Meta signs multiyear NVIDIA deal to power next-gen AI data centers

Meta and NVIDIA have entered a multiyear, multigenerational partnership to develop hyperscale AI infrastructure across Meta’s on-premises data centers and cloud environments. This collaboration involves deploying large volumes of NVIDIA CPUs—including Arm-based Grace and future Vera processors—and millions of Blackwell and Rubin GPUs to scale AI training and inference systems that power Meta’s global platforms. The partnership emphasizes deep codesign across compute, networking, and software, integrating NVIDIA Spectrum-X Ethernet switches into Meta’s Facebook Open Switching System to enhance network efficiency and throughput for AI clusters. The goal is to build data centers optimized for training large frontier AI models and running inference at scale, supporting personalization and recommendation systems used by billions of Meta users. The deal also focuses on improving energy efficiency and performance per watt, with Meta undertaking the first large-scale rollout of Grace CPUs and planning potential large-scale adoption of Vera CPUs by 2027. Meta will deploy NVIDIA GB300-based systems and unify its architecture across data centers and NVIDIA Cloud Partner environments to