NVIDIA’s next-gen superchips anchor a $21B AI partnership between Meta and CoreWeave.

Meta has entered a roughly $21 billion agreement with CoreWeave, deepening its commitment to AI infrastructure as demand for scalable compute and inference capacity accelerates.

The deal expands on an earlier partnership valued at up to $14.2 billion and ensures CoreWeave will provide dedicated cloud resources to Meta through 2032, supporting the company’s growing AI ecosystem.

The rollout will span multiple data center regions and include early deployment of NVIDIA’s Vera Rubin platform, a next-generation AI supercomputing architecture designed to deliver higher efficiency, resilience, and scale.

Vera Rubin is expected to help drive the transition from conventional generative AI to more advanced “agentic” systems capable of autonomously executing complex workflows.

A major emphasis of the agreement is inference capacity, underscoring the industry’s shift toward continuously operating AI models in production environments rather than focusing primarily on training.

According to CoreWeave CEO Michael Intrator, the partnership reflects increasing reliance by major technology firms on specialized AI cloud infrastructure to support high-intensity workloads.

The agreement highlights a broader structural trend, as large companies move to secure long-term compute capacity to underpin AI-driven products and services.

CoreWeave shares rose करीब 3% in pre-market trading following the announcement.