Feb 26 (Reuters) - CoreWeave expects capital expenditure to double this year, it said on Thursday, as it spends heavily to scale up its AI cloud platform to handle the massive computing power its customers demand for training and deploying advanced AI models.
Its shares slumped more than 8% in after-hours trading. They are down 8% so far this year.
The cloud infrastructure technology company expects capital expenditure between $30 billion and $35 billion in 2026, up from $14.9 billion in 2025, driven by purchases of Nvidia's AI chips, rapid data center buildouts and energy procurement for powering them.
"We made the decision to go ahead and to build faster so that we can deliver more infrastructure," CEO Michael Intrator told Reuters in an interview.
"It puts some short-term pressure on the margins," he said, but added that the build-out was helping the company secure stable contracts.
"Q1 is going to be the low point, and then it's going to build from there," he said, referring to margins. CoreWeave's adjusted operating income margin fell to 6% in the December quarter from 16% a year ago.
All of the company's substantial capital expenses were tied to already signed customer contracts, CFO Nitin Agrawal said in the company's earnings call.
Pitted against Big Tech and their deep pockets - these companies are expected to spend over $600 billion this year - CoreWeave has to work quickly to keep up with surging demand.
The company's revenue backlog surged to $66.8 billion as of December 31, from $15.1 billion a year earlier, driven by long-term cloud‑compute agreements.However, the revenue is dependent on CoreWeave meeting delivery obligations and its data centers coming online on time.
CoreWeave finished 2025 with more than 850 megawatts of active power across 43 data centers, and 3.1 gigawatts of contracted capacity, most of which it expects online by 2027.
"Right now, CoreWeave is being punished for either having too little capex or too much capex," D.A. Davidson analyst Alexander Platt said. "The fact they're not seeing issues bringing capacity online - hence the high capex - is a positive signal."
DIVERSIFYING BACKLOG
CoreWeave is a so-called neo-cloud that provides tech companies with the hardware and cloud capacity needed for AI technologies, and counts hyperscalers such as Microsoft and Alphabet's Google among its rivals.
But unlike these big tech firms, CoreWeave offers its clients exclusive access to the entirety of Nvidia's advanced chips, called GPU clusters, without having to share capacity.
This specialized and cost-effective alternative has helped it attract clients ranging from AI labs to large enterprises.
"We've continued to diversify our backlog in the second half of 2025," Intrator told Reuters, noting that 70% of the backlog was from financially strong, low-risk customers.
The companysaid adjusted net loss ballooned to $284 million in the fourth quarter ended December 31 from $36 million a year ago.
CoreWeave is also working to lower its financing costs as it scales, with management highlighting its improving access to cheaper capital. As of September last year, its debt burden stood at around $14 billion.
"We expect to continue to reduce our weighted average cost of capital along the way," Intrator said on a call with analysts.
The company reported a revenue of $1.57 billion for the fourth quarter. It projected first-quarter revenue between $1.9 billion and $2.0 billion, below estimates of $2.29 billion, according to data compiled by LSEG.
(Reporting by Anhata Rooprai in Bengaluru and Krystal Hu in San Francisco, additional reporting by Juby Babu in Mexico City; Editing by Sayantani Ghosh and Janane Venkatraman)
