Tech giants like Microsoft (MSFT 0.03%) and Meta have been splurging on artificial intelligence (AI) accelerators and building out massive AI data centers but also have been relying on start-up CoreWeave to meet soaring demand for AI computing capacity. CoreWeave has built a specialized cloud computing platform focused primarily on running AI workloads, and almost all its revenue comes from long-term contracts with major technology companies. At the end of 2024, the company was operating more than 250,000 GPUs within its data centers.

AI accelerator market-leader Nvidia (NVDA 1.69%) invested in CoreWeave back in 2023, and now, the AI cloud platform is planning to go public and will likely target a valuation of at least $35 billion. The company is expected to raise over $3 billion from its initial public offering (IPO), which will help fuel further expansion. Revenue exploded by more than 700% in 2024 to $1.9 billion, and CoreWeave plowed $8.7 billion into capital expenditures as it expanded its footprint to meet customer demand.

CoreWeave is profitable on an operating basis, although interest payments on its debt eats up all its operating profit. While operating profitability, predictable long-term contracts, and booming demand for AI computing capacity are positives for the company, investors considering buying CoreWeave stock once it goes public should tread carefully.

Extreme customer concentration

CoreWeave generated 77% of its revenue from just two customers in 2024, and one of them was Microsoft. The latter is spending big on its own AI data centers but also plans to ramp up its leasing activity. Microsoft CEO Satya Nadella recently noted in an interview that the company was planning to lease a lot of capacity in 2027 and 2028.

Even though Microsoft is committed to plenty of leasing in the future, relying on so few customers is a major risk.

A business model that may not last

CoreWeave's long-term contracts with customers are take-or-pay, which means that customers pay for a certain amount of reserved capacity regardless of whether they make use of that capacity. In addition, customers provide a sizable prepayment, which helps fund the infrastructure build-out necessary to support the contract.

Given that cloud giants and other tech companies are currently scrambling to obtain AI compute capacity, CoreWeave has enough leverage to impose this kind of arrangement. If the company is forced to move to a pay-as-you-go model without multiyear customer commitments, its revenue would become much less predictable.

What could cause such a shift? An oversupply of AI computing capacity.

Such a scenario may seem unlikely today, but there's a reason Microsoft is planning to bump up leasing activity in a few years. In that same interview with Nadella, the Microsoft CEO noted that he expects the rampant spending on expanding AI computing capacity across the industry to bring prices down. In a world with too much AI computing capacity, you want to be leasing, not owning.

In an oversupply scenario, CoreWeave's customers are unlikely to be willing to commit to long-term contracts at fixed rates. CoreWeave's business model is tailor-made for the AI boom, but the company is going to face a reckoning once demand no longer outpaces supply.

Be careful with CoreWeave -- and Nvidia, too

Nvidia's sales of AI accelerators remain strong, but cracks in the AI growth story are starting to appear. Reports that Microsoft has cancelled some AI data center leases have popped up.

While Microsoft hasn't pulled back on its $80 billion infrastructure spending plan for this year, it could be shifting future plans to include more leasing and less building. That's exactly what the company would do if it was growing more worried about oversupply wrecking the value of its AI investments.

If pricing on AI computing capacity slumps due to oversupply, pricing on Nvidia's AI accelerators would also come under pressure. CoreWeave wouldn't benefit from lower AI accelerator prices because it would already be dealing with too much capacity. The only winners would be companies like Microsoft, by leasing capacity and paying market rates.

CoreWeave grew at an incredible pace in 2024. It's possible the AI boom will continue for another year, another two years, another three years, etc. But eventually, the incredible sums of money being thrown into AI computing capacity will push supply above demand. CoreWeave's business model works great when there's a shortage, but it could fall apart if there's a glut.

Should you buy the CoreWeave IPO? All I can tell you is what I'm doing, and I'm a hard no.