Meta inks $21B with CoreWeave

Meta agreed to expand a long‑term deal worth roughly $21 billion with CoreWeave to buy outsourced AI cloud capacity through 2032, anchoring large inference workloads on a specialist provider. (investors.coreweave.com) Reuters reports the pact reflects Meta’s push to scale after a model release that underperformed expectations and shows hyperscalers still buy third‑party capacity for speed and flexibility. (reuters.com)

Meta is still building its own data centers, but it just agreed to rent far more artificial intelligence computing from CoreWeave through December 2032 for about $21 billion. The surprise is not the size alone; it is that one of the world’s richest tech companies is still outsourcing a critical piece of its artificial intelligence workload. (investors.coreweave.com) CoreWeave is not a consumer app company or a social network. It is a specialist cloud provider that rents out graphics processing units, the chips used to train and run big artificial intelligence models, and it has built its business around supplying that capacity fast. (cnbc.com) This deal is about inference workloads, which means the computing needed after a model is built, when millions of people start asking it questions. Training is like opening a restaurant kitchen for the first time; inference is serving dinner every night without the line backing up out the door. (investors.coreweave.com) Meta already had a similar CoreWeave agreement worth about $14.2 billion from September 2025, so the relationship now totals roughly $35 billion. That tells you this was not a one-off emergency purchase but a decision to keep a specialist supplier close as demand grows. (bloomberg.com) (cnbc.com) CoreWeave said the new capacity will be spread across multiple sites and will include some of the first deployments of Nvidia’s Vera Rubin platform, the next generation of Nvidia systems for artificial intelligence computing. Meta is not just buying more servers here; it is reserving early access to newer hardware. (coreweave.com) That helps explain why Meta would rent instead of waiting to build everything itself. A third-party cloud provider can often deliver ready-to-use capacity faster than a company can finish a new data center, fill it with chips, connect the power, and bring it online. (reuters.com) (coreweave.com) Meta has room to spend because its 2026 capital expenditure forecast is $115 billion to $135 billion, up from $72.22 billion in 2025, and the company has said infrastructure costs include payments to third-party cloud providers. In other words, renting CoreWeave capacity is not separate from Meta’s buildout; it is one lane of the same spending surge. (usnews.com) (datacenterdynamics.com) The timing also matters because Reuters says Meta has been trying to catch up after an underwhelming artificial intelligence model release last year. When a company feels behind, buying outside capacity is one of the fastest ways to shorten the wait between a model idea and a product people can actually use. (reuters.com) For CoreWeave, this is the kind of contract that can change a company’s scale overnight. The company went public in March 2025, and now it has one customer relationship with Meta worth about $35 billion across two agreements. (cnbc.com 1) (cnbc.com 2) The bigger picture is that the largest cloud buyers are also becoming giant cloud renters when artificial intelligence demand spikes. Even companies that own fleets of data centers still pay specialists like CoreWeave for speed, flexibility, and early access to scarce chips when the race moves faster than concrete and power lines. (reuters.com) (investors.coreweave.com)

Get your own daily briefing

Scout delivers personalized news, insights, and conversations tailored to your role and industry.

Download on the App Store

Shared from Scout - Be the smartest in the room.