Meta Locks In $21B CoreWeave AI Cloud Through 2032, Anchors Vera Rubin Rollout
CoreWeave and Meta have signed a $21 billion expanded AI cloud agreement, extending their existing relationship through December 2032. The deal represents one of the largest committed AI infrastructure contracts ever disclosed and significantly deepens Meta’s bet on third-party compute over owned data centres.
The capacity spans multiple distributed locations — an architecture designed for performance, resilience, and geographic redundancy across Meta’s AI workloads. The agreement includes some of the initial deployments of NVIDIA’s Vera Rubin platform, the next-generation successor to Blackwell, putting CoreWeave at the front of the Vera Rubin rollout queue.
Why This Changes the Math
Meta has historically been aggressive about owning its own infrastructure. Its MTIA 400 chip deployment, announced this week, targets GenAI inference in-house. The CoreWeave commitment suggests that owned silicon is not sufficient — the scale of Meta’s AI ambitions requires overflow from hyperscale clouds and AI-native providers simultaneously.
At $21 billion over approximately 6.5 years (2026-2032), the implied annualised spend is roughly $3.2 billion per year with CoreWeave alone. That sits alongside Meta’s existing $60B+ annual capex guidance for 2026 and its growing Llama compute footprint.
CoreWeave Context
CoreWeave went public on Nasdaq (CRWV) in March 2026. This deal was disclosed via an SEC filing, not a press release — meaning the $21B figure is a binding contractual commitment, not a soft framework. CoreWeave CEO Michael Intrator described it as evidence that “leading companies are choosing CoreWeave’s AI cloud to run their most demanding workloads.”
The timing matters: CoreWeave closed an $8.5 billion investment-grade GPU loan at A3 credit rating just weeks earlier, using long-term customer commitments like this one as collateral. Meta is effectively helping underwrite CoreWeave’s debt structure.
Vera Rubin is the Strategic Wildcard
NVIDIA’s Vera Rubin succeeds Blackwell with a unified CPU-GPU architecture (Vera CPU + Rubin GPU). Early access to Vera Rubin clusters — before most hyperscalers have deployed them at scale — gives Meta a compute advantage window for frontier training runs in 2027-2028. CoreWeave’s ability to move faster than AWS, Azure, or GCP on new NVIDIA silicon is a key part of its value proposition here.