Meta commits an extra $21 billion to CoreWeave as AI compute demand soars

Big new deal — on top of the last one

It has been reported that Meta will spend an additional $21 billion with CoreWeave on AI cloud infrastructure, covering 2027 through 2032. That commitment comes on top of a prior $14.2 billion arrangement that runs through 2031. The scale is staggering. Why buy so much from an outside vendor when you’re already building your own data centers? Because supply, speed and specialization matter — and Meta wants both.

Why CoreWeave — and why now

CoreWeave runs data centers packed with hundreds of thousands of Nvidia GPUs, the kind of kit hyperscalers need to train and run big models. The company also serves Google, Microsoft, OpenAI and others, and CEO Mike Intrator told CNBC that customers keep coming back “because of the quality of the product that we deliver.” Meta recently unveiled a new model, Muse Spark, and has said it will boost capex dramatically — between $115 billion and $135 billion this year — so external capacity looks like insurance against the bottlenecks of a gold-rush moment.

What it means for CoreWeave and the market

The deal helps CoreWeave diversify away from Microsoft, which accounted for 62% of its 2024 revenue; after the new business, no single customer will represent more than 35%, Intrator said. The company finished 2025 carrying about $21 billion of debt and borrowed another $8.5 billion in March to fund infrastructure tied to new contracts — a heavy load, but investors have rewarded the story: CoreWeave’s stock is up about 24% this year while the broader market has lagged. Is this more proof the AI arms race has no off switch? Maybe so — and both suppliers and buyers are leaning in hard.

Sources: cnbc.com