Meta's big CoreWeave deal
Meta signed a multibillion‑dollar deal with CoreWeave — reported as $21 billion of committed spend within a larger ~ $35 billion arrangement — to secure AI cloud infrastructure through 2032 and accelerate inference capacity. Social posts highlight the scale and long horizon of the agreement, reflecting how major cloud consumers are locking long‑term capacity for inference workloads. Long contracts like this can reshape competitive dynamics in the AI cloud market by prioritizing guaranteed capacity over spot availability. (x.com)
Meta just agreed to spend another $21 billion with CoreWeave, a company most people had never heard of two years ago, to rent artificial intelligence computing power from 2027 through 2032. That comes on top of a previous $14.2 billion arrangement, taking the relationship to about $35 billion. (cnbc.com) CoreWeave is not a chatbot company. It is a landlord for Nvidia graphics processing units, packing data centers with the chips that train and run artificial intelligence models for customers like Meta, Microsoft, Google, and OpenAI. (cnbc.com) This deal is about inference, which is the expensive step after a model is built. Training is like teaching a chef one recipe; inference is every meal that chef has to cook after the restaurant opens. (thenextweb.com) Meta is building its own giant facilities at the same time. In January, the company said 2026 capital spending would jump to $115 billion to $135 billion, after spending $72.22 billion in 2025. (investor.atmeta.com) So why rent from CoreWeave if you are already pouring more than $100 billion into your own infrastructure. CoreWeave chief executive Michael Intrator gave the blunt answer to CNBC: the risk of not having enough compute is too high, so even companies that can build still want guaranteed outside capacity. (cnbc.com) That is what makes the timeline so important. The new commitment starts in 2027 and runs to 2032, which means Meta is not buying a temporary burst of chips for one launch but reserving years of supply the way an airline locks in gates at a crowded airport. (cnbc.com) For CoreWeave, long contracts like this turn a fast-growing renter of chips into something closer to a utility with booked demand. The company said on February 26 that its revenue backlog had reached $66.8 billion, while full-year 2025 revenue was $5.13 billion. (investors.coreweave.com) CoreWeave still has concentration risk, which is the danger of leaning too hard on a small number of giant customers. In its 2025 annual report, the company said Microsoft alone accounted for about 67% of revenue in 2025. (sec.gov) Meta’s contract changes that customer mix and gives CoreWeave a second anchor tenant with a very long lease. It also gives Meta a way to add capacity faster than waiting for every new in-house data center to be built, powered, and filled with chips. (cnbc.com) The larger shift is that artificial intelligence cloud capacity is starting to look less like on-demand computing and more like pipeline space. When the biggest buyers lock up years of supply in advance, smaller customers are left competing for what is not already spoken for. (thenextweb.com) That is why a contract number matters here more than a product launch. A $21 billion commitment through 2032 says Meta thinks the bottleneck in artificial intelligence is no longer ideas on a whiteboard but access to enough machines to answer billions of prompts after the models go live. (cnbc.com)
