Meta Commits $21B More to CoreWeave, Bringing Total AI Cloud Spend to $35B
The expanded deal runs through 2032, locks in early access to Nvidia's Vera Rubin chips, and makes Meta roughly 40% of CoreWeave's projected capacity for the rest of the decade.
Deal Overview
CoreWeave (NASDAQ: CRWV) announced on April 9 an expanded long-term agreement with Meta Platforms (NASDAQ: META) to provide approximately $21 billion in AI cloud capacity through December 2032. This builds on a prior $14.2 billion deal signed in late 2025, bringing Meta’s total committed spend with CoreWeave to roughly $35 billion. The infrastructure will span multiple data center locations and include some of the earliest deployments of Nvidia’s Vera Rubin platform. CoreWeave also announced plans to raise $3 billion in fresh debt.
Unlike earlier contracts focused on training large language models, this agreement is specifically optimized for inference: running AI models in real-time at scale. Meta shares rose roughly 3% on the news; CoreWeave gained about 5%. CoreWeave CEO Mike Intrator framed the deal simply: hyperscalers can build their own data centers, but they still need external capacity. “For some reason, all these people who can buy compute also feel the need to buy it from us.”
Why It Matters
The inference pivot: The AI infrastructure market is shifting from training (building models) to inference (running them). Meta’s $21 billion commitment is the clearest signal yet that inference compute will dwarf training spend over the next decade. As Meta deploys AI agents across Facebook, Instagram, and WhatsApp for tasks like real-time translation, content moderation, and autonomous shopping, the demand for low-latency, always-on compute is effectively unbounded.
CoreWeave’s concentration risk: Meta now represents roughly 40% of CoreWeave’s projected capacity through the end of the decade. That is extraordinary customer concentration for a company that IPO’d only weeks ago. If Meta’s AI strategy stumbles, or if it decides to bring more compute in-house, CoreWeave’s revenue base compresses dramatically. The flip side: $35 billion in contracted revenue provides a level of visibility that almost no infrastructure company has ever had at this stage.
The GPU landlord model works: CoreWeave’s entire thesis is that hyperscalers will always need more compute than they can build themselves. This deal, plus similar arrangements with Google, Microsoft, and OpenAI, proves the model. CoreWeave shares are up 28% in 2026.
- Customer concentration: ~40% of capacity committed to one client. Meta's internal buildout (including a $10B Texas data center) could reduce future external demand.
- Debt load: $3B in new debt on top of existing leverage. CoreWeave is a capital-intensive business that must keep building ahead of demand.
- $35B in contracted revenue: Unprecedented visibility for an infrastructure company. Revenue is locked in through 2032 with one of the world's largest technology companies.
- Vera Rubin early access: Being among the first to deploy Nvidia's next-generation platform gives CoreWeave a performance edge that competitors cannot match for 12 to 18 months.