Meta and NVIDIA Forge Multi-Year AI Infrastructure Partnership, Deploying Millions of GPUs

A Partnership Spanning Generations of Chips

Meta and NVIDIA announced a multiyear, multigenerational strategic partnership on February 17, 2026, encompassing GPUs, CPUs, networking, and software across Meta's U.S. data center buildout. Analysts estimate the deal's value at roughly $50 billion, embedded within Meta's stated plan to spend up to $135 billion on AI infrastructure in 2026 alone.

Hardware at Scale

• GPUs: Millions of NVIDIA Blackwell and next-generation Rubin GPUs for training and inference
• CPUs: NVIDIA Grace CPUs (Arm-based) deployed as standalone data center servers—a first for any company at this scale
• Networking: NVIDIA Spectrum-X Ethernet switches integrated with Meta's Facebook Open Switching System platform

A First: Grace CPUs as Standalone Servers

Meta becomes the first company to deploy NVIDIA Grace CPUs in standalone server configurations—without pairing them with GPUs—at large scale. CEO Jensen Huang stated the partnership enables "deep codesign across CPUs, GPUs, networking and software," and that Meta would see "significant performance-per-watt improvements" across its data centers.

WhatsApp Privacy and Confidential AI

Grace CPUs' confidential computing capabilities are intended to power WhatsApp's private AI processing, enabling on-device-style privacy guarantees at data center scale. Meta's $600 billion U.S. AI investment plan through 2028 will heavily rely on this infrastructure.

Source: NVIDIA Newsroom