You’ve probably noticed that burning stuff makes both heat and light. You can get more or less of each, depending on how you set things up, but the truth is we usually don’t want either. We want electricity.

We’ve gone whole hog on turning heat into electricity. Most of the electricity in the world is made by burning something to heat up water. When a water molecule heats up, it changes from liquid to vapor that is 1600× bigger by volume (that’s why popcorn pops). This massive expansion moves a steam turbine producing mechanical energy, that’s connected to a generator that spits out electrons. Whether you start with wood, coal, oil, gas or uranium, that’s where your electricity comes from.

In the last couple decades, we’ve also gotten pretty good at turning light directly into electrons with solar panels. The sun is super reliable, but the ass end of Earth gets in our way at dusk every day. What if we could make make our own sunshine?

The ingenious Lightcell burns fuel, like hydrogen, but is optimized to make more light and less heat. Many times brighter than the sun, it can convert a lot more of the chemical energy from fuel into electricity.

Swapping a Tesla battery out for a Lightcell would weigh 1000 pounds less and give it a 2,500 mile range. You could “recharge” in minutes at the gas pump! That might not be your ideal electric dream but the implications for drones and planes are extraordinary. There are lots of cases where liquid fuels are necessary or ideal. Lighter, quieter, safer, more efficient and reliable than internal combustion engines, Lightcell could go anywhere a generator is used and make better use of the fuel.

Thousands of new datacenters filled with millions of NVIDIA chips made with billions of transistors are being built to make trillions of dollars — doing multiplication.

1,000,000,000,000,000,000,000,000,000,000,000,000 multiplications.

You’ve probably never head the word undecillion, but that’s how many multiplications they will do all-in.

Every AI you’ve ever talked to is a pile of numbers getting multiplied together — billions of multiplications for every word it spits out. We got the idea in the 1950s by trying to imitate neurons in the brain, and for fifty years it barely worked because computers were too slow. Then video games happened. GPUs were built to draw explosions in Quake, and by coincidence, the math for drawing explosions is the same math for simulating neurons. With AI, neural networks took over the world, not because multiplication is the right way to think, but because we had warehouses full of chips that were good at it. This coincidence got expensive, now we’re multiplying numbers that never needed to be multiplied.

EvoChip went around the whole mess. They compile training data straight into logic, chains of the operations transistors do in a single clock cycle, essentially for free. Nothing gets pruned, quantized, or compressed, because there is no neural network to shrink — and there’s hardly any multiplying left to do. The model is the circuit.

SidePath, a Dell partner, tested EvoChip against the fastest TensorFlow configurations they could build, across seven public datasets. EvoChip came out 13× to 41× faster on ordinary CPUs, with 35× to 301× fewer parameters, at the same accuracy. A $200 FPGA cranks out over 3 billion inferences per second. An Arduino, the computer from your kid’s science fair project, ran a model in 521 bytes of RAM.

Speed isn’t even the interesting part. There’s a reason your plane doesn’t have an AI copilot: nobody can prove what a neural network will do next, including the people who made it, and you can’t certify a black box for jobs that kill people when they fail. An EvoChip model is a logic circuit. Same input, same answer, same number of clock cycles, every time. You can hand the schematic to a regulator. If AI is ever going to fly planes or run pacemakers, this is how it will have to work.

When we invested in EvoChip in 2024, this was 1000× speedups in the lab. Now the benchmarks are public, the products are launching, there are 100+ patents on the core, and LLMs are next. If you know people trying to fit AI into something that runs on a coin cell battery, certify it for something that can kill you, or who just can’t get their hands on a zillion H100s, send them to EvoChip.

You’ve read about a hundred miracle battery breakthroughs — we haven’t invested in any of them — you can’t buy any of them. Any new chemistry needing new factories is dead on arrival — nobody is going to rebuild a trillion dollars of manufacturing infrastructure to try out a clever new molecule. This is why the world is drunk on lithium.

The batteries you can actually buy — in your phone, your laptop, your car, your toothbrush — nearly all trace back to one man: John Goodenough, the electrochemist whose lab invented the lithium cobalt oxide cathode in 1980. Sony commercialized it, and the lithium age began.

Goodenough didn’t stop. His lab spent the next four decades methodically deleting problem elements from the periodic table of batteries. In 1984, his lithium manganese oxide cathode got rid of most of the cobalt. In 1997, his lithium iron phosphate cathode — LFP, the chemistry inside Teslas and BYDs — eliminated cobalt and nickel entirely. He was engineering critical minerals out of batteries decades before ”critical minerals” became a Washington buzzword. The final act was going after lithium itself: his lab’s sodium Prussian White cathode in 2013 became the foundation of CATL’s sodium-ion program, and in 2017 a researcher in his lab named Leigang Xue invented Potassium Prussian White — a high-energy cathode with no lithium, no cobalt, no nickel, no copper. The reviewers called it ”a significant breakthrough for the battery field.“ Goodenough himself rated its promise on par with LFP and LCO — his own inventions. Two years later he won the Nobel Prize. One lab in Austin invented the cathode in basically every battery you’ve ever owned, and then invented its successor.

Lithium is scarce, miserable to refine, and the supply chain — lithium, cobalt, nickel — runs overwhelmingly through China. Lithium cells also have a well-earned reputation for catching fire, which gets exciting when you stack a few thousand of them in a building full of expensive computers.

Group1 was founded to commercialize Xue’s invention, with Xue as co-founder. Potassium is about a thousand times more abundant than lithium. It’s potash — fertilizer — farmers buy trainloads of it. You ate some at breakfast; there’s half a gram in every banana. Nobody controls the supply because it’s everywhere, including a massive deposit in Michigan that the Department of Energy is backing with $1.26 billion to mine.

Group1’s potassium cells are made on existing lithium-ion production lines. Potassium slides into the same graphite anodes lithium uses — a trick sodium can’t pull off, which is why sodium-ion needs new factories and potassium doesn’t. Same slurry mixing, same coating, same assembly, same machines. No retooling. Group1 built the world’s first potassium-ion 18650 — same size and voltage as the lithium cell in a Tesla — and has already produced them on the commercial production lines of one of the biggest battery manufacturers on Earth. The trillion-dollar moat is their trillion-dollar head start.

The first customers are AI data centers. Every data center keeps a wall of backup batteries standing by for the moment the grid hiccups — power failures cost the industry about $20 billion a year. Today that wall is either lead-acid, a chemistry from 1859, or lithium cells wrapped in fire suppression. Potassium cells deliver huge bursts of power on demand, which is exactly what backup requires, without the fire anxiety and without a supply chain that routes through a geopolitical rival. Group1 is even closing the loop with the Michigan potash mine: their batteries will provide backup power for the facility that digs their feedstock out of the ground. American potassium, American cells, powering American infrastructure.

Goodenough spent forty years removing every troublesome element from the battery, and lithium was the last one left. We invested so Group1 can finish the job.

Some of you will have seen a recurring theme in my worldview over the last 20 years. The computational models we call AI suck, right up until they don’t. At that moment they can become superpowers. Often the kingpin is applying enough computation, using enough good data, or both. We will pass from the exaflop era to the zettaflop era this year, enabling a multitude of models to thrive that were infeasible until now.

MacroX is building models with alternative data to get to the heart of very important economic factors. With 195 countries in the world, approximately zero report accurate employment statistics. Many range between incompetent and fraudulent in reporting GDP. In any case, they report far less frequently than a diligent investor would prefer. MacroX can derive reliable trends in near real time using high dimensional, alternative data that is correlated in non-intuitive ways. For trading firms, this will become indispensable.

We expect MacroX to thrive based on their initial product which is rolling out into production now. That said, we are investing because we see a lot more places this will go as industries develop advanced computational models and use them for real time decision making. This is imperative to make them bigger, faster, cleaner, better.

Your friends at asset management firms will love you if you tell them about MacroX.

Power plants usually start by burning some crap. Wood, whale oil, coal. These days they mostly burn euphemistic natural gas. That heats up some water until it becomes steam. The steam expands, creating pressure to push a turbine, which is basically a big fan. The turbine turns a generator, which is essentially an electric motor run backwards and electricity comes out the other end. About 40% of the energy in the heat becomes electricity.

These electrons go down a fat wire to your neighborhood, then a smaller wire to your house, an even smaller wire to your power outlet, and finally a modest wire into your toaster. A toaster is made of a zillion tiny wires that turn those electrons back into heat. About 2/3 of that heat flies through the air into your bagel, and the rest just heats up your kitchen.

A datacenter is a colossal toaster. Currently, about 2% of all the electricity in the world goes into datacenters. Roughly 60% goes through the tiniest wires of all, computer chips. Nearly all of it becomes heat eventually. The rest goes into “cooling” equipment, which ironically, also makes more heat. A datacenter turns 97% of the electricity it consumes into heat.

So why don’t we just use the heat coming out of the datacenter to make more electricity? Well, this heat isn’t very hot, maybe 120°F. The Rankine cycle (eloquently described above) needs 1200°F steam. For a couple hundred years people have fantasized about solving this, but all the inventions still need temperatures over 450°F.

Spar Systems has invented the only efficient way to recover low temperature waste heat. These guys will install a machine at your datacenter, siphon off the heat you are trying to get rid of anyway and turn it into electricity that you can use to power toasters, or, you know, the datacenter. How much? They should be able to reduce your electricity bill by about 25%.

The team is focused on delivering megawatt scale systems next year, and the customer demand is directly proportional to NVIDIA’s market cap. Someday, when they run out of datacenters, the same technology can make shallow geothermal energy practical.

Hyperbole fails me when trying to describe the importance of computer chips to the world today. Everything made possible by computers relies on chips. Chips rely on transistors. Transistors rely on silicon. Silicon relies on Lithography.

Lithography the process of putting an image onto the surface of the silicon. Pretty much like the way a silkscreen puts “Team Building Exercise 1999” on a T-shirt. Except that this image has to be the highest resolution, with the smallest microscopic features, of anything humans produce.

“Moore’s Law” usually refers to increasing transistor density. Basically, how can we make transistors half as big as they were 18 months ago? Every time we figure that out, computers get twice as powerful.

The state of the art uses Extreme Ultraviolet (EUV) light to do the lithography. The machine that can do this cost $50 billion to develop. It has 500,000 parts. Only the Large Hadron Collider is more complicated. To buy one costs $250 million and you’ll be stuck on a waiting list that is $40 billion long. The machine comes from ASML in the Netherlands and they don’t have a single competitor, in the entire world.

That machine shoots a tiny ball of molten tin into a vacuum and blasts it with two lasers. This produces a flash of 13.5 nanometer ultraviolet light that gets aimed at the surface of a silicon wafer. You are looking at the pinnacle of human engineering achievement. Now you know how the chip for your iPhone is made.

ASML advanced from 193nm to 13.5nm light to make this possible, but there’s a problem. The diffraction limit of 13.5 nanometer light was set by either God or Issac Newton and there’s nothing we can do about it. We can’t print features smaller than that and there’s no practical way to do lithography with a shorter wavelength. When people say that Moore’s Law is over, this is why. We can’t keep making smaller transistors.

The semiconductor industry knows this, so they’ve tried to solve the problem by handing it over to the marketing department where the laws of physics don’t apply. You’ve seen them progress from 45nm to 30nm to 20nm over the last decade, then all of the sudden, 12nm, 7nm, 5nm & soon 3nm chips are coming. Well guess what, it is all just marketing bullshit.

This measurement in chips used to be half the distance between the centers of two features. Once marketing took over, they started measuring half the distance between the edges of two features. Instant improvement! Then they started measuring other random stuff. Other kinds of improvements in chip design helped to gloss over the fact that we are no longer able to shrink the size of transistors by 50% every 18 months anymore.

Today, there are extraordinary geopolitical machinations to control chip production. The U.S. has new tariffs and export controls akin to those for fighter jets and ICBMs (both are largely made of chips anyway). Access to chip production is as critical to superpowers as oil.

Lace Lithography use helium atoms instead of ultraviolet light to etch features onto silicon. The beam Lace will use to make chips is about the width of a single helium atom, or 0.1 nanometer. This is the only known technology that can go well beyond Extreme UV and put Moore’s Law back on track. ASML is worth over half a trillion dollars. Lace Lithography will be their successor.

Along with Vsquared, Runa & Future Ventures, we are their first investors.

I have been obsessed with robot hands for years. The human hand has 27 bones, three dozen muscles, and 17,000 touch sensors. It can thread a needle, crack an egg, and play flamenco guitar. The hand is the tool that built every other tool. Every other part of the robot problem has been getting solved — robots can see now, they can plan, they can walk, they can even do backflips on TikTok. Look closely at the hands on any viral humanoid robot video and you’ll see mittens. Grippers. Claw machines.

Robot hands have been stuck for fifty years, and the reason is a design philosophy. Engineers are trained to simplify: minimize the actuators, minimize the moving parts, control everything precisely. Every robot hand ever built follows this religion, with 15 to 25 degrees of freedom, modeled loosely on human joints. The hands cost as much as a sports car and still can’t pick a strawberry without juicing it. Robot builders kept assuming dexterity lives in the joints. It doesn’t. Your dexterity comes from a ridiculous, redundant tangle of muscles — most of them in your forearm — coordinated by a nervous system doing real-time control no roboticist would attempt.

Tom Zhang did. He grew up in a cave house in rural China and started building robots as a teenager in Singapore. His company, Daxo Robotics, builds hands the way nature does: throw in way more muscle than you need and let intelligence sort it out.

Daxo’s hand is driven by 108 artificial muscles — about three times as many as your hand — twenty in each finger, contracting and relaxing like biological muscle fibers. None of them is assigned a fixed job. An AI plays the role of the nervous system, discovering muscle combinations on the fly instead of executing pre-programmed motions. The redundancy means it degrades the way you do, not the way machines do: kill 20% of the motors and the hand keeps working, the AI just reroutes around the dead ones, like you favoring a sprained finger. The hand bends in ways your bones won’t allow, holds multiple objects independently, and does in-hand manipulation that no other robot hand can pull off. Tom refuses to even use the human hand as his benchmark — “if we engineer to match a human hand, we’ll always be suboptimal.”

The prototype cost about $1,200 to build, because it uses dirt-cheap $5 motors instead of the precision direct-drive units the rest of the industry worships. The first hand went together in two days. New iterations take hours. Research hands that do less cost six figures. When the hand finally works and costs less than your laptop, every humanoid robot company on Earth becomes a customer.

This is a kingpin for robotics. Stephen Wolfram signed on as an advisor. I signed up to invest. If you’re building robots that need to do real work with real hands, talk to Daxo.

How did the Romans build the Pantheon? Made of non-reinforced concrete – it remains standing after 2000 years, in a seismic zone – while our concrete with steel rebar struggles to last a century. No one has been able to figure it out for two millennia, but this team at MIT just solved the mystery and recently published about it in Science.

The implications are staggering. DMAT can make stronger, self healing concrete. It can be lower cost and is a very easy change to existing production operations, this is not true for any other significant advancement we’ve seen in cement. Even more exciting, this radically reduces the carbon emissions from making cement, which is the largest offender of all construction materials, almost 9% of global CO2 emissions and growing.

We are their first investors. Please listen to this conversation I had with Admir Masic — he’s the delightful genius who figured it out.

We’ve been promised that 3D printing would change everything for about forty years now. What we got is a machine that takes five hours to print a plastic dinosaur, one excruciating layer at a time. It’s like an inkjet doing one page an hour, except every page has to be glued to the previous one. The glue between layers is exactly where printed parts break, which is why 3D printing gave us a zillion prototypes and almost no actual products. Too slow, too weak, too expensive. Factories yawned and went back to injection molding.

Researchers at Lawrence Livermore figured out something wild: you can run a CT scanner backwards. A CT scan spins around you, taking 2D slices to build a 3D picture of your insides. Volumetric manufacturing does the reverse — project the slices as light patterns into a spinning vat of photosensitive resin, and a complete object materializes in the middle. The whole thing at once. No layers. No nozzle dragging around. The part you waited five hours for prints in 3 minutes.

MAV Unlimited built machines that work this way to print 100x faster than the fastest commercial 3D printers, at a tenth of the cost. I cannot wait for you to see them in action. Parts just magically appear in the vat.

No layers turns out to be a superpower beyond speed. Layerless parts are stronger, since there are no seams to fail. They come out optically smooth, which means you can print lenses, prisms, and light pipes that need zero polishing. MAV can vary the hardness of the material millimeter by millimeter, printing soft and rigid zones in a single part. They’re the only additive technology that can print directly over electronics and metals — the object forms around whatever you put in the vat.

The killer app is hiding in the AI buildout. Chips increasingly talk to each other with light, and the dirty secret of photonics is that attaching the optics is the expensive part — aligning a fiber to a chip takes minutes per connection on half-million-dollar machines, eating half of the module cost. MAV prints gradient-index couplers that catch light over a wide zone, so parts can be slapped on by a regular pick-and-place robot instead of lovingly aligned by a precision instrument. The Air Force is already on board; printing custom optics and parts at a forward operating base, from locally sourced materials, is the kind of thing that makes military logistics people emotional.

This is the way manufacturing always should have worked. Light is fast. Now the printers are too.

It is important to first understand that We are all made of energy.

Solar panels have been improving and hold great potential but they are rendered useless by the relentless and uncannily predictable onslaught of night. There’s a simple way to fix this – put the solar panels in space.

Space solar is an idea that has been around for decades, but it has been laughably expensive to launch enough solar panels on rockets for anyone to take it seriously. Still, the idea is enticing, because solar panels orbiting Earth get sun 24×7, 365 days a year. They receive 8 times as much energy as terrestrial panels. They can beam power down to Earth anywhere on the planet using radio waves, even in the middle of the night, during a snowstorm – right through clouds. No need for the miraculous breakthrough energy storage that wind & solar farms on Earth require.

What about safety? Beam forming technology makes it possible to send a diffuse beam of radio waves to a large antenna on the ground. The 10GHz band can’t penetrate your skin for more than a millimeter. I’ve stood in front of this beam and at most you can detect a slight warmth, much less than standing in the sunshine. Humans should never end up in the path of the beam, but if they do there is little to no health risk.

What about the ridiculous cost? Even uncontested in the private space launch market, SpaceX has gotten launch cost down 20x from the Space Shuttle. They’re at $2300, $1800, $1500 per kilogram. This idea gets irresistibly cost effective at about $300 per kilogram. The SpaceX target for Starship is $20 per kilogram. In something measured in years, not decades, space solar will become the lowest cost baseload energy anywhere on Earth. It will also be the primary economic driver for private space launch.

Why Virtus Solis? Most of what we’ve seen elsewhere is still in research but we are thrilled with this team’s focus on commercializing this technology as fast as possible. The engineering team is ambitious but humble and practical. They’ve consistently impressed us with better answers than we expected to our hardest technical questions. You can learn more from their blog or this interview with one of the founders.

The creation and destruction of electronics is a nasty yet unavoidable business. The creation involves mining for metals and the destruction entails polluting the environment. This is a growing global problem. What if there were a non-toxic, low-energy, scalable, and less expensive process to extract these metals out of the earth and electronics waste?

We’re thrilled to be working with Descycle, a clean technology company developing “Green” metals processing based on a novel class of chemistry, Deep Eutectic Solvents (DES) – simply mixtures of salts without water, in which metals behave differently than in water-based systems. Battery metals (e.g. nickel, copper, cobalt, rare earths, etc.); precious metals (e.g. gold, silver, palladium, etc.) can all be extracted through this process. This technology can obviate the use of pollutive smelters, toxic chemicals, and strong acids in metals processing.

Descycle has teamed up with mining companies and data center decommissioning partners to be the first-mover in many of these markets. The value of the positive environmental impact alone is hard to quantify. What is much easier to quantify is the value of the gold, palladium, etc. that is guaranteed to be in the e-waste. One ton of printed circuit boards will yield 253 grams of gold, 62 grams of. Palladium, and 1.2 grams of silver. We know this because we put in there. Mining a ton of stone might yield 9 grams of gold on a good day but we always strike gold in electronics waste.

We’ve known the Descycle geologists and management team for over 10 years. They have a long track record in finance and mining and are also investors in Descycle. There’s also great potential that eutectic processes possess in extracting iron, zinc, aluminum, arsenic, mercury, lead, chromium, and sulphur from contaminated environments. We’re elated to join Descycle in addressing one of the world’s biggest problems.

Recorded on April 19, 2025

The post Jensen’s Law appeared first on .

This is a place where hundreds of startups that are doing very difficult, very ambitious deep tech projects have gotten their start. It’s a community that’s helped each other out. Scott is one of the most soulful guys I know. He’s just trying to help create an environment where these things can thrive.

And I’m so proud of him and what he’s done. I think more people need to know about Newlab. It’s one of my favorite things whenever one of our companies gets involved in Newlab. When you decide that you want to leave your job and go work in a deep tech startup, you should just check out the menu at Newlab.

There’s so much we talk about. By the end, you will really understand Newlab. You’ll believe in them and what they’re doing. You’ll want to be a part of it.

Towards the end of this episode, we also talk about the coolest, newest thing that Deep Future is, working on.

Important Links

Newlab

Recorded on October 4, 2024

The post Newlab is the Deep Tech Epicenter – Scott Cohen appeared first on .

Recorded on August 16, 2024

The post General Solicitation Rules for VC – ØF appeared first on .

Cement is the most used material on the entire planet. Between 8 and 13 percent of global CO2 emissions are associated with making cement. Obviously, the most used construction material. We now know – thanks to Admir – how to make cement that lasts virtually forever, use less of it, use less steel, it doesn’t cost more, you can make it in any cement plant, and the kicker is it’s about 20 percent less CO2.

So this is a major breakthrough, Admir is the founder of a company called DMAT, which is commercializing this. I’m so proud to be one of their first investors with the Deep Future Fund.

I couldn’t find a more delightful human to hang out with all year long. Admir’s an amazing guy. He started his life out in Bosnia, became a refugee as a teenager in Croatia during the Bosnian war, and then managed to somehow, go get educated in Italy. Then he became, a researcher at the Max Planck Institute in Germany before becoming a professor at MIT, where he is now studying all the things that you can do with cement and other amazing things we can learn from material science in antiquity. We spend a lot of time talking about cement, but I promise it’s interesting.

We also spent some time talking about Admir’s childhood growing up, as a refugee in Eastern Europe and it’s such a soulful conversation. I’m so thrilled to share it with you guys and I hope you’ll share it with your friends because everyone can get some inspiration from this story.

Important Links

• DMAT

Recorded August 27, 2024 at MI

Recorded on August 27, 2024

The post Classic Rock, the Secret of Roman Concrete – Admir Masic appeared first on .

Recorded on August 16, 2024

The post You Must Buy a Ninja Creami – ØF appeared first on .

Now he has a company working on making robotics much more practical to integrate with AI, meaning how do you get the language models connected up to the vision models? So when you talk to a robot about what it needs to do, how do you correlate that input to what it’s seeing in the world.

Avi’s new company called GroundLight is building the toolkit you need to do that.

Important Links

• GroundLight

• Avi Geiger on LinkedIn

Recorded July 18, 2024 in Seattle

Recorded on July 18, 2024

The post Robots that See what You Mean – Avi Geiger appeared first on .

Recorded on August 16, 2024

The post I’m a Perplexity Junkie – ØF appeared first on .

Recorded on August 16, 2024

The post Electric Motortoys – ØF appeared first on .

Recorded on August 16, 2024

The post Mega SSN Leak – ØF appeared first on .