Meta’s $21B CoreWeave deal
Meta signed a massive $21 billion cloud-computing agreement with CoreWeave to deepen its AI training and inference capacity, signalling continued heavy demand for scalable GPU-backed cloud services. The deal underscores how hyperscalers are locking in external capacity as they race to deploy larger models and push more inference to production. (reuters.com)
Meta just agreed to spend another $21 billion with CoreWeave, a cloud company that rents out artificial intelligence computing power, and this new contract runs through December 2032. The added commitment sits on top of an earlier $14.2 billion arrangement, taking the two companies’ total disclosed business to about $35 billion. (cnbc.com) CoreWeave is not a household-name cloud giant like Amazon Web Services or Microsoft Azure. It is a specialist that packs data centers with Nvidia graphics processing units, the chips used to train and run large artificial intelligence models. (cnbc.com) This deal is not mainly about building one model in a lab. CoreWeave said Meta will use the capacity to scale inference workloads, which means serving answers, recommendations, and generated content after a model is already trained. (coreweave.com) Inference is the expensive part people notice less because it happens after the headline demo. Every time a chatbot replies, an image tool renders, or a recommendation engine re-ranks a feed, those graphics processing units are doing live work in the background. (coreweave.com) Meta is spending like a company that thinks computing power is now a core product input, not a support function. In its January 2026 results, Meta said it expects 2026 capital expenditures of $115 billion to $135 billion, up from $72.22 billion in 2025. (atmeta.com) That helps explain why Meta is doing two things at once. It is building its own giant facilities, including a $10 billion Texas data center reported by CNBC, while also paying outside suppliers like CoreWeave for extra capacity from 2027 through 2032. (cnbc.com) CoreWeave’s side of the story is just as revealing. The company said the Meta capacity will be spread across multiple locations and will include some of the first deployments of Nvidia’s Vera Rubin platform, which is the next generation of artificial intelligence system Nvidia has been pitching to big customers. (coreweave.com) A contract that large changes the shape of CoreWeave itself. CoreWeave reported $5.13 billion in 2025 revenue, so a fresh $21 billion agreement is several years of last year’s sales booked into one customer relationship. (finance.yahoo.com) That concentration cuts both ways. CoreWeave’s backlog had already swollen to $66.8 billion by the end of 2025, and third-party analyses of its filings said Microsoft accounted for about 67% of 2025 revenue, so landing Meta at this scale both deepens demand and starts to diversify who writes the biggest checks. (sacra.com) The simplest way to read this is that even companies rich enough to build their own server campuses still do not want to rely only on their own construction schedules. Meta’s new contract says the bottleneck in artificial intelligence is still access to working graphics processing units in real buildings with power, cooling, and network links already in place. (cnbc.com)
