Meta Expands Partnership with NVIDIA, Secures First-Mover Rights for Grace CPUs

On Tuesday (February 17), Meta and NVIDIA announced a new long-term strategic partnership covering on-premise, cloud, and artificial intelligence (AI) infrastructure.

According to the press release, Meta will construct hyperscale data centers optimized for both training and inference to support the company’s long-term AI infrastructure roadmap. This collaboration will involve the large-scale deployment of millions of Blackwell and Rubin GPUs, alongside NVIDIA's Grace CPUs. Furthermore, NVIDIA Spectrum-X Ethernet switches will be integrated into Facebook’s Open Switching System (FBOSS) platform.

The statement noted, “This partnership represents the first hyperscale deployment of NVIDIA Grace and improves performance-per-watt for every generation of processors through co-design of CPU ecosystem libraries and investments in software optimization.” The two companies are also collaborating on the deployment of NVIDIA’s Vera CPUs, with large-scale rollout expected by 2027, further expanding Meta’s energy-efficient AI computing footprint and advancing the broader Arm software ecosystem.

NVIDIA CEO Jensen Huang stated in the release, “No company is deploying AI at the scale of Meta—combining frontier research with industrial-grade infrastructure to build the world’s largest systems for billions of users. Through deep co-design across CPU, GPU, networking, and software, we are bringing the full NVIDIA platform to Meta’s researchers and engineers to help them build the foundation for the next generation of AI frontiers.”

Mark Zuckerberg added, “We are excited to expand our work with NVIDIA, leveraging their Vera Rubin platform to build leading clusters that provide personal super-intelligence for everyone in the world.” Meta also indicated in the statement that it will integrate NVIDIA’s security technologies into the AI features of its messaging app, WhatsApp. Following the announcement, shares of Meta and NVIDIA rose in after-hours trading, while AMD shares at one point fell by over 4%.

It is worth noting that this is not the first collaboration between the two companies—Meta has utilized NVIDIA GPUs for over a decade—but this agreement marks a deeper and more extensive technical partnership between these two Silicon Valley giants. The most significant new highlight of this deal is the CPU; Meta will become the first company to deploy Grace standalone in its data centers, rather than integrating it with GPUs in the same server.

Last month, Meta announced that it expects its capital expenditures to reach as high as $1350 billion this year. At the time, Ben Bajarin, a chip analyst at Creative Strategies, remarked: “We anticipate that a substantial portion of Meta's CAPEX will be directed toward NVIDIA's data center expansion. These chips are specifically designed for inference workloads and intelligent agent tasks. Meta's large-scale adoption validates NVIDIA's 'full-stack' infrastructure strategy of positioning both CPUs and GPUs simultaneously.”

However, Meta is not relying solely on NVIDIA. Reports surfaced last November that Meta was considering the use of Google’s Tensor Processing Units (TPUs) in its 2027 data centers, news which caused NVIDIA’s stock to drop 4% at the time. Furthermore, Meta is developing its own in-house chips and continues to use AMD products. Amid ongoing NVIDIA supply constraints, major AI firms are actively seeking "second-source" suppliers.