Crypto-Mining GPUs in Closet Spark $1.5B IPO: CoreWeave Co-Founder Reveals Journey

CoreWeave's stock market debut on Friday was more of a gentle nod than a triumphant shout. Priced at $40 on Thursday, the shares fell short of the anticipated $47 to $50 range. The company also scaled back the number of shares it offered.

In total, CoreWeave managed to raise $1.5 billion, securing a $14 billion market cap on its first day. This was less than the hoped-for $3 billion+ and a higher valuation they had aimed for. Shares started trading at $39 (yikes!) and ended the day at $40. The reception was, well, lukewarm at best.

Despite this, CoreWeave's IPO stands out as the largest AI-related listing so far and the biggest U.S. tech IPO since the bustling days of 2021.

In a plain white hoodie, sitting in an unremarkable conference room, and speaking with a noticeable Jersey accent, chief strategy officer Brian Venturo shared with TechCrunch that he feels incredibly fortunate.

It all started when he and his hedge fund buddies found themselves with time on their hands after their previous venture fizzled out.

Venturo had been a portfolio manager at Hudson Ridge Asset Management, an energy industry hedge fund founded by CoreWeave co-founder and CEO Michael Intrator. They developed a machine learning model to guide their investment choices in the data-heavy energy sector. There, they crossed paths with their other co-founder, Brannin McBee, who was running the data firm they utilized.

But as the U.S. entered its fracking boom, they shut down Hudson Ridge, leaving them, as Venturo puts it, "with a lot of time on our hands."

Their next venture? Crypto. They were keen to dive in but first wanted to grasp the commodity aspect, so they kicked off mining operations right on the pool table in their Manhattan office.

Thousands of GPUs in a warehouse

It's like the old saying about potato chips—one GPU led to ten, and ten led to a thousand. The rigs migrated from the pool table to a closet.

"Before we knew it, we were in the most cliché spot imaginable—my grandfather’s garage in New Jersey," Venturo chuckled. As their finance buddies got interested, they ramped up their operations.

"We were the world's largest Ethereum miner for about two and a half years," he said. "At one point, we had 50,000 Nvidia consumer GPUs."

These were chips designed for gaming on consumer PCs, not for running 24/7 in "a warehouse with no air-conditioning or ventilation," he explained. So, the co-founders engineered "crazy automation and health-checking systems" to manage these low-grade GPUs in tough conditions.

The team knew they wanted to repurpose their GPU empire for other ventures, like AI training. But they needed to learn how.

They reached out to EleutherAI, an open-source group developing a large language model. CoreWeave offered GPU access in exchange for insights into AI training, and they formalized a partnership in 2022.

"We thought we were just going to learn how the infrastructure worked," Venturo said. But EleutherAI connected them with hundreds of AI startup builders, sparking a major turning point for CoreWeave.

The goodwill from working with EleutherAI turned these startups into paying customers. It was "total luck that started the training business," Venturo noted.

Stability AI caught wind of CoreWeave through EleutherAI and became a client. The founders needed more capital to enhance their infrastructure.

Over a dinner with Magnetar investors, Venturo "was literally pounding on the dinner table," convincing them of AI's future. He said Magnetar ended up writing them a $100 million check.

Open source paves the way

OpenAI discovered CoreWeave through its open-source community involvement. Microsoft learned about the company through OpenAI, and subsequently became its biggest customer, being OpenAI’s largest investor and sole cloud provider at the time.

That's changed now. OpenAI recently inked a $12 billion deal with CoreWeave, knocking Microsoft off its perch as the top customer.

Today, CoreWeave boasts 32 data centers and 250,000 GPUs, including Nvidia’s hard-to-get Blackwell chips, which support AI reasoning, according to the company.

Venturo admits that CoreWeave's staggering $7.6 billion in debt, much of which is due within two years, has been a hot topic, as reported by the Financial Times. Against CoreWeave’s $1.9 billion in revenue (and $15 billion under contract), this debt is a major reason for investor caution.

However, Venturo insists that CoreWeave has structured each customer deal to cover the debt used to purchase the necessary GPUs. More than that, he recognizes that three hedge fund guys turned crypto miners, now running a pivotal AI training infrastructure, have been on a rollercoaster journey.

"There’s so many pieces of luck along the way, it’s crazy," he said.