Meta is deploying Nvidia standalone CPUs at scale across its infrastructure.
Meta has become one of the first hyperscalers to deploy Nvidia's standalone Grace CPUs at scale, according to an announcement made Tuesday. The company has already deployed these Grace processors in CPU-only systems and is working with Nvidia to field its upcoming Vera CPUs beginning next year. This expansion builds on Meta's previous deployments of Nvidia's Grace-Hopper Superchips as part of its Andromeda recommender system. Meta now plans to field "millions" of Nvidia GB300 and Vera Rubin Superchips across its infrastructure.
The shift marks a significant departure from Nvidia's typical deployment model. Apart from a few scientific institutions, Nvidia's nearly three-year-old Grace CPUs have predominantly shipped as part of "Superchips" which featured integrated Hopper and Blackwell GPUs onboard. Meta is now using Nvidia's CPU-only Grace systems to power both general purpose and agentic AI workloads that don't require a GPU. According to Ian Buck, Nvidia's VP and General Manager of Hyperscale and HPC, "Grace is an excellent backend datacenter CPU. It can actually deliver 2x the performance per watt on those back end workloads." Grace CPUs are equipped with 72 Arm Neoverse V2 cores clocked at up to 3.35 GHz, available in a standalone configuration with up to 480 GB of memory, or as a Grace-CPU Superchip combining two processors with up to 960 GB of LPDDR5x memory and up to 1 TB of memory bandwidth between the two dies.
Nvidia's Vera CPU, officially unveiled at CES earlier this year, represents a significant upgrade with 88 custom Arm cores, support for simultaneous multi-threading, and confidential computing functionality. Meta will leverage Vera's confidential computing capability for private processing and AI features in its WhatsApp encrypted messaging service. Meta's adoption of Nvidia CPUs runs counter to the broader industry trend, which has increasingly pivoted to custom Arm CPUs like Amazon's Graviton or Google's Axion.
The expanded partnership reflects Meta's massive infrastructure investments, with the hyperscaler deploying additional Nvidia GPUs and Spectrum-X network infrastructure alongside the CPU deployments. This aligns with Meta's capex target of $115 billion to $135 billion for 2026. While Nvidia remains tight-lipped about the collaboration's scale, the company is expected to gain tens of billions in revenue from the deal. At north of $3.5 million per rack, a million GPUs would represent approximately $48 billion—a substantial figure even for Nvidia, which earned $31.9 billion in net income on revenues of $57 billion in the quarter ended October 26.
Despite the deep Nvidia partnership, Meta is not an exclusive shop. The company maintains a significant fleet of AMD Instinct GPUs in its datacenters and was directly involved in designing AMD's Helios rack systems, which are due out later this year. Meta is widely expected to deploy AMD's competing rack systems, though no formal commitment has yet been made.