and the fuselages of planes from WW2 onwards were made out of the light but strong metal.

Today, a key driver for aluminum production is still electricity cost, just like crypto mining and other electron-hungry industries. Alcoa has been closing factories in the US and opening them next to currently cheap electron sources like hydropower in Iceland, Canada and Norway.

What’s clear from the early history of aluminum is that falling prices and scalable processes enabled rapid adoption by more markets. Also, the hype in the 1850s preceded the actual product-market fit starting the 1890s.

Low cost and scalable production were vital for the growth of solar PV and lithium-ion battery tech. I outlined above why I think it will be just as important for CO2 utilization.

I believe this will also be true for promising materials like graphene.

6.2 Graphene: Potential In Many Markets

Has there ever been a Nobel Prize won with a simpler tool than Scotch tape?

This humble tape was what the pioneers of graphene, Andre Geim and Konstantin Novoselov, used in 2004 to first isolate it by peeling off flakes from a chunk of graphite. They would win the Nobel Prize in Physics only six years later.28

Graphene is a single layer of graphite, the most stable form of pure carbon found in nature. When rolled up, a sheet becomes a carbon nanotube. Less than 20 years ago, researchers thought that single-layer materials could not even exist. Now, because of a range of useful properties, graphene could be well on its way to being used in many things.

Single-layer graphene is extremely thin at 0.33 nanometers. That is, over two million layers could fit into a standard 0.76mm thick credit card. Within that nanoscale layer, carbon atoms are arranged in the flat honeycomb lattice shown above.

Useful properties

Graphene’s potential starts with its properties, though it's worth noting these properties are the theoretical limits. As we'll see below, it is finding some uses in niche markets, but the properties are generally not as exceptional in real world uses.

Pound for pound, it is harder than diamond, but more elastic than rubber. It’s 100x stronger than steel but lighter (i.e. less dense) than aluminum. A hypothetical one meter squared graphene hammock could support about 4kg of weight, enough for a cat to nap on.

It is the best conductor of heat at room temperature — twice that of diamonds — and amongst the best electrical conductors, surpassing all metals. Graphene also has a very high surface/weight ratio  as only 6 grams are needed to cover a football field. Finally, its pure form is impermeable — it does not let anything through, not even tiny helium atoms.

Traction in a range of use cases

Given these properties, graphene is already in a wide array of commercial products, though mostly contained to smaller niche markets for now.

In energy storage, graphene’s ability to dissipate heat has been used by Huawei for smartphone battery heat management since 2016. Some fast-charging power banks recharge phones in as little as 17 minutes using its high electrical conductivity.

Lyten, a lithium-sulfur battery startup employing graphene in the anode for better energy density and faster charging. They are testing their batteries with the US military, aiming for their batteries to enter the EV market in 2026 with up to 3x the energy density of current LIB cells.

Startups like Skeleton Technologies are enhancing supercapacitors with graphene. Skeleton claims to have the best performing supercapacitors in the world, backed up by testing by the US Navy. They use a type of graphene that has a nanoscale 3D structure in their products. Skeleton has a $200M order backlog and has raised €70M so far.

In performance apparel and equipment, graphene has found many high end niche uses. Head offers tennis racquets used by Novak Djokovic, Naomi Osaka and other other elite players which have graphene in the shaft and head.

Vollebak sold a reversible jacket with graphene on one side, that traps heat extremely well and uses graphene’s antibacterial properties. Graphene clothing startups have Kickstarted hundreds of thousands of dollars. Inov8 shoes, used by some of the world’s elite endurance runners, added graphene to some shoe outsoles. More recently, they released soles with graphene-foam that doubled shoe longevity and are more resistant to wear.

Graphene improves performance in other composite materials and coatings too. Leading car wax companies are using graphene in their products for longer lasting paint protection. Graphene-based paints are being used for their protective anti-corrosiveness for ship hulls and offshore wind turbines. Only fractional amounts are needed as additives to concrete to strengthen it by 30%, reducing the amount of concrete (and cost) required.

Graphene also shows promise as a biosensor for medical and healthtech uses. Grapheal has developed a graphene-based 5 minute rapid COVID antigen test that won an innovation award at CES 2022.

Yet problems remain before the graphene market can continue growing. Production has been limited to hard-to-scale processes like chemical vapor deposition. Quality of the material, with many commercial variants, is hard to assess for buyers, and scammers abound.

Electrochemical processes could play a role in bringing down cost of production, and some promising early signs include a flash graphene process that uses energy from electrons to superheat rubber tires or landfill waste partially into graphene that can then be used to strengthen cement.

The overall scaling challenges are likely solvable.

The history of energy transitions and aluminum shows that promising materials will likely find their way into more markets as costs continue to fall.

6.3 Batteries As A Structure

Another interesting emerging material is not exactly a material.

The world will use exponentially more batteries in the next several decades. Why not integrate batteries directly into materials?

Tesla is developing battery packs that do double work as structural support, reducing mass by 10% and improving range 14%. This allows cost savings by using 370 fewer parts. This hints at the early days of integrating batteries into structural materials.

Researchers, with EU Commission and US Airfare backing, have made breakthroughs in “massless” structural batteries. The lab-scale batteries are made of carbon fiber, aluminum foil and fiberglass. The resulting material is already more stiff than timber, and the researchers think it could match aluminum’s rigidity too.

7 Transition Enablers

The clean energy transition moves fastest when all boats are rowing in the same direction. Startups, R&D, policy, financing and more.

Many things enable the transition, but I will touch on two: AI + software and funding.

7.1 Software and AI Are Key Enablers

Software and AI (including machine learning) are important enablers for the clean energy transition.

Tesla and other EV startups have leveraged their strengths29 in hardware, software and AI to offer features that legacy automakers lack in-house expertise to replicate. Tesla’s camera-based FSD system and custom chips are only possible with a highly skilled team of hardware, AI and software engineers. Software-based realtime traction control has become even more finely tuned with EVs — see Rivian’s torque vectoring as an example.

Computing power used in training AI systems has skyrocketed since a 2012 breakthrough in an image processing algorithm. As shown below, research by Open AI has found it has doubled every mere 3.4 months since then. That’s remarkable, and shows no signs of slowing down. Further, the amount of compute required to train a neural network to the same level of performance on a computer vision benchmark algorithm has halved every 16 months.

In the electricity sector, software is helping to lower friction. Seattle based Level 10 offers a booming marketplace that connects project developers with corporate PPA buyers. Arcadia allows software developers to integrate utility data into their apps — so that people and businesses can plug in to local renewables projects without changing utility.

Real-time software (e.g. Tesla’s Autobidder) allows battery storage plants to monetize and offer electricity services useful to the grid. Distributed energy resources like batteries, solar PVs and energy efficiency devices are being aggregated by Sunrun, Swell, Tesla, Stem and others to monetize participation in the grid.

Patch and Watershed both offer software platforms for companies to remove carbon and track emissions, while Pachama uses AI-powered satellite imagery to track corporate emission reductions.

Meanwhile, machine learning (ML) is increasingly being used to accelerate R&D for materials and chemicals discovery, speeding up time to market. Citrine, Kebotix, Exabyte and more startups are working on this problem. This type of compute is energy-intensive, and enough demand for ML in this domain could help drive demand for clean electrons from solar and wind. Another example is KoBold Metals, which accelerates the exploration of minerals essential for LIBs using AI modeling of mining sites.

Simulating chemical and physical reactions is also a very promising avenue for ML applications. Meta AI partnered with CMU in a project to accelerate the discovery of more effective catalysts to create fuels from renewables. Deepmind, an AI research lab owned by Alphabet, increased wind turbine output by 20% using ML. Their deep learning expertise is also helping a Swiss fusion research group autonomously control the magnetic coils to harness a stable plasma for fusion.

Given the pace of change, I can only hope to have highlighted some aspects of how AI and software are enabling the energy transition.

7.2 The Boom in Climate Tech Funding

Access to plentiful and patient capital is vital for climate tech startups building physical things.

There’s plenty of equity-based funding available, with over $60B invested worldwide in the first half of 2021 as shown below. Around one in seven of total VC dollars are being allocated to climate tech startups and there are now over 78 climate tech unicorns out of a total of less than a thousand unicorns overall.

Much of the funding is for mobility and transport startups, no surprise given the stratospheric growth of LIB use in EVs and stationary storage as we discussed. But tellingly, all verticals covered by the PWC report, except for the built environment, grew at over 90% in funding dollars year on year.

Helping startups bridge the gap to a series A are a number of accelerators, incubators and public sector funds and grants available.

VCs have learned to become more patient with climate tech, recognizing that investments in this space require more time for an exit than in internet companies. As we showed with CATL and Tesla’s incredible growth, returns can actually be greater in climate tech vs traditional ones and zeros VC. Indeed, a cleantech investor noted that the supposed bust of Cleantech 1.0 companies does not look as bad in 2021, after just over a decade. Returns on the $25B invested during the entire original boom  were about 20x in mid 2021, but would have been underwater back in 2016.

Interestingly, climate tech startups will raise more capital in the first half of 2021 alone than during the entire Cleantech 1.0 “boom”. Also anecdotally, an increasing number of people working in traditional tech (from Silicon Valley in particular) are pivoting into joining or starting climate tech companies.

Debt financing for larger companies is also booming, with over $269B invested in 2020, and 2021 looking like it will grow well beyond that. Parallel to the boom in clean capital, it has become increasingly difficult for fossil fuel incumbents to get access to capital themselves. Banks invested more in green bonds and loans in 2021 than in fossil fuel companies, and new coal plants can’t get financing whilst existing ones are increasingly uneconomic.

As challengers start hitting accelerated growth, the market share of incumbents start to fall quickly. This is happening for coal right now as solar and wind are sustaining high growth rates, and for ICE cars as EV sales boom.

Finally, the public markets also take notice of market share, as shown in the chart above. Historically, capital tends to desert incumbents when new challengers hold single-digit market share. This sounds surprising, but has played out many times, as shown in this table.

8 A Vision for 2045

There’s not enough optimism in looking forward to better futures.

This surprised me. The problem is massive, but I believe the solutions in progress and that will be developed are up to the challenge. Solar, wind and batteries will decarbonize most of humanity’s energy use and create massive opportunities in new and old industries.

This lack of optimism inspired me to write about possible futures by 2045. I draw a lot on what is written in this post. Go easy on this part as these are visions, and are not predictions!

8.1 Insanely cheap solar, wind and storage power almost everything

In 2045, all electrons are generated from renewables in all wealthier countries, and most middle income ones.

Solar, wind and batteries have made electricity incredibly cheap. Cheap to buy from utilities. Cheap to purchase directly from developers to use in energy-intensive industries. Cheap to buy your own clean energy for resilience. Coal and natural gas plants the world over have retired, unable to compete.

Solar PVs are mounted on most rooftops, and most new roofs have photovoltaics built-in. Many walls and facades also are harnessing energy from photons. Solar plants dot the small amount of land required for supplying most of the grid’s energy needs.

Intermittency has been solved. Wind is harnessed offshore even more so than on gusty plains worldwide. Most large scale solar installations include co-located stationary storage, cheap enough to discharge electrons for 12-24 hours. Scatterings of long duration storage help fill the “Dunkelflaute” weeks that are still and cloudy.

Solar and battery microgrids provide power for countless companies, buildings and remote locations.

8.2 EVs of all types and are taking over transportation

In 2045, EVs are everywhere on land, water and in the skies.

Just like the rapid shift from horses to cars, it is almost impossible to spot an ICE vehicle. EVs are just better — having more range at lower cost and can backup your home in a pinch.

Electric boats glide across harbors and along rivers.

Most vehicles are autonomous, making transport the safest in history. Trucks drive themselves between cities. Cheap, on-demand mobility is now a reality. Traffic and driving speeds in urban areas improve dramatically.

Small eVTOLs have replaced all shorter commercial air routes. People fly from Austin to Houston or London to Paris in less than an hour, and SF to San Jose in minutes. Energy density and material breakthroughs have enabled electric planes with dozens of seats to be cost-effective on longer routes too. People flying from SF to LA, Sydney to Melbourne or Shanghai to Beijing enjoy shorter door-to-door times at lower cost.

8.3 Battery storage is everywhere, but in EVs above all

In 2045, batteries serve as the primary connective tissue for energy. Massive scale storage banks sit alongside equally large solar and wind plants.

Storing even more energy than grid-scale batteries are the scores of LIBs in EVs everywhere. They provide energy to homes, buildings, companies and cities when needed, and vehicle owners are incentivized to share juice. No one worries about battery degradation, thanks to multi-million mile warranties and worry-free range.

8.4 Most existing industries are electrified, all are low-carbon

In 2045, renewable-powered electrification has happened faster than most predicted.

Long haul trips use fuels made using CO2 from the air, transformed into fuels using electrochemistry or biochemistry. These fuels have become cheaper than fossil alternatives. Shipping too, uses cheaper low-carbon fuel from either DAC CO2-derived fuels or clean ammonia from green hydrogen.

Buildings use heat pumps that work even in very cold climates, making gas furnaces obsolete for space heating, just as cheap clean electrons makes water heating with gas uneconomical.

Even industry is electrifying rapidly by 2045. Much steel is produced by cost-effective electrolysis, just as aluminum product-market-fit took off in the 1890s. Concrete is carbon-neutral thanks to CO2 additives. Mines, often remote, have almost all switched from fossils to electron-powered equipment and vehicles. Industrial heat uses a combination of green hydrogen and improved technologies directly converting electricity to heat.

All chemical refining and production is fossil-free, synthesized from the air or microbes and powered by electrons (and heat when needed) from renewables.

8.5 CO2 from the air is a key commodity

In 2045, using carbon is more important than ever. But unlike during the fossil era, it is now clean and cheap, opening up even more ways to use the vast CO2 stock we have in the sky.

Carbon removal has scaled up, slowly at first, then all at once, to multi-gigaton levels. Much of the removal is done using DAC. Electrochemical DAC in particular plays a key role, buoyed by the improving unit economics of clean, cheap electrons. These units are modular and small, mostly deployed near point of use or sequestration to minimize transport time and cost. Car and home air filters have CO2 capture built in, as do HVAC units in buildings of all sizes.

In 2045, captured CO2 from DAC is below $50/ton. Some CO2 has been sequestered underground or mineralized into rocks. But increasingly, it is used as a raw material. CO2 strengthens concrete and replaces the carbon needed for most existing petrochemicals and fuels. It is used to make polymers that were already known, but also new materials that have been commercialized.

We're also putting microbes to work to convert CO2 directly or indirectly in bioreactors. Microbes feast on CO2 from the air to cultivate cell-based meats, seafood and dairy. Pathways using synthetic biology are also being used to produce chemicals, plastics, fuels and performance materials.

Fossil fuels have very few if any competitive uses in making carbon-based products.

Nature-based carbon removal is also working. Regenerative land practices help restore wildlife corridors from barren land. Cultivated kelp forests are revitalizing marine ecosystems.

All of this means that in 2045 we are on the way to reusing the excess 2.4 teratons of CO2 in the air. Still, there’s many decades left before this vast resource will have been put to use.

8.6 Ubiquitous graphene, fusion and more

In 2045, graphene has far outgrown earlier niche markets. It is used in a large range of applications in energy storage, electronics, apparel, medical and high-performance materials.

Indoor air is less stuffy thanks to CO2 filters in HVAC systems. Streets are quieter and cities free of smog.

Indoor and vertical farming are the cheapest way to grow vegetables and berries. Cell-based and plant-derived proteins have almost eliminated the need for animal farming.

Additive and prefab manufacturing are cheaper than conventional construction approaches. Standardized, modular designs will look great, and emerging materials are affordable, clean and perform better than good ol' concrete and steel.

Nuclear fusion tech has advanced enough to be an emerging energy source in 2045, with some plants having enough energy gain to be supplying electricity. It is not yet cost-competitive with solar, wind or batteries but a new generation of founders see paths to get there.

Conclusion

There’s much to be done, but the future is bright.

Congratulations on making it to the end of this incredibly long post!

After deep diving here into the energy transition, I think we can be more optimistic about the future.

In the coming years and decades, we will have access to the cleanest and cheapest energy in history. The biggest opportunities and biggest changes are still ahead. Some of the world's brightest engineers and company builders are working on these as we speak.

Reach me on Twitter, LinkedIn or email me [at] tsungxu [dot] com.

The following founders and engineers at the frontier the energy transition reviewed sections of this article and made them better. Thank you, in no particular order, to Brian Heligman, Duncan Campbell, Jacob Brown, Douglas Onyango, Ryan Anderson, Grant Faber, Shanu Mathew, David Izikowitz and Ivo Stranic. Thanks also to Minh Lam for proofreading and last but certainly not least A-M Delas for being editor-in-chief.

Due to popular demand, you can download a PDF of the article.