Meta and CoreWeave Seal $21 Billion Deal for AI Inference

American cloud provider CoreWeave and tech giant Meta Platforms have concluded an expanded, long-term agreement for the provision of AI computing capacity. The contract is valued at approximately 21 billion US dollars and runs through December 2032. The deal is considered one of the largest agreements in the field of AI infrastructure and underscores the growing demand for specialized computing power in the industry.

What Was Agreed

CoreWeave, which has been trading on the Nasdaq under the ticker CRWV since its IPO, will provide Meta with dedicated cloud capacity across multiple locations. The agreement includes, among other things, initial deployments of the new NVIDIA platform “Vera Rubin,” one of the most powerful next-generation GPU architectures. The distributed approach across multiple data centers is designed to optimize performance, resilience, and scalability for Meta’s AI operations.

The agreement builds on and significantly expands an already existing business relationship between the two companies. Further details were disclosed in a mandatory filing with the US Securities and Exchange Commission (SEC).

The Shift from Training to Inference

Particularly noteworthy is the strategic focus of the deal. The primary emphasis is no longer on training new AI models, but on operating them at scale — what is known as inference. This involves using already-trained models to answer queries in real time, generate content, or make decisions.

This shift reflects an industry-wide development. While massive investments flowed into the training of large language models in recent years, the focus is increasingly shifting toward the productive use of these models in real-world applications. Inference workloads place their own demands on infrastructure: they require low latency, high availability, and the ability to process millions of requests simultaneously.

“This is another example of leading companies choosing CoreWeave’s AI cloud to run their most demanding workloads.” (Michael Intrator, Co-Founder and CEO of CoreWeave)

Why Inference Is Becoming the New Key Discipline

The transition from training to inference has several causes. First, the major foundation models are now largely developed and are being integrated into products and services. Second, the number of users of AI-powered applications is growing rapidly, increasing the demand for scalable computing power for ongoing operations. Third, the cost of inference at a growing user base is a decisive competitive factor.

• Scaling in operation: Companies like Meta must keep AI models available in real time for hundreds of millions of users.
• Specialized hardware: Inference workloads benefit from GPU clusters optimized for throughput and efficiency.
• Distributed infrastructure: Multiple data center locations reduce latency and increase resilience.
• Long-term planning reliability: Multi-year contracts like this one allow companies to build their AI strategy on a stable infrastructure foundation.

Context and Significance

The contract between Meta and CoreWeave sends a signal to the entire industry. It demonstrates that major technology companies are willing to make long-term, billion-dollar commitments for specialized AI infrastructure. For CoreWeave, the agreement provides substantial revenue security over several years and confirms the company’s positioning as a central infrastructure provider within the AI ecosystem.

For Meta, in turn, the deal is part of a broader strategy to deeply integrate AI into its products and platforms — from social networks and messaging to mixed-reality applications. The choice of an external cloud provider alongside its own data centers suggests that capacity requirements are increasingly outpacing internal infrastructure.