NVIDIA and Meta Partner on Millions of GPUs, New CPUs for AI

NVIDIA and Meta have unveiled a sweeping, multiyear strategic partnership that will reshape AI infrastructure at scale. Under the deal, Meta will build hyperscale data centers optimized for both training and inference, deploying millions of NVIDIA Blackwell and Rubin GPUs along with Arm-based Grace CPUs and future Vera CPUs (targeting 2027). Meta will also integrate NVIDIA’s Spectrum-X Ethernet networking platform into its Facebook Open Switching System, providing predictable low-latency performance and improved power efficiency. The collaboration extends to software: Meta has adopted NVIDIA Confidential Computing for WhatsApp, enabling AI-powered features while maintaining user data confidentiality. Engineering teams from both companies will codesign and optimize state-of-the-art AI models across Meta’s core workloads, leveraging NVIDIA’s full-stack platform.

For Meta, this partnership is about unifying its AI compute architecture across on-premises data centers and cloud deployments, simplifying operations while maximizing performance and scalability. The adoption of Grace CPUs marks the first large-scale deployment of Arm-based servers solely for data center production applications, delivering significant performance-per-watt improvements. The integration of Spectrum-X Ethernet ensures network efficiency and throughput for AI workloads. By also pushing confidential computing beyond WhatsApp to other use cases, Meta aims to offer privacy-enhanced AI at massive scale. NVIDIA CEO Jensen Huang said, 'No one deploys AI at Meta’s scale,' and Mark Zuckerberg called the partnership a step toward 'personal superintelligence for everyone.'