Crusoe

#1Cully Cavness grew up in a family where oil and gas weren't abstract industries.

#2His father worked in the business. So had his grandfather. Around the dinner table, energy was something people actually talked about. For Cavness, becoming the third generation of his family to work in oil and gas seemed like a perfectly natural path.

#3Then he went to Middlebury College.

#4Climate change was an unavoidable subject there. Environmental activism, including the work surrounding Bill McKibben, was part of the atmosphere. Cavness found himself caught between two worlds he couldn't easily dismiss.

#5He knew the people in the oil and gas industry. They were his family.

#6But he was also becoming increasingly troubled by what fossil fuels meant for the climate.

#7After college, a Watson Fellowship gave him a chance to explore that tension around the world. He studied different energy systems, seeing geothermal and hydropower in Iceland, coal in China, wind and solar in Spain, and other forms of energy infrastructure along the way. He later worked in geothermal development.

#8Eventually, though, he returned to oil and gas.

#9And there he encountered a problem that was difficult to ignore.

#10Oil wells can produce natural gas along with oil. Sometimes there's a pipeline nearby that can carry that gas to market.

#11Sometimes there isn't.

#12Or the pipeline is already full.

#13When getting the gas to a buyer costs more than the gas is worth, producers can end up burning it at the well site.

#14Flaring.

#15To Cavness, it looked absurd.

#16There was usable energy coming out of the ground, and the industry was setting it on fire.

#17But he also understood why companies did it. This wasn't necessarily a problem you could solve by simply telling an oil producer to stop. If there was no economical way to move the gas, the incentives didn't change.

#18The more interesting question was:

#19Could not flaring the gas somehow become more profitable than flaring it?

#20Around the same period, Cavness had another energy problem in his own house.

#21He was experimenting with Bitcoin mining in his basement.

#22The machines consumed enormous amounts of electricity. They threw off heat. The change was noticeable enough that his wife noticed what the experiment was doing to their home and electricity use.

#23At work, Cavness was watching energy get burned because there was nowhere useful to send it.

#24At home, he had a computer desperately hungry for energy.

#25Those two scenes didn't look related.

#26Until they did.

#27Cavness had known Chase Lochmiller since high school in Colorado.

#28They had climbed together when they were younger, but after school their careers had gone in very different directions.

#29Lochmiller studied mathematics and physics at MIT.

#30For a while, he imagined becoming a theoretical physicist. He did scientific research, but eventually discovered something about himself: the pace of fundamental research didn't suit him particularly well.

#31So he moved into a much faster world.

#32He worked in quantitative trading, using mathematics and computing to model financial markets. He later earned a master's degree in computer science at Stanford, specializing in artificial intelligence.

#33Then he became a general partner at the cryptocurrency investment firm Polychain Capital.

#34That gave him another kind of education.

#35Bitcoin's proof-of-work system turns electricity and computation into an economic product. If you're running a mining operation, electricity isn't some background utility bill.

#36It's one of the central variables in the business.

#37In 2018, Lochmiller left Polychain.

#38Then he went after a goal he'd been thinking about for years: Mount Everest.

#39His fascination with mountaineering stretched back to childhood. He had already attempted Everest in 2014, but that expedition ended after an avalanche in the Khumbu Icefall killed 16 Sherpas.

#40Four years later, he returned.

#41This time, he reached the summit.

#42But the idea for Crusoe didn't arrive in some cinematic revelation at the top of Everest.

#43It came after Lochmiller returned to Colorado and was thinking about what he wanted to do next.

#44That's when Cavness called.

#45The two old friends went climbing in the Rockies.

#46Cavness wanted to hear about Everest.

#47But according to Lochmiller's later recollection, much of their conversation ended up being about something else entirely.

#48Flaring.

#49Cavness knew why oil producers were stranded with natural gas they couldn't economically sell.

#50Lochmiller knew why Bitcoin miners were constantly searching for cheaper electricity.

#51Put those two pieces of knowledge together and the usual energy question started to look backward.

#52Normally, you begin with the energy.

#53There's gas over here.

#54How do we get it to the customer over there?

#55Build a pipeline. Connect it to the grid. Transport the energy toward the place where it can be consumed.

#56Cavness and Lochmiller flipped the direction.

#57If moving the gas to the computer is too difficult, why not move the computer to the gas?

#58In 2018, they founded Crusoe.

#59The basic architecture sounded almost ridiculously simple.

#60Take natural gas that otherwise would have been flared.

#61Run it through a generator.

#62Produce electricity right there at the well site.

#63Then place a modular data center next to the generator.

#64Now the gas didn't need to travel hundreds of miles.

#65The computation could happen beside the energy source, and the resulting information could travel over the internet.

#66Cavness would later describe the idea as a digital pipeline.

#67Instead of transporting a molecule, turn its energy into computation and transport bits.

#68Even the name Crusoe fit the concept. Like Robinson Crusoe making use of the resources available on an isolated island, the company was trying to make something valuable from resources stranded in remote places.

#69There was still one major question.

#70What computation should all that electricity actually perform?

#71Looking backward from today's AI boom, it can seem almost inevitable that Crusoe would put GPUs beside energy sources.

#72It wasn't inevitable at all.

#73In 2018, Bitcoin was a far more practical first workload.

#74Bitcoin mining consumes huge amounts of electricity, so cheap energy creates a direct economic advantage.

#75It is also unusually indifferent to geography.

#76A miner doesn't need to sit in downtown San Francisco or Northern Virginia. If the machine has electricity and enough connectivity to participate in the Bitcoin network, it can operate somewhere extremely remote.

#77That mattered in oil fields, where high-capacity networking could be difficult or expensive.

#78And Lochmiller already understood the economics of proof-of-work from his time in crypto.

#79So Bitcoin wasn't simply an embarrassing first business that Crusoe would later outgrow.

#80For the infrastructure Crusoe was building in 2018, it was an unusually good first customer for electricity that nobody else wanted.

#81The company began installing computers in oil fields.

#82And doing that turned out to teach Crusoe much more than how to mine Bitcoin.

#83An oil field is not where anyone would choose to operate a data center if convenience were the goal.

#84Crusoe had to deal with the gas.

#85It had to generate electricity.

#86It had to manage electrical equipment and distribute power to machines.

#87Those machines produced heat, so they had to be cooled.

#88They needed networking.

#89When equipment failed, someone had to keep a remote computing facility running in a place never designed to host one.

#90There was another problem: oil and gas wells change over time.

#91Production can decline.

#92So a data center attached to a particular source of gas couldn't necessarily behave like a conventional building expected to remain in the same place for decades.

#93Crusoe made its systems modular and movable.

#94Site after site, the company found itself dealing with layers of infrastructure that are often treated as separate industries:

#95energy,

#96generation,

#97electrical systems,

#98cooling,

#99networking,

#100compute hardware,

#101and remote operations.

#102Bitcoin was what the machines were calculating.

#103But the organization was learning something broader:

#104how to keep large amounts of power-hungry computing alive wherever the energy happened to be.

#105That didn't mean a Bitcoin mine could later be turned into an AI cluster just by swapping one box for another.

#106AI training is much harder.

#107Thousands of expensive GPUs have to work together. They need extremely fast interconnects. Cooling density rises dramatically. Reliability becomes much more important when an enormous distributed training job depends on the entire cluster.

#108Crusoe would have to build those capabilities later.

#109But underneath them was a lesson it had already learned the hard way.

#110Before you can solve the compute problem, you have to solve the energy problem.

#111Crusoe's Bitcoin business started working.

#112That should have made the company's strategy simpler.

#113Instead, Lochmiller kept complicating it.

#114Even around the Series A period, he was talking to investors about artificial intelligence and other forms of high-performance computing.

#115Later recollections from early investors capture how odd this sounded at the time.

#116The Bitcoin business was working.

#117Why not just focus on that?

#118In 2019, there was an even stranger preview of what might come later.

#119A Microsoft infrastructure executive discussed with Crusoe the possibility of using flared natural gas to power Azure data centers.

#120This was around the time Microsoft made its first $1 billion investment in OpenAI.

#121Lochmiller pushed the thought further.

#122Perhaps Crusoe could provide compute for OpenAI itself.

#123Nothing came of it.

#124The market wasn't ready for that idea at anything like the scale that would arrive later.

#125Crusoe kept mining Bitcoin.

#126But it didn't stop experimenting with other kinds of computing.

#127By 2021, names that had little to do with Bitcoin were appearing in Crusoe's own materials.

#128MIT's Computer Science and Artificial Intelligence Laboratory.

#129Folding@Home.

#130OpenCV.

#131They were early users of computing infrastructure outside Crusoe's mining business.

#132Then, in April 2022, the company announced a major financing round and laid out plans to expand CrusoeCloud.

#133The workloads it described included AI research, machine learning, computational biology, drug discovery and simulation.

#134It was already offering NVIDIA GPUs such as the A100.

#135The date matters.

#136April 2022.

#137Roughly seven months before ChatGPT was released.

#138That makes the neat version of the story — crypto crashed, ChatGPT appeared, and Crusoe suddenly pivoted to AI — hard to sustain.

#139Bitcoin was still a major part of the company.

#140Crusoe hadn't abandoned it.

#141Instead, a second computing business had been developing alongside it.

#142Then the world changed.

#143ChatGPT arrived at the end of 2022.

#144In 2023, the race to build generative AI systems exploded.

#145Companies scrambled for NVIDIA GPUs.

#146But buying GPUs turned out to be only the beginning.

#147Thousands of accelerators needed high-speed networking so they could behave like one enormous machine.

#148They produced extraordinary amounts of heat.

#149They needed buildings designed for much higher power densities.

#150And every one of those machines needed electricity.

#151A lot of it.

#152Crusoe expanded its H100 and A100 capacity and deployed NVIDIA Quantum-2 InfiniBand networking. It arranged hundreds of millions of dollars in financing to acquire more GPUs.

#153And suddenly its unusual history mattered.

#154Many companies had started with the compute and were working downward through the stack.

#155We have the GPUs.

#156Where can we put them?

#157Where can we get enough power?

#158Crusoe had spent years moving in the opposite direction.

#159We have energy.

#160What computation should we put beside it?

#161A question developed in remote oil fields had become a much larger question for the AI industry.

#162In 2024, Crusoe took on a project in Abilene, Texas, at a scale that made its early oil-field installations look almost unrecognizable.

#163Working with Lancium, it began developing a massive AI data center campus.

#164The initially announced project was 200 megawatts.

#165The broader campus grew to a planned 1.2 gigawatts.

#166Eight buildings.

#167Roughly four million square feet.

#168High-density AI infrastructure.

#169Direct-to-chip liquid cooling.

#170Infrastructure designed for enormous numbers of advanced AI accelerators.

#171The physical scale had changed completely.

#172The questions hadn't.

#173Where does the power come from?

#174How do you distribute it?

#175How do you remove the heat?

#176How do you keep huge amounts of computing running reliably?

#177Abilene also produced one of the stranger echoes in Crusoe's history.

#178The project wasn't simply announced from day one to the public as "Crusoe is building Stargate."

#179Crusoe began developing a hyperscale AI campus, and as the commercial structure evolved, Abilene became the flagship campus for Stargate, with infrastructure on Oracle Cloud Infrastructure supporting OpenAI workloads.

#180The first phase went live in 2025.

#181Back in 2019, Lochmiller had floated the idea of Crusoe providing compute for OpenAI.

#182Nothing happened.

#183Six years later, Crusoe was building infrastructure that OpenAI would actually use.

#184There is no evidence that the first conversation directly caused the second event.

#185But the symmetry is difficult to miss.

#186On March 25, 2025, Crusoe announced a decision that would have sounded bizarre during its early years.

#187It was divesting its Bitcoin mining operation and Digital Flare Mitigation business to NYDIG.

#188This wasn't the disposal of a tiny failed experiment.

#189The business had grown to more than 425 modular data centers and more than 250 megawatts of capacity, with operations spanning seven U.S. states and two countries.

#190About 135 employees were expected to move to NYDIG.

#191Crusoe was handing over the business that had given the company its reason to exist.

#192Not because Bitcoin had simply failed.

#193AI infrastructure had become so capital-intensive and so demanding of people and organizational attention that trying to pursue both businesses meant splitting the company.

#194Crusoe chose AI.

#195But in a strange way, selling the Bitcoin business wasn't a rejection of what the company had learned from Bitcoin.

#196It was the opposite.

#197The workload disappeared.

#198The infrastructure philosophy survived.

#199In October 2025, Crusoe announced the initial close of a $1.375 billion Series E at a valuation above $10 billion.

#200By 2026, the numbers had grown again.

#201Crusoe said it had contracted roughly 4.9 gigawatts of AI infrastructure capacity. Its broader development pipeline, including projects at earlier stages, exceeded 40 gigawatts.

#202Beyond Abilene, the company announced another 1-gigawatt AI campus in Childress, Texas.

#203It was exploring more onsite generation.

#204Large-scale power protection systems.

#205Even partnerships around nuclear-powered AI infrastructure.

#206Because as AI grew, the problem Crusoe had obsessed over from the beginning returned in a new form.

#207Where does all the electricity come from?

#208If a data center has to wait years for a grid connection, owning GPUs doesn't solve much.

#209So the old oil-field logic comes back.

#210Instead of treating electricity and computing as two systems that happen to meet at the end, what if you design them together from the beginning?

#211In September 2026, reports said Crusoe had secured more than $3 billion in new funding at a valuation of roughly $30 billion.

#212Around the same time, reports said the company had struck a five-year AI cloud agreement with Jane Street worth about $13 billion.

#213Those two latest figures are reported numbers rather than the kind of detailed valuation announcement Crusoe itself made for its Series E, so they need that qualification.

#214But the scale of the transformation is hard to miss.

#215In 2018, Cavness and Lochmiller were looking at natural gas being burned in remote oil fields because nobody could economically use it there.

#216Their answer wasn't to figure out how to transport all that energy to a conventional data center.

#217They transported the computers to the energy.

#218At first, those computers mined Bitcoin.

#219That business forced Crusoe to learn about generation, power distribution, cooling, networking and operating large amounts of computing in difficult places.

#220Years later, the computers changed.

#221Bitcoin ASICs became AI GPUs.

#222Small modular oil-field data centers became gigawatt-scale AI campuses.

#223Eventually, Bitcoin itself disappeared from Crusoe.

#224But the question that created the company never did.

#225If anything, the AI boom made it much more valuable:

#226Why move the energy to the computer if you can move the computer to the energy?

#227Crusoe wasn't the only crypto company to find a second life in AI. CoreWeave once ran GPUs for cryptocurrency mining too. So how did it become, within a few years, one of the AI industry's most important specialized cloud providers?