Meta, the operator of Facebook and Instagram, signed a multi-year deal with Nvidia to build next-generation AI data centers.
The two companies said on the 17th (local time) that Meta will deploy the latest graphics processing unit (GPU) "Blackwell" and the next-generation "Rubin" by the millions to expand its AI clusters. The deal goes beyond GPUs to include Nvidia's "Grace" CPU. Meta decided to adopt Grace as a standalone server chip for its data centers, seen as the first case among major data center operators to use Nvidia CPUs exclusively.
Meta said it applied Nvidia's Spectrum-X networking platform across its infrastructure to improve operational and power efficiency, and used confidential computing to both protect user data on WhatsApp and implement AI features.
Meta CEO Mark Zuckerberg said, "We will build a leading cluster with the Vera Rubin platform to deliver personalized superintelligence," and Nvidia CEO Jensen Huang said, "We will support Meta's next-generation AI foundation through co-design of CPUs, GPUs, networking, and software."
Meta is building the 1 GW-class "Prometheus" data center in Ohio and the 5 GW-class "Hyperion" data center in Louisiana. It once weighed adopting Google TPUs to keep Nvidia in check, but the new deal reaffirms its status as a key customer.