It was a sunny day in August 2017, and CoreWeave CEO Michael Intrator was beginning to think he might actually set his Manhattan office tower on fire.
Dozens of powerful Nvidia GPUs were plugged into every available outlet, churning through the convoluted math required to mine cryptocurrency, and they were throwing off a ton of heat. With the air conditioning turned off for the weekend, the sixth-floor Financial District office had turned into a 130-degree hotbox. “I flipped. This wasn’t really built for this,” Intrator says. “I was just like, ‘We’ve got to go. We’ve got to go.’ ”
He and his cofounders, Brian Venturo and Brannin McBee, scrambled, frantically disconnecting servers that were almost too hot to touch and loading them into a pickup truck before heading to New Jersey, where a garage properly ventilated with a gigantic exhaust fan waited to cool them.
Nearly a decade later, some things haven’t changed. Intrator is still running GPUs, but on an exponentially larger scale. Now he’s got some 250,000 of them warehoused across the 33 data centers that CoreWeave operates, renting out access to their number-crunching prowess to compute-hungry customers. Once popular for rendering graphics for video games and mining crypto, GPUs (graphics processing units), commonly designed by chip giant Nvidia, have become a crucial engine for artificial intelligence. Access to them can determine a company’s fortunes, and CoreWeave, which trades on the Nasdaq, has built a $50 billion (market cap) business by becoming one of their go-to brokers, serving the likes of Microsoft, OpenAI and Meta in the race for AI ascendency.
There is still the risk of fire—at least in a monetary sense. After pioneering a new kind of financing that helped CoreWeave borrow $29 billion, largely in loans backed by its GPUs, Intrator now balances shareholders’ qualms about deepening debt and widespread fears about an AI bubble with his broader ambition: to provide compute not just to the tech giants but to every company integrating AI into its operations.
It’s not an unreasonable goal. In less than a decade, the Livingston, New Jersey–based business has spun up a data center empire, booking $1.9 billion in revenue in 2024 (net losses: $860 million, for a negative 45% margin) and $2.2 billion in the first half of 2025 (net losses: $605 million, or –28%). By 2031, it hopes to bring in another $30 billion, mostly from contracts with A-list customers like IBM and Meta and a string of AI startups such as Cohere and Mistral. Its stock price has more than doubled since its March IPO, trouncing the broader Nasdaq’s return of 23.9% and minting at least five billionaires in the process. Intrator, 56, is now worth $6.7 billion, landing him on The Forbes 400 list of the richest Americans for the first time, along with cofounder Venturo, 40, who’s worth $4.2 billion. Cofounders McBee and Peter Salanki, as well as early investor Jack Cogen, are also new billionaires thanks to CoreWeave.
The company’s remarkable growth has stemmed from ferocious global demand for GPUs. Startups have been strapped for those sophisticated chips since ChatGPT exploded in popularity in 2022, with even OpenAI in February delaying rolling out a major model upgrade because it ran out of chips to run it on. That means whoever has GPUs holds the keys to an enormous business opportunity: The AI cloud computing market was $230 billion in 2024 and, assuming the boom continues, is poised to explode to $400 billion by 2028, according to Wall Street analysts. “I’ve never seen a supply/demand imbalance like this in the 40 years I’ve been investing,” says Mark Klein, CEO of SuRo Capital, which invested a total of $25 million in CoreWeave.
Thanks to good instincts, a substantial appetite for risk and a dash of luck, Intrator realized early that his crypto company’s GPU stockpile could be good for more than minting tokens. A strategic relationship with Nvidia helped with the rest, positioning CoreWeave as one of the predominant cloud providers for the AI era.
At first Silicon Valley investors weren’t convinced that CoreWeave could hold its own against well-capitalized cloud titans like Amazon, Microsoft and Google, which remain both arch-competitors and, in some cases, important customers. “The overwhelming majority of [investors] thought we were insane,” Intrator says. Spurned by VCs, he turned to the debt markets, what he calls “East Coast money.” Top-tier private creditors like Blackstone and Magnetar financed the multibillion buildout of CoreWeave’s infrastructure so it could capitalize on AI’s insatiable appetite for compute. Debt financing also meant that Intrator and his cofounders held on to more precious equity; Intrator retains 13%, Venturo 7%. The scale is the key, Venturo says. “The capital intensity of this business is all that matters.”
Critics are spooked by the $11.2 billion in debt on CoreWeave’s balance sheet. And with interest rates ranging from 7% to 15%, expenses remain high: The company spent more than $250 million paying interest on that debt last quarter on just $19 million in operating income. Plus about half of CoreWeave’s assets are GPUs, which are said to depreciate over six years—or less, now that Nvidia is upgrading its chips faster. D.A. Davidson analyst Gil Luria doesn’t mince words: “This is a value-destructive entity.”
Naysayers also point out that 62% of CoreWeave’s $1.9 billion revenue in 2024 came from one extremely large customer: Microsoft. That dependence won’t last long, though: OpenAI has committed to spending a total of $15.9 billion on CoreWeave’s infrastructure over the next five years and received a $350 million stake in the company. CoreWeave recently inked a deal to supply computing power to Google, and IBM has been using CoreWeave’s GPU clusters as well. “Our researchers can focus on the AI and less on the infrastructure,” says IBM Cloud CTO Hillery Hunter.
Then there’s the existential question: Is the AI boom actually a bubble? In the first half of 2025, investors plowed $104 billion into artificial intelligence startups at towering valuations, per PitchBook, betting that AI really will turn out to be as revolutionary as the internet or the automobile and upend entire industries. That hasn’t happened yet. Despite spending billions on generative AI, many businesses haven’t seen material returns on it. Most AI companies still haven’t turned a profit. Even Sam Altman, AI’s chief carnival barker, is warning that overexuberant investors could “get burned.” That’s especially worrisome for CoreWeave, given its precarious, highly leveraged position as the industry’s premier GPU provider. Put another way: If this bubble were to pop, CoreWeave’s business would rupture along with it.
Intrator’s talent for creative number crunching first emerged when he was a commodities trader at Natsource, a New York–based carbon credit trading and greenhouse gas asset manager. There, from 1998 through the Enron scandal to 2014, he learned how to turn pollution into a financial instrument—just as he would later do with GPUs.
After his first venture, a natural gas hedge fund, failed, Intrator and his cofounders pivoted, forming Atlantic Crypto in 2017 to mine ether. “We founded the company on a credit card,” Intrator says. “It was like, ‘Okay, I’ll buy a couple GPUs. What the hell?’ ”
When crypto experienced one of its periodic crashes in 2018, the cofounders bought thousands of GPUs from miners going out of business for pennies on the dollar and raked in an estimated $80 million in revenue from mining. “Ethereum kept them alive long enough for them to make the transition to AI,” says early investor Nic Carter.
Given ether’s volatility, Intrator realized the more lucrative bet was not fickle cryptocurrency but the equipment used to mine it. Their servers could be configured for different use cases: rendering 3D images, animation or early AI efforts. In 2019, they swapped out their “crappy” mining GPUs with data center–grade chips on a dollar-for-dollar basis. “Looking back on it, it was genius,” Venturo says.
Their first break into AI came through a collaboration with Eleuther AI, a Washington, D.C.–based research nonprofit, in late 2021. CoreWeave shelled out $2 million to buy a bunch of Nvidia’s A100 chips and built a cluster of 96 GPUs for the organization, which was training a powerful open-source large language model. It did what Eleuther wanted but was plagued by bugs, networking issues and storage problems. At one point the nonprofit’s employees had to use an external storage device to manually offload data to ensure the system wouldn’t crash. “That was the tuition to learn how to build the infrastructure,” Intrator says.
It was also a path to some early customers. Through Eleuther, CoreWeave found Stability AI, the creator of a viral early AI text-to-image generator called Stable Diffusion. That’s when Venturo saw the first outlines of an enormous business opportunity. But CoreWeave would need to double down on Nvidia’s chips to seize it. It would be “the most important thing this company is ever going to do,” he says.
To buy those GPUs, CoreWeave needed cash—a lot of it, and fast. In late 2022, Evanston, Illinois–based asset manager Magnetar Capital invested $100 million, giving CoreWeave the ammunition to buy thousands of Nvidia H100 chips and provide computational power for AI at scale. It later co-led a $2.3 billion round, CoreWeave’s first GPU-backed loan. Intrator’s genius: convincing Magnetar to secure the funds using the very chips he was buying as collateral. “They laid the groundwork for debt financing in the GPU space,” says Darrick Horton, CEO of cloud computing rival TensorWave.
It was an unprecedented and risky move, but when ChatGPT took AI mainstream in November 2022, Intrator’s gamble paid off. In January 2023, CoreWeave signed a $300 million deal with early AI darling Inflection to build a 22,000-GPU AI cluster, one of the world’s largest at the time. “We went from having nothing sold to having everything sold five times over,” Venturo says.
By December 2022, CoreWeave had bought so many GPUs from Nvidia that its leadership took notice. The cofounders got a call out of the blue; Nvidia CEO Jensen Huang wanted to talk. Twenty minutes later, they were on a Zoom call with him, pitching the cofounder of the world’s largest semiconductor company on investing in a no-name former crypto miner.
Huang saw in CoreWeave an opportunity to strengthen Nvidia’s dominance in the AI market by propping up a future buyer for its complex chips that had the technical expertise to run them efficiently, Intrator says. Nvidia invested some $350 million in CoreWeave for a 5% stake (now worth $2.5 billion) and granted it “elite” status on its partner network, vaulting it to the front of the line for the company’s most cutting-edge chips. (Nvidia declined to comment for this story.)
It’s quite the incestuous relationship. (Intrator prefers “symbiotic.”) Nvidia is a vendor, an investor and a customer; it has paid CoreWeave at least $320 million for its infrastructure and agreed to buy its unsold capacity through 2032. But it is by no means a relationship of equals. In fiscal 2025 (ended January), Nvidia made $72.9 billion profit on $130.5 billion revenue. Its current market capitalization exceeds $4.3 trillion. D.A. Davidson analyst Luria puts it directly: “CoreWeave wouldn’t exist if Nvidia didn’t want it to.”
As CoreWeave landed larger contracts, it took on more debt to fulfill them. In 2023, it struck a deal with Microsoft, which reportedly plans to spend $10 billion on its cloud computing services through 2030. That was enough for Blackstone to lead $2.3 billion in debt financing in 2023, followed by a fatter $7.5 billion check in 2024. “We are underwriting contracts, capacity and power, not a particular ‘story,’ ” says Jas Khaira, a senior managing director at Blackstone, adding that the due diligence process ensured the contracts were airtight and customers like Microsoft couldn’t easily walk away.
Inside a coolly lit data center in Orangeburg, New York, about 40 minutes north of Midtown Manhattan, rows of what look like hospital IV bags hang across long banks of densely packed black shelves. Each bag dispenses a drip of neon-green cooling fluid to a rack of GPU-filled servers when needed, keeping them at the safe, always-on temperature at which they help pay for CoreWeave’s considerable investment in them.
Just how much revenue those servers can deliver depends on two things: how long AI’s demand for compute will outstrip supply and how long it will take AI companies to catch up. Microsoft, for instance, is projected to spend some $80 billion on its own data centers in 2025. Open AI reportedly signed a $300 billion deal with Oracle recently to provide compute—part of its President Trump–endorsed Stargate partnership—calling into question the future of its contracts with CoreWeave. Aside from the hyperscaling incumbents, CoreWeave must also battle cloud computing rivals like Crusoe, Nebius and Lambda Labs, many of which have landed contracts with CoreWeave’s big tech clients, who use multiple cloud providers to reduce dependence on just one. “Renewal and retention is a risk,” says Macquarie analyst Paul Golding.
“We think about risk in a more sophisticated way than almost anyone else in the space.”
And then there’s that debt. CoreWeave has no plans to stop borrowing against its GPUs—debt could rise to $30 billion in two years, per JPMorgan projections. Some analysts are worried about the risk of doing that, especially amid ongoing net losses ($2.1 billion in total since 2023) in a fast-moving industry.
Intrator has heard and dismissed all this before. “No shit, I’m borrowing money,” he says, gesticulating defiantly at his Wall Street haters while en route to CoreWeave’s Manhattan office. McBee backs him up. “There’s no risk,” he says, improbably.
They claim each CoreWeave contract (average length: four years) prices in most of the costs of delivering on it—GPUs, data center buildouts and interest payments—plus depreciation and maintenance. If all goes as planned, then, CoreWeave is left with some profit plus the value of its GPUs. Generally, contracts are structured to generate approximately $2 of revenue for every dollar of debt to be repaid, McBee says. The risk, of course, is that the AI boom goes bubble and pops—and those contracts are never fulfilled, leaving CoreWeave with a ton of idle servers and a mountain of debt.
Citi analyst Tyler Radke forecasts CoreWeave could be profitable as early as next year. It’s losing money because of interest payments and because infrastructure costs a lot to build upfront. “We think about risk in a more sophisticated way than almost anyone else in the space,” Intrator says.
Maybe. But many credit rating agencies categorize CoreWeave’s latest bonds as junk. Even if the boom times continue, things could go south fast if the company can’t build out what it has promised, as quickly as promised.
Then there are the known unknowns: regulatory hurdles, manufacturing mishaps, local opposition. Protests alone have already stopped or slowed $64 billion in data center projects, according to Data Center Watch. And CoreWeave is already facing opposition over a new facility in Lancaster, Pennsylvania, where it’s investing $6 billion. When it’s finished, it is expected to consume four times as much power as all the region’s homes, says Lancaster resident Emma Burgess. Residents of the community known for its Amish history recently packed city council meetings to voice their concerns: too little community input, too much electricity, too much pollution (both air and noise). McBee says there hasn’t been pushback at the site, but acknowledges “you’re gonna break things along the way.” Because CoreWeave’s AI data centers need to be on 24/7/365, they can’t be subject to the fluctuations that come with wind and solar power. So for the time being “it’s gas,” McBee says. “That’s what we prefer.”
In May, CoreWeave acquired AI software developer Weights & Biases for $1 billion, a deal it hopes will diversify its customer base. A planned $9 billion acquisition of data center owner Core Scientific—which could bring some 1.5 gigawatts of additional contracted power—would more than quadruple CoreWeave’s current active power, provided regulators and shareholders allow it. Next up: investing in AI startups with a combo of cash and GPU access.
Things are volatile. CoreWeave’s market cap has swung by some $70 billion since its rocky March IPO. Tariffs could further unsettle its bottom line. Its biggest customer just signed a huge deal with a rival fast encroaching on its turf. Delays, of both capital and equipment, could kill CoreWeave’s head start. None of this has fazed Intrator, who’s confident his GPU-backed debt will pay off and keep his company growing alongside rampant demand for AI compute. As the markets rise, so too will CoreWeave, buoyed by the prescient bets that made it a $50 billion behemoth in just six years. But it’s still beholden to the biggest gamble of all—that the AI bubble won’t burst.