Meta Platforms expanded its long-running partnership with Nvidia, agreeing to deploy millions of chips across its U.S. data-center buildout, including what Nvidia called the first large-scale use of its Grace CPUs as standalone processors. The multiyear pact also covers next-generation Vera Rubin rack-scale systems and Spectrum-X Ethernet networking; financial terms weren’t disclosed. The commitment aligns with Meta’s plan to spend up to $135 billion on AI in 2026 and $600 billion in the U.S. by 2028 as it builds 30 data centers, with major sites underway in Ohio and Louisiana. Shares of Meta and Nvidia rose in late trading while AMD fell, as the deal helps secure supply of Nvidia’s current Blackwell GPUs and forthcoming Rubin components. The companies will co-design to optimize models for Meta’s platforms, including AI features on WhatsApp, even as Meta pursues in-house silicon and evaluates alternatives such as AMD and Google’s TPUs.
Related articles:
NVIDIA Grace CPU for AI and HPC
Meta Introduces Llama 3
NVIDIA Grace Hopper Superchip
