Nvidia said on the 20th that it signed a long-term strategic partnership with Meta to build next-generation artificial intelligence (AI) infrastructure.
Meta plans to build hyperscale data centers under its long-term AI infrastructure roadmap. Nvidia will collaborate on both the on-premises (internal network) and the cloud that Meta is building.
Under the partnership, Nvidia will supply its central processing units (CPUs) and millions of Blackwell and Rubin graphics processing units (GPUs). These will be used to run Meta's Facebook Open Switching System platform. Nvidia's Spectrum-X Ethernet switches (a high-speed networking platform optimized for Generative AI) will also be integrated.
Nvidia and Meta are also working together to deploy Arm-based Nvidia Grace CPUs for data center production applications. Meta aims to improve performance per watt across its data centers through this effort. This collaboration marks the first large-scale deployment of the Nvidia Grace stand-alone architecture. The two companies are also working to deploy the Vera CPU in 2027.
Jensen Huang, Nvidia chief executive officer (CEO), said, "There are no corporations deploying AI at the scale of Meta," and added, "Meta is integrating cutting-edge research with industrial-scale infrastructure to deliver some of the world's largest-scale personalization and recommendation systems for billions of users." He continued, "Through close co-design across CPUs, GPUs, networking, and software, Nvidia is providing the full-stack platform that enables Meta researchers and engineers to build the foundation for next-generation AI innovation."
Mark Zuckerberg, Meta chief executive officer (CEO), said, "We are very pleased to expand our collaboration with Nvidia to build a state-of-the-art cluster leveraging the Vera Rubin platform, enabling us to deliver personalized superintelligence to everyone around the world."
Nvidia said, "We are working together to extend Nvidia confidential computing capabilities beyond WhatsApp across the Meta portfolio, supporting AI for large-scale privacy enhancement," and added, "We are also conducting deep co-design to optimize and accelerate state-of-the-art AI models."