Meta is set to deploy standalone Nvidia Grace CPUs in its production data centers, marking a significant step in Nvidia’s push to become a major standalone data center CPU vendor. The move is part of a new multi-year strategic partnership between the two companies, aimed at accelerating Meta’s AI infrastructure development. As part of the agreement, Meta will scale up its use of Nvidia’s Blackwell and upcoming Vera Rubin GPUs, with plans to deploy “millions” of these accelerators across its global network of hyperscale data centers. This expansion is part of a broader AI investment strategy that could total up to $135 billion by 2026. Nvidia’s data center head, Ian Buck, confirmed that Meta is already seeing performance-per-watt improvements of up to 2x on certain CPU workloads using the Grace platform. Early testing of Nvidia’s next-generation Vera CPU has also shown “very promising” results, with large-scale Vera-only deployments expected to begin as soon as 2027. The Grace CPU, available in a standalone C1 configuration, features 72 Arm Neoverse V2 cores and supports up to 480GB of LPDDR5X memory. For higher-performance needs, Nvidia offers the Grace CPU Superchip, which combines two dies via NVLink-C2C interconnect, delivering 144 cores, up to 960GB of memory, and aggregate memory bandwidth of up to 1024 GB/s. Vera, Nvidia’s newest data center CPU, takes performance and security further with 88 custom Arm cores and 176 threads, support for up to 1.5 TB of LPDDR5X memory, and memory bandwidth of up to 1.2 TB/s. It also features PCIe Gen 6 and Compute Express Link 3.1 connectivity. Notably, Vera is the first Nvidia CPU to provide a fully confidential or trusted computing environment across entire rack-scale systems. Beyond CPUs, Meta will also deploy Nvidia’s Spectrum-X Ethernet switches throughout its infrastructure. These switches, which include co-packaged optics, eliminate the need for active cabling by integrating optical transceivers directly into the switch. This innovation can reduce power consumption in data center racks by up to 10%, freeing up more energy for AI workloads—critical as data centers push for maximum compute density and efficiency. In addition to hardware, Nvidia will provide its in-house expertise in AI model design and optimization to help Meta’s engineers fine-tune and enhance the performance of its core AI applications. This partnership underscores how deeply Nvidia’s influence extends beyond GPUs and into the foundational layers of modern AI infrastructure. As Meta scales its AI ambitions, the integration of Nvidia’s full-stack technology—from CPUs and networking to software and model development—demonstrates a growing trend: AI infrastructure is increasingly shaped by a single ecosystem, one that’s becoming indispensable across the tech industry.