For the first time, Nvidia is beginning to sell AI CPUs independently.
For the first time, Nvidia is beginning to sell AI CPUs independently.
Meta has entered into a multiyear agreement to enhance its data centers with millions of Nvidia’s Grace and Vera CPUs as well as Blackwell and Rubin GPUs. While Meta has previously utilized Nvidia’s hardware for its AI products, this agreement “marks the first significant Nvidia Grace-only deployment,” which Nvidia claims will bring “notable performance-per-watt enhancements in [Meta’s] data centers.” The agreement also encompasses plans to incorporate Nvidia’s upcoming Vera CPUs into Meta’s data centers by 2027.
Meta is additionally developing its own in-house chips for executing AI models, but as reported by the Financial Times, the company has encountered “technical challenges and delays in rollout” regarding its chip initiative. Nvidia is similarly confronting issues related to depreciation and chip-back loans utilized to finance AI expansions, alongside competitive pressures. CNBC mentions that Nvidia’s stock fell four percent following a November report indicating that Meta was contemplating the use of Google’s Tensor chips for AI, and late last year, AMD disclosed chip partnerships with both OpenAI and Oracle.
Both Nvidia and Meta have not revealed the financial details of the agreement, but the projected AI expenditures from Meta, Microsoft, Google, and Amazon this year are expected to exceed the total cost of the Apollo space program.
