Meta Platforms has entered a multiyear agreement with Nvidia to significantly expand its data center capabilities using millions of Grace and Vera CPUs alongside Blackwell and Rubin GPUs. This deal marks the first large-scale deployment focused entirely on Nvidia’s Grace CPU platform, which is expected to deliver substantial performance-per-watt gains within Meta’s data centers. Plans also include integrating Nvidia’s next-generation Vera CPUs into Meta facilities in 2027.
While Meta continues developing its own in-house chips for running AI models, technical challenges and delays have slowed the rollout of its custom silicon strategy. Nvidia faces pressures from depreciation and chip-backed loans used to fund AI infrastructure, alongside growing competition from Advanced Micro Devices and Google. Despite speculation that Meta might use Google’s Tensor chips for AI workloads and AMD’s recent chip supply deals with OpenAI and Oracle, the financial terms of the Nvidia-Meta agreement were not disclosed. Meta’s AI-related spending this year, along with Microsoft, Google, and Amazon, is expected to surpass the cost of the Apollo space program, estimated at over $250 billion in today’s dollars.
