Data Center Delays Trigger Sell-Off Of AI Cloud Firm CoreWeave

CoreWeave is facing a data center capacity crunch as the Nvidia-backed firm races to deliver on tens of billions of dollars of new artificial intelligence computing contracts for the world’s largest tech companies.

The AI cloud firm trimmed its annual revenue forecast following revelations that it won't have the data center capacity needed to fulfill a major customer contract scheduled to commence this year.

Speaking on the firm’s third-quarter earnings call Monday evening, CoreWeave’s leadership blamed the capacity shortfall on a third-party data center provider’s failure to deliver new inventory on the promised timeline following delays in the development process. 

The specific data center firm responsible for the delays hasn't been disclosed, nor has the identity of the Big Tech customer impacted by the capacity constraints. CoreWeave said the customer has agreed to restructure its contract so the delay won't decrease the cloud firm's overall revenue from the deal.

But that revenue is now expected to come in early 2026 when the delayed capacity is delivered, leading CoreWeave to slash its projected revenue for this year by more than $200M.

Those lower numbers, along with concerns about the firm’s continued ability to acquire needed data center capacity along the timelines customers demand, sent the company’s share price plummeting more than 16% Tuesday.

It fell another 3% Wednesday, putting it down nearly 40% over the last month. 

“Everybody is frustrated — the data center provider is frustrated, we’re frustrated, the client is frustrated,” CoreWeave CEO Mike Intrator told Bloomberg.

CoreWeave, which has chipmaking giant Nvidia among its largest shareholders, specializes in cloud computing for AI workloads, effectively offering its graphics processing units to other operators for lease. Branding itself as “the AI hyperscaler,” CoreWeave went public in March, and it is the largest player in the emerging neocloud sector.

The data center concerns marred what was otherwise a strong quarter for the New Jersey-based firm. 

CoreWeave’s revenue more than doubled in Q3, jumping 134% year-over-year to $1.4B. The firm added $25B in new contracts during the quarter, inking several deals with Big Tech customers that are increasingly looking to CoreWeave for their AI computing needs. Last quarter, CoreWeave secured a $14.2B deal with Meta Platforms and expanded a contract with ChatGPT-maker OpenAI by $6.5B. 

The company also made substantial progress toward diversifying its customer base, addressing a major red flag for investors. At the start of this year, Microsoft alone accounted for nearly 85% of CoreWeave’s contracted revenue. Now, the firm’s leadership said no single customer represents more than 35%.

CoreWeave is still losing money, reporting a net loss of $110M in Q3. But that loss was far less than the $359M it lost in the same quarter last year. 

CoreWeave remains in the red largely due to its aggressive pursuit of data center capacity. The company added 120 megawatts across eight new facilities to its data center footprint last quarter, bringing its total capacity to 590 MW. Its contracted power pipeline grew by more than 600 MW to 2.9 gigawatts, with more than a gigawatt expected to come online within the next 24 months, Intrator said on its earnings call Monday.

CoreWeave’s massive capacity push has been funded largely by private equity-backed debt, with firms like Blue Owl partnering on the development of new campuses. This rapid debt-fueled expansion has made CoreWeave a frequently mentioned name in the growing narrative of an AI bubble, with the entire neocloud segment viewed as particularly high-risk should demand slow for AI computing. 

In one sense, the disclosed capacity shortages could counter these bubble fears. The delay in a customer contract due to data center setbacks proves CoreWeave’s infrastructure spending is driven by existing demand rather than speculation.

But the data center delays also highlight what has long been considered an area of risk for CoreWeave: The firm owns almost none of its data center infrastructure, relying almost entirely on capacity leased from third-party providers. 

This lack of control over its own data center infrastructure was among the key risk factors identified by investors — and by CoreWeave itself — ahead of the firm’s initial public offering in March.

“The quarter revealed something that investors have feared for a while — operational risk,” Barclays analysts said, according to Reuters. “This is the first time for the young AI infrastructure industry that this has come up and will likely remind investors that these large scale AI data centers are not easy engineering projects.”

While tech behemoths like Amazon and Google self-develop a share of their own data center portfolios, there are questions as to whether CoreWeave has the expertise needed to execute these projects at scale. 

CoreWeave had hoped to address these concerns by acquiring crypto miner-turned-AI data center provider Core Scientific. CoreWeave was already contracted for 840 MW of capacity from Core Scientific, and the $9B, all-stock deal would have given the cloud provider ownership of 1.3 GW of capacity in Core Scientific’s U.S. data center portfolio.

But Core Scientific’s shareholders rejected that deal late last month. 

While CoreWeave’s leadership has repeatedly declined to identify the third-party provider behind the delayed data center inventory, multiple analysts and industry observers believe the provider is Core Scientific. 

Intrator said Monday that CoreWeave is still looking to make self-development a larger part of its infrastructure expansion strategy. Calling the delays a “stumble” that is “going to be cleaned up rapidly,” he said the company is learning lessons from these delays and applying them to future projects, such as a need to diversify its supply chain to avoid extending development timelines. 

“We’re not saying that we’re going to go self-build and not use third-party data center providers,” Intrator said. “What we are saying is that self-build is a component of the way that you go about de-risking delivery across the broader portfolio.”