CoreWeave’s public debut was not the explosive market event the company had hoped for. It priced at $40 on Thursday, below the $47 to $50 range announced, trimmed the number of shares offered, opened at $39 on Friday and closed at $40.
Even with that muted reception, the IPO mattered. CoreWeave raised $1.5 billion, reached a $14 billion market cap on Day 1, and became the largest AI-related listing to date, as well as the biggest U.S. tech IPO since 2021.
From a hedge fund setback to a GPU business
The company’s origin story does not begin with a conventional cloud computing plan. Brian Venturo, CoreWeave’s chief strategy officer, had worked as a portfolio manager at Hudson Ridge Asset Management, an energy industry hedge fund founded by CoreWeave co-founder and CEO Michael Intrator.
At Hudson Ridge, the team built a machine learning model to help choose investments in the data-heavy energy industry. They also met Brannin McBee, who ran the data firm they used and later became a CoreWeave co-founder.
After the U.S. moved into its fracking boom era, Hudson Ridge closed. Venturo described the result simply: the group had “a lot of time on our hands.” That downtime pushed the founders toward crypto, not first as speculators, but as people trying to understand how crypto was produced from the commodity side.
Their first experiments were small. Venturo said they began mining on a pool table in a Manhattan office. Then the setup expanded. One GPU became 10, 10 became 1,000, and the equipment moved from the pool table to a closet.
The Ethereum mining years
The operation eventually reached what Venturo called “the most cliché place possible”: his grandfather’s garage in New Jersey. Friends in finance wanted to participate, so the founders bought more equipment.
At its peak, CoreWeave’s earlier operation was not a side project. Venturo said, “We were the largest Ethereum miner in the world for like two and a half years.” He also said that, at one point, the company had 50,000 Nvidia consumer GPUs.
Those chips were not designed for the work CoreWeave was asking them to do. They were consumer GPUs meant for video games on personal computers, not for constant use in harsh warehouse conditions. Venturo described the environment as “a warehouse with no air-conditioning or no ventilation.”
That constraint forced the team to build systems around the hardware. The founders created automation and health-checking tools to keep lower-grade GPUs operating in difficult conditions. The technical discipline developed for crypto mining later became part of the company’s shift toward AI training infrastructure.
Open source became the bridge to AI
The founders knew they wanted to use their large GPU base for more than mining. AI training was one possible direction, but they needed to learn how the infrastructure worked in practice.
That led CoreWeave to EleutherAI, an open source group working on a large language model. CoreWeave offered GPU access in exchange for help understanding AI training and announced a partnership in 2022.
Venturo said the company expected to learn infrastructure basics. Instead, EleutherAI connected CoreWeave to a broader group of people building AI startups. That connection became a turning point for the business.
The goodwill from working with EleutherAI helped turn some of those startups into paying customers. Stability AI heard about CoreWeave through EleutherAI and became a customer. As demand grew, the founders needed more capital to build stronger infrastructure.
Venturo described a dinner with Magnetar investors where he argued for the future of AI. He said Magnetar wrote a $100 million check.
Microsoft, OpenAI and the customer shift
CoreWeave’s open source work also created a path to larger customers. OpenAI learned about the company through its work with the open source community. Microsoft then learned about CoreWeave through OpenAI.
Microsoft became CoreWeave’s biggest customer because it was OpenAI’s biggest investor and sole cloud provider at the time. That relationship has since changed. OpenAI recently signed a $12 billion deal of its own with CoreWeave, which moved Microsoft out of the top customer position.
CoreWeave now says it has 32 data centers and 250,000 GPUs, including Nvidia’s difficult-to-obtain Blackwell chips, which supports AI reasoning. That scale helps explain why the company became central to AI training infrastructure so quickly.
But scale also brings pressure. The company has $7.6 billion in debt, with much of it due to be repaid in two years, the Financial Times reports. Against $1.9 billion in revenue, even with $15 billion under contract, that debt is a major reason investors have been cautious.
Venturo’s answer is that CoreWeave structures each customer deal to cover the debt used to buy the GPUs required for that customer. In other words, the company frames its borrowing as tied directly to contracted demand for computing capacity.
Why the IPO still matters
CoreWeave did not get the larger raise or higher valuation it wanted. The company had hoped for a $3 billion+ raise and a much higher valuation, but ended its first day with a $1.5 billion raise and a $14 billion market cap.
Still, the listing gives public investors a direct way to evaluate a company built around AI training infrastructure. It also shows how quickly the AI market has pulled unusual businesses into the center of technology finance.
The story remains unusual because CoreWeave’s path was not linear. It moved from a closed hedge fund, to crypto mining, to managing huge numbers of consumer Nvidia GPUs, to serving AI startups, Microsoft and OpenAI.
Venturo summed up the company’s path with a simple admission: “There’s so many pieces of luck along the way, it’s crazy.” The IPO may have been lukewarm, but the business behind it reflects one of the clearest shifts in computing demand: GPUs that once mined Ethereum are now part of the race to train and run AI systems.