As organizations rush to embrace AI, it’s putting the cloud — or clouds, rather — under strain.
Amazon Web Services, Microsoft, Google and Oracle, facing an unprecedented spike in demand for the server chips that train and run AI-powered software, are limiting their availability for customers. Microsoft has been particularly candid about its struggles, warning of service disruptions if it can’t get enough AI chips for its data centers.
Startups are feeling the pressure, too — including CoreWeave, a GPU-focused cloud compute provider. After raising $221 million a Series B round in April and a $200 million extension to that round in May, CoreWeave has secured $2.3 billion in debt financing, the company says.
The credit facility, which was led by existing investors Blackstone and Magnetar Capital with participation from Coatue, DigitalBridge Credit and funds and accounts managed by Pimco and Carlyle, comes only weeks after CoreWeave’s announcement that it plans to build a $1.6 billion data center in Plano, Texas. In a blog post provided to TechCrunch, co-founder and CEO Michael Intrator says that the debt facility will provide “financial headroom and flexibility” to meet CoreWeave’s goals of reaching 14 data centers by the end of the year.
“[We’ll commit the loan] entirely toward purchasing and paying for hardware for contracts already executed with clients and continuing to hire the best talent in the industry,” Intrator wrote. “No one was expecting this level of demand for GPU compute, but our strategic investments to increase capacity continue to pay off — and we’re delivering where others cannot.”
CoreWeave was founded in 2017 by Intrator, Brian Venturo and Brannin McBee to address what they saw as “a void” in the cloud market. Venturo, a hobbyist Ethereum miner, cheaply acquired GPUs from insolvent cryptocurrency mining farms, choosing Nvidia hardware for the increased memory (Nvidia is an investor, unsurprisingly).
Initially, CoreWeave was focused exclusively on cryptocurrency applications. But it pivoted within the last several years to general-purpose computing as well as generative AI technologies, like text-generating AI models. CoreWeave’s GPU might was conducive to this, as GPUs’ ability to perform many computations in parallel make them well suited to training today’s most capable models. (Nvidia’s benefitted massively, briefly becoming a $1 trillion company.)
Fast-forward to today and CoreWeave provides access to over a dozen SKUs of Nvidia GPUs in the cloud, including H100s, A100s, A40s and RTX A6000s, for use cases like AI and machine learning, visual effects and rendering, batch processing and pixel streaming. CoreWeave applies its infrastructure to special projects, as well, like an “AI supercomputer” of over 3,500 H100s it unveiled in partnership with Nvidia last month.
It’s tough for any cloud provider to compete with the incumbents in the space — i.e., Google, Amazon and Microsoft. For perspective, AWS made $80.1 billion in revenue last year, while Google Cloud and Azure made $75.3 billion and $26.28 billion, respectively.
Those figures are multiples above CoreWeave’s valuation (~$2 billion), obviously, let alone its war chest ($576.5 million).
To drive the point home, according to a Statista report from the fourth quarter of 2022, AWS had a 32% market share, Azure had a 23% share and Google Cloud had a 10% share.
Making matters more precarious, as my colleague Ron Miller recently wrote, many companies are looking for ways to cut back on spending in an uncertain economy. In 2023, the market for cloud infrastructure slowed to 21% growth, a precipitous drop from the 36% growth in the year prior.
That’s led to consolidation in the space. In July, DigitalOcean, the cloud hosting business, acquired Paperspace, a New York-based cloud computing and AI development startup, for $111 million in cash.
But that’s not to say it’s impossible for a smaller player to succeed — see Scaleway, Clever Cloud and Vultr. And CoreWeave’s wisely redoubled its efforts to build infrastructure supporting a red-hot sector: generative AI. According to Brian Venturo, CoreWeave’s CTO, the company’s newer data centers host as many as ~20,000 GPUs in one location — well above what cloud providers have traditionally offered.
“The soaring computing demand from generative AI will require significant investment in specialized GPU cloud infrastructure — where CoreWeave is a clear leader in powering innovation,” Jasvinder Khaira, a Blackstone senior managing director, said in an emailed statement.
While tech giants invest in in-house supercomputers and AI chips to train their generative AI models, smaller outfits are turning to cloud providers like CoreWeave. And they have a lot of cash to burn. According to PitchBook data, about $1.7 billion was generated across 46 deals for generative AI startups in Q1 2023, with an additional $10.68 billion worth of deals announced in the quarter but not yet completed.
Inflection AI, an AI startup helmed by DeepMind co-founder Mustafa Suleyman, is one of those outfits. Inflection trained its AI assistant product, Pi, on CoreWeave’s infrastructure. And it’s working with Nvidia and CoreWeave to build an AI training cluster with 22,000 H100 GPUs.
Beyond infrastructure, CoreWeave attempts to differentiate itself with offerings like its accelerator program, which launched in late October. The accelerator — which operates on an open-ended basis, with no deadlines — provides companies compute credits in addition to discounts and other hardware resources on the CoreWeave cloud.
CoreWeave says it employs “just over” 115 people now — up 150% in the last year or so — thanks in part to its acquisition of cloud rendering platform Conductor Technologies in January. Intrator says that the plan is to keep hiring “throughout the year,” bolstered by the debt facility.
