Mike Schroepfer on atoms, not bits: The Gigascale thesis for rebuilding the physical economy

‍

CTVC: Want to give your own introduction?

Schroepfer: Hey, I'm Mike Schroepfer. Most people call me Schrep. I popped out of Stanford with a CS degree in the late '90s and built companies for the next 25 years. Started my venture-backed startup in 2000, built Firefox for a while, then joined Facebook in 2008 and built most of the hardware, software, infrastructure, onboarded Oculus and Instagram, built our AR research lab. Did a lot of different things and decided a little bit ago to take on my next adventure. I thought the world was waking up to the important problem being atoms, not bits, and that was the future we were gonna build. There was a confluence of demand and technological innovation that meant we got a shot to rebuild our infrastructure in a way that our kids and our kids' kids are gonna be real happy about, and do so without costing anyone any money — in fact, saving folks a lot of money along the way. So that's what we're doing.

CTVC: And I think one of the things I remember you calling out in your story on LinkedIn was that you all were thinking about this atoms, not bits thesis for quite some time, and now it feels like it's playing out in the real world. Everyone's talking about physical economy, power demand. But back then it might have felt pretty non-obvious. What really made you take that bet in the first place? Especially coming from Meta, where everyone thinks software is eating the world. What made you think atoms and hardware would matter?

Schroepfer: At Meta we started the Facebook AI Research lab in 2013, many years before ChatGPT, and it's a similar process of looking at the fundamentals. Where are the fundamental technologies? Where are the economics going? Even five, six years ago it was obvious that solar was getting really cheap, batteries were getting really cheap, electricity demand was growing even pre-data centers. If you just talk about onshoring EVs, people replacing gas stoves with induction stoves, replacing gas heaters with heat pumps — just that, ex data centers, was gonna create an unprecedented demand on the grid in the US. It's one of these things where when you're at the bottom of the S curve, it looks like nothing's changing, nothing's changing, and then all of a sudden everything's changing all the time. The trick is to catch it a couple years into the flat part of the S curve and be laying the groundwork so you're ready to go. The fundamentals were there, plain to see.

CTVC: While you were at Meta, were there hints of that data center explosion, AI explosion coming? Did you see that demand growth story taking off as much as it has, or has that been a surprise?

Schroepfer: The scale of it has surprised a lot of people. I knew it would grow, but I wouldn't say I called it perfectly at this pace. We made an investment in Pantelosa, which is an open ocean data center suitable for inference, not training — made that many years ago. That bet was premised on: inference capacity is gonna be where it's at, and this is the cheapest way to build inference compute at terawatt scale. Everyone was chasing training when it first came along, and we just said, "Wait a second, these models gotta run somewhere." That's where the real infrastructure's gonna come out. So directionally correct, had the magnitude off a little bit.

CTVC: You were super involved in actually building data centers in your seat at Meta. I seem to recall a particular anecdote about a loading dock being a specific height. Tell us about your time actually building real things and how that's porting over into the types of founders you're excited to work with today.

Schroepfer: It builds real humility when you have to work in the real world, because everyone wants to think it's this really clever, gnarly intellectual problem. But the real world is a thousand annoying small things. I showed up at Meta in 2008. We were mostly a software company. Our first data center came online in Prineville, Oregon in 2011 — a building where we bought the land, designed the building, designed the largest fleet of servers in the building, the Type 1 Freedom Server. We had next-generation free air cooling, no chillers in the building. I showed up for an inspection before the building was done and there were wooden blocks on the loading dock. I asked what was up, and they said, "We graded the loading dock at the wrong angle so the truck can't unload." The thing you see in every Walmart in the world, we didn't get right in our brand new state-of-the-art data center. Imagine that, but a thousand a day — concrete didn't work, the chiller broke, the electrician didn't show up, local opposition. It builds a lot of humility. Just get to the problems and grind through them, and be willing to be in the details.

Meta is now building data centers in tents. Think about a wedding tent venue — you can put one up in an afternoon or a couple of days. Building a building takes a year or two. When time to compute really matters, why spend all that time erecting steel when mostly you just need to stop rain hitting the servers, which a tent can do just fine. Clever solutions to hard problems, but a lot of grind-it-out detail work.

The last thing, and this really informs the investments we look at: we delivered data centers consistently. I built 17 facilities around the world — Singapore, Europe, the United States — tens of millions of square feet, mostly on time and on budget. The reason we did it is we did the same thing over and over again. Everything in your life that has gone down in price is mass-produced. Everything that isn't — bespoke healthcare, custom homes — is a bespoke project. The more we can build new infrastructure technology as mass-produced things in factories, the easier it gets.

CTVC: That brings an interesting question — the role of innovation versus deployment, and when that tips into deployment. One of the challenges in the last seven years of climate tech is you have a ton of early-stage venture capital pouring into new innovation — 40 different SMR companies — but then you don't get to tip into deployment, and you don't have repetition. With that thesis, how do you see early-stage venture playing a role when it's really repeatability that gets this stuff to scale?

Schroepfer: That's where the architecture of the system really matters. On SMRs, we made a bet on Radiant Nuclear — it's a micro reactor, you can put it on a C-17. People talk about factory building, but this is literally putting the entire reactor core on a truck. Building fission reactors is really hard; building an existing one takes about a decade. The only way to go faster is to go smaller. Because they're so small and can repeat, and you can literally build everything in a factory, we think they'll be one of the fastest to deploy. Their plan is to deploy dozens of units in the first year. That's still not the scale I'd like to see, but it's way better than one, which is where a lot of these projects get to.

On the grid side — one of my favorites — anytime you try to order a power transformer, it's its own special snowflake project. Custom engineering for every one. What size, what voltage, mechanical engineers, 18 months lead time. Part of it is because it's 100-year-old technology and you literally encode a bunch of the system in the hardware — the voltages on each end are burned into the structure of the transformer. Then you've got a company like Karan Power saying, "We've been shipping power electronics in electric vehicles at the 10 million unit scale. Those modern power electronics are software programmable. I can drop a five megawatt unit on a pad. You can tell me anywhere from 400 volts DC to 35,000 volts AC — change it on your laptop." And then they sell the same product to a solar farm, a car charging farm, a data center. That's the future we're building.

CTVC: On the capital stack front — first of a kind, second of a kind. There might have been catalytic capital, government capital to finance these things, but the last two years have been really challenging. You still need lots of capital to manufacture this stuff, and venture capital isn't always the best fit. How are you thinking about that capital stack gap?

Schroepfer: Good news and bad news. There are tons of entrepreneurs watching this who are struggling to raise capital — as a former founder who got lots of nos on my Series A, I had someone fall asleep during my pitch. I remember exactly where it was, who it was, when it was. That was 26 years ago. It sucks to raise money and get nos.

The good news is I'm not trying to systematically solve this problem. I'm not trying to get every company funded. We're trying to pick outliers — the company that's gonna run the gauntlet when no one else could. Commonwealth Fusion Systems is building fusion, the most high-risk technical path but the most promising to deliver clean power to the grid. They've raised billions of dollars because they have an unbelievable team taking the most scientifically proven approach. We've seen companies like Dioxycle, turning carbon emissions into clean chemicals — nobody likes to talk about chemicals, nobody likes to talk about reusing carbon — but Sarah is an unbelievable CEO who keeps getting non-dilutive financing from governments and has signed offtakes with L'Oréal and others. So I think of this less as what's the systematic problem, and more as: for this particular company and this founder, do they have the right to make it all the way because their product and team are just that good? Then we go to work and get behind them.

CTVC: You talk a lot about green discount as the antidote to green premium. What specifically are you looking for in these outliers?

Schroepfer: There are so many problems in the world that when you break them down and apply 2026 technology, you say, "There's a way to do this that is 5X better, cheaper, or faster." Let me give a specific example. Offshore inference data centers — there's a lot of talk about space-based data centers right now. It's 100 times cheaper to put a ton of material on a floating platform in the ocean than to launch it into space. So if anyone thinks space-based data centers are a great idea, I have a technology that starts with a 100X cost advantage. Even if I'm wrong by 10X, and then wrong by another 2X, I still have a 5X advantage. When you break it down — what is this thing made of, how do I construct it, what are the analogs in offshore oil platforms — you can weigh the steel, make it the same way you make a pylon for an oil platform, buy factory equipment from Italy to make exactly this. No special fixtures. It's a really simple machine.

One other example that's very different. Silkoa is making neodymium in Alameda, California — 6.5 kilogram chunks ready to go into a magnet, ready to go into an electric motor. For those who don't know, this is the key component for electric motors: wind turbines, EVs, everything. All of it's made in China right now. The way it's made in China is a terribly polluting, deadly process — fluorinated gases, factories with open roofs because it's so dangerous. If you want to do that process in the United States, you have to double-wall the containers and do all the safety requirements, which makes it insanely expensive. Silkoa has a totally different process using mine scraps and recycled goods. You can walk around the factory. It's totally safe. And because of that, their pitch is: we can make magnet-grade neodymium cost-competitive with China, and it doesn't emit anything. Because it's so simple, there's less stuff to fix and amend. They're hard to find, but once you find them, they're kind of magic.

CTVC: Is climate still your theme or thesis, or how are you tightening or refining that?

Schroepfer: Climate tech has always been an awkward framing. We've talked about alt proteins, fusion, neodymium production, offshore data centers, and you're like, "That's climate tech?" It's a weird framing, and I don't have a decidedly better one.

The way I describe it: we're a for-profit business. I do plenty of philanthropy, but this is a place where I've taken other people's money on a promise to give them back a lot. My 101 of business is we need to make stuff for cheaper than we sell it for, and there has to be lots of demand. That's the fundamental of every business we go after — it's a good business where the buyer doesn't need to care about the environment. They buy it because it makes their life better. We just add one extra filter: if this thing wins, what happens to pollution, to resilience, to public health? If the answer is nothing, or something bad, then it's not for us. I call it deep tech with a climate lens. I don't have the perfect soundbite yet.

CTVC: How much do you still think about the mitigation versus adaptation framing? And how much does that play into investing work at Gigascale, given it can be challenging from a returns perspective?

Schroepfer: We think about those terms, but I distill it down to: what happens here? We're gonna replace a polluting gas power plant with a zero-emission power source. Or we're gonna help people better understand wildfire risk so they can respond and prevent. We just did a grid resilience company putting sensors on the grid to detect when a grid fault causes a wildfire — if you catch it early, you can put it out before it becomes devastating. That emits a bunch of GHGs when you burn wood, and it's really bad for everyone in the area. Is that mitigation or adaptation? I don't know. Pretty sure it's a really great thing.

We use those words, but we don't have columns for them in an investing memo. I keep it simple: why is this business gonna be huge and make a lot of money, and what's the good we're gonna see in the world as a result — from human health, resilience, GHGs? One challenge I've had is people have a zero-sum mentality, like if you're doing X you're not doing Y. My view is there's so much stuff we need to do. Protect from wildfires, reduce pollution, reduce GHGs, deal with sea level rise — pick one and go for it. If you build things that people want, you get an economic engine, and then we don't have to make these choices.

CTVC: From when you were thinking about spending more time outside Meta than inside, comparing that time to now — thumbs up, thumbs down on the caliber of founder talent wanting to work on climate then versus now?

Schroepfer: It's gone in cycles. In 2020, 2021 that's all people wanted to talk about, and we were daily getting people wanting to work on it. Then AI took off, climate seemed harder than everyone thought, and a lot of people went away. But in the last 12 to 18 months we're seeing an even better version of it in two ways. One, people are coming at it with a much more economic lens — economics and whether it's a good business come first, because you can't just depend on the IRA to fund your business. Two, Tesla, SpaceX, Neuralink — there's a never-ending stream of amazing people coming out of those companies who've been trained on how to build new things, and they want to build new things in this area.

My problem is I don't have enough hours in the day to meet all the cool stuff I'm seeing. If you're not getting a response from me on email or Slack, it's because I'm overwhelmed, and everything is great, and I'm just trying to figure out which more awesome thing to go spend time with.

The other thing: AI has made this the best time in the world to start a new company. The leverage you get from two people building a company now versus seven months ago is remarkable. We have CEOs vibe coding firmware for their hardware companies. It's unbelievable.

And look — because we're friends I can say this — stepping away from a very large platform, it wasn't always easy. There was a period where I was like, "Did I make a big mistake?" I had to show up, introduce myself to LPs, and say, "Hi, my name is Mike Schroepfer, this is what I did." That sorts the ego a little bit. But that's part of what I wanted, to prove I could do it. I also just thought the opportunity was here.

It's much easier now because a bunch of our companies are working. You don't have to believe me anymore — go meet one of our founders and they'll tell you how their business is blowing up. But I do want to note: it has been hard, which means it will be hard again. I'm not assuming it's like this all the time. The reality of what's going on in the world is still here, and the economics are here. That's what we're gonna live with in five years, not whatever the current vibes are.

CTVC: What's the one thing you think people are missing or get wrong in this space?

Schroepfer: The most common mistake I see brilliant people make, particularly technologists, is not taking a systems view. We get pitched all the time: "I have a new thing that's five times better than the prior thing." And when you zoom out a little, you realize they made 1% of something 5X better, which is sort of irrelevant. The actual buyer cares about the big thing, and you made one part. I made one bolt in a 30,000-part car five times cheaper ... Nobody cares.

That's really hard for someone who spent five years of a PhD working on that thing. So the one thing I'd ask everyone to do is always zoom out and take a systems view. At the end of the day, someone is paying money for this thing. What does that person care about, and does your thing solve their top one or two problems? If it doesn't, it's gonna be a lot harder. Climate change is a systems-level problem. So the solutions have to be too.

CTVC: What would you buy, sell, or hold right now?

Schroepfer: I would buy compute supply, memory, power assets, land with good solar irradiance, anyone who can build grid capacity. What would I sell or hold? I'm more of a buyer than a seller. I can't think in the negative. I only think about where the opportunities are. I'm not a short-seller, sorry.