CoreWeave is plummeting after spending $35 billion, despite disappointing sales

Shares of CoreWeave fell more than 8 percent in after-hours trading on Thursday after the Nvidia-backed AI cloud infrastructure company unveiled plans to nearly double capital spending this year even as it posted a modest revenue gain in the fourth quarter.

The stock is down about 8 percent so far this year as investors weigh aggressive expansion plans against mounting losses and supply risks associated with its huge backlog.

Capex will reach $35 billion

CoreWeave said it expects capital expenditures of between $30 billion and $35 billion in 2026, sharply higher than the $14.9 billion it spent in 2025. The increase will be largely driven by the purchase of advanced AI chips from Nvidia, the rapid expansion of data centers and the energy purchases required to power them.

CEO Michael Intrator told Reuters the company had chosen to accelerate its buildout to meet rising customer demand for AI computing.

“We have made the decision to move forward and build faster so we can deliver more infrastructure,” Intrator said, acknowledging that pressure on margins would weigh on near-term margins.

Adjusted operating profit margin fell to 6 percent in the December quarter, compared to 16 percent a year earlier. Intrator describes the first quarter as likely to mark the bottom for margins before a recovery later in the year.

Chief Financial Officer Nitin Agrawal said all planned capital expenditures were tied to customer contracts already signed, providing some insight into future revenue streams.

Losses are increasing, despite the sales setback

CoreWeave reported fourth-quarter revenue of $1.57 billion, slightly ahead of expectations. However, its adjusted net loss rose to $284 million in the quarter ended December 31, compared with a loss of $36 million a year earlier, due to heavy initial infrastructure investments.

For the first quarter, the company expects revenue between $1.9 billion and $2.0 billion — lower than analyst estimates of $2.29 billion, according to LSEG data — adding to investor concerns about near-term growth momentum.

The $66.8 billion backlog is under scrutiny

The company’s revenue backlog rose to $66.8 billion as of Dec. 31 from $15.1 billion a year earlier, fueled by long-term cloud computing deals. Management says 70 percent of its order book now comes from financially strong, low-risk customers as the company works to diversify its customer base.

Still, the backlog is dependent on CoreWeave bringing data center capacity online on time and meeting delivery commitments – a complex task amid soaring global demand for AI infrastructure.

At the end of 2025, CoreWeave had more than 850 megawatts of active power capacity across 43 data centers, of which 3.1 gigawatts were contracted – the majority of which is expected to be operational in 2027.

“Right now, CoreWeave is being punished for either having too little or too much capex,” said Alexander Platt, analyst at DA Davidson. “The fact that they see no problems in bringing capacity online – hence the high capital investments – is a positive signal.”

Competing with great technical firepower

CoreWeave operates as a so-called ‘neo-cloud’, offering AI labs and enterprises access to dedicated clusters of Nvidia’s advanced graphics processing units (GPUs), without having to share capacity as on hyperscale platforms.

It competes with deep-pocketed technology giants such as Microsoft and Alphabet, whose cloud arm Google is a major player in AI infrastructure. Collectively, major tech companies are expected to spend at least $630 billion on AI-related infrastructure this year, increasing competition.

Unlike its larger rivals, CoreWeave offers customers exclusive access to full GPU clusters, a proposition that resonates with AI startups and large enterprises looking for predictable performance for model training and deployment.

Debt and capital costs in focus

As of September last year, CoreWeave’s debt was approximately $14 billion. Management said it is working to lower financing costs as it becomes more scalable, citing better access to cheaper capital.

“We expect to continue to reduce our weighted average cost of capital over time,” Intrator said.

With input from authorities.