Importance Score: 72 / 100 🔴
CoreWeave, a prominent artificial intelligence (AI) infrastructure provider specializing in GPU cloud computing, commenced trading on Friday with a subdued debut, rather than significant fanfare. The company priced its initial public offering (IPO) at $40 per share on Thursday, falling short of the anticipated $47-$50 price range. Furthermore, the number of shares offered was also reduced, indicating a recalibration of expectations for the AI IPO market.
In total, CoreWeave’s IPO generated $1.5 billion and established a market capitalization of $14 billion on its first day of trading. This figure is less than the initially hoped-for capital raise exceeding $3 billion and a correspondingly higher valuation. Shares experienced an initial dip, opening at $39, before closing at the IPO price of $40, signaling a tepid market response.
Despite the lukewarm reception, CoreWeave’s IPO stands as the largest listing in the AI sector to date and represents the most substantial U.S. technology IPO since the boom period of 2021.
Brian Venturo, Chief Strategy Officer, in casual attire and speaking with a New Jersey accent, conveyed a sense of gratitude during an interview, acknowledging the company’s journey to this point.
He explained that the company’s origins stemmed from a period of downtime experienced by him and his associates from the hedge fund industry after their previous venture concluded.
Venturo had served as a portfolio manager at Hudson Ridge, an energy sector hedge fund founded by CoreWeave co-founder and CEO Michael Intrator. At Hudson Ridge, they developed a machine learning (ML) model for investment decisions within the data-intensive energy industry. It was there they collaborated with Brannin McBee, their co-founder, who managed their data firm.
However, the emergence of the fracking boom in the U.S. led to the closure of Hudson Ridge, leaving the team with considerable free time, according to Venturo.
Their next endeavor was cryptocurrency. They aimed to enter the market but first sought to understand its fundamental production, as Venturo described, “from the commodity side, how is this made?”. This curiosity led them to initiate crypto mining operations within their Manhattan office.
Scaling Up: From Pool Table to Warehouse with GPUs
Starting modestly, their GPU collection expanded rapidly. A single GPU quickly multiplied to ten, and then to a thousand. Their mining equipment relocated from a pool table to a closet.
“Before we knew it, we were in a very stereotypical situation, operating out of my grandfather’s garage in New Jersey,” he recounted humorously. Interest from finance colleagues spurred further expansion and equipment acquisition.
“We were the largest Ethereum miner globally for approximately two and a half years,” Venturo stated. “At one point, we possessed 50,000 Nvidia consumer GPUs.”
These graphics processing units, designed for consumer gaming PCs, were being utilized continuously in demanding environments. “A warehouse with no air conditioning or no ventilation,” as Venturo described. Consequently, the co-founders developed sophisticated automation and monitoring systems to maintain these consumer-grade GPUs under harsh operational conditions.
The team recognized the potential of their extensive GPU infrastructure beyond cryptocurrency mining, particularly for applications like AI training. However, they needed to acquire expertise in this domain.
This led to a collaboration with EleutherAI, an open-source collective focused on large language models (LLMs). CoreWeave offered access to its GPUs in exchange for knowledge sharing on AI training methodologies, announcing a formal partnership in 2022.
“Initially, we anticipated simply learning about the infrastructure mechanics,” Venturo explained. However, EleutherAI’s network of hundreds of AI startup developers provided an unexpected catalyst for CoreWeave’s business evolution.
The goodwill generated from the EleutherAI collaboration translated into paying customers from these burgeoning startups. According to Venturo, entering the AI training business was “total luck.”
Stability AI, another prominent player in the AI landscape, became aware of CoreWeave through EleutherAI and subsequently became a customer. This growth necessitated further capital investment to enhance their infrastructure.
Meetings with Magnetar investors followed, where Venturo passionately advocated for the future of AI. Magnetar subsequently invested a significant sum, which Venturo recalled as a $100 million investment.
Open Source Collaboration Fuels Growth
OpenAI also became acquainted with CoreWeave through its engagement with the open-source community. Microsoft, in turn, learned about CoreWeave via OpenAI. Microsoft became a primary customer, being a major investor and the exclusive cloud provider for OpenAI at the time.
This dynamic has since shifted. OpenAI recently finalized a substantial $12 billion agreement with CoreWeave, surpassing Microsoft as its largest client.
Currently, CoreWeave operates 32 data centers and manages 250,000 GPUs, including Nvidia’s advanced Blackwell chips, which are crucial for AI reasoning applications, as per the company.
Venturo addressed concerns regarding CoreWeave’s reported $7.6 billion debt, a significant portion of which is due within two years, as highlighted by the Financial Times. In relation to CoreWeave’s $1.9 billion in revenue (and $15 billion in contracted revenue, according to the company), this debt has contributed to investor caution.
However, Venturo asserted that CoreWeave has structured each customer agreement to adequately cover the debt incurred for purchasing the necessary GPUs. Reflecting on their trajectory, Venturo acknowledged the improbable journey from hedge fund associates turned cryptocurrency miners to leaders in AI training infrastructure.
“There’s been an incredible amount of luck involved along the way, it’s truly astonishing,” he concluded.