CoreWeave and Meta sign $21bn expanded AI infrastructure agreement

CoreWeave and Meta expanded their existing cloud-compute partnership through a new agreement CoreWeave valued at approximately $21 billion, with capacity commitments running through December 2032. The deal adds dedicated capacity across multiple data-centre sites and, notably, will include some of the earliest deployments of Nvidia’s next-generation Vera Rubin platform, the chip architecture expected to succeed Blackwell.

Neither company detailed the split of GPU generations or specific site locations in the announcement; the disclosed terms were filed separately with the US Securities and Exchange Commission. CoreWeave chief executive Michael Intrator framed the expansion as evidence of continued demand for specialised AI cloud capacity, saying that “leading companies are choosing CoreWeave’s AI cloud to run their most demanding workloads.”

The agreement was one of several large infrastructure deals CoreWeave and other cloud providers signed with frontier labs and hyperscalers in the same month, part of a broader pattern in which AI developers were locking in multi-year compute commitments — often running to the early 2030s — well ahead of the hardware generations involved becoming available. For Meta, the deal extended its reliance on external cloud capacity alongside its own data-centre buildout, at a point when the company was also launching its first closed-weight frontier model, Muse Spark, under new AI leadership.