NVIDIA and Meta Announce Major Multi-Year AI Partnership
NVIDIA has announced a major multi-year partnership with Meta involving the deployment of millions of its Blackwell and upcoming Rubin GPUs. The deal also marks the first large-scale, standalone use of NVIDIA's Grace CPUs for AI agent workloads and will integrate its Spectrum-X Ethernet networking fabric.
- This partnership is a component of Meta's staggering $600 billion investment in U.S.-based AI infrastructure, which is planned to be rolled out by 2028. - The collaboration will extend to future hardware, with Meta planning a large-scale deployment of NVIDIA's upcoming "Vera" CPUs in 2027. This is part of a deep co-design effort between the two companies across CPUs, GPUs, networking, and software. - A key aspect of the deal is the implementation of NVIDIA's Confidential Computing technology, which will be used initially within WhatsApp to enable AI features while keeping user data encrypted during processing. - The NVIDIA Grace CPUs, based on Arm's Neoverse V2 cores, are designed to deliver double the performance per watt compared to traditional servers, a critical factor for Meta's hyperscale data centers. The upcoming "Vera" CPU will feature 88 custom Arm cores. - NVIDIA's Spectrum-X Ethernet platform is engineered to boost network performance by 1.6 times for generative AI, which is crucial for interconnecting the massive clusters of GPUs that Meta is deploying. - This agreement solidifies NVIDIA's position with a major client following reports in late 2025 that Meta was in talks to purchase billions of dollars worth of Google's Tensor Processing Units (TPUs). - The Blackwell GPUs offer a significant performance leap over the previous Hopper generation, with up to 20 petaflops of AI performance and a second-generation Transformer Engine for enhanced efficiency with large language models. The upcoming Rubin GPUs are expected to deliver 50 petaflops of FP4 performance. - Meta's goal with this massive hardware deployment is to build what CEO Mark Zuckerberg calls "personal superintelligence," aiming to power the next generation of AI products and recommendation systems for its billions of users.
