In a multiyear deal, Meta Platforms will deploy “millions” of processors from semiconductor giant NVIDIA to power its next wave of AI infrastructure, further deepening ties between the two industry powerhouses. The agreement spans Meta’s on-premises and cloud environments, targeting hyperscale data centres that will handle both AI training and inference workloads.
According to Meta CEO Mark Zuckerberg, the company plans “to build leading-edge clusters using NVIDIA’s next-generation platforms to deliver personal superintelligence to everyone in the world.” Both NVIDIA’s current Blackwell generation of AI accelerators and its upcoming Vera Rubin AI platform will be deployed by Meta, with the latter even planning to increase deployment of NVIDIA’s Arm-based Grace CPUs across production data centres, aiming to improve performance per watt.
Meta will also integrate NVIDIA’s advanced networking devices across its network infrastructure, supporting higher efficiency and throughput for large AI workloads. The social media giant will also adopt NVIDIA’s “Confidential Computing Technology,” designed to secure data and applications, to enable AI features on its WhatsApp platform while maintaining data confidentiality.
According to Bloomberg, Meta already accounts for about 9% of NVIDIA’s revenue and is the chip maker’s second-largest customer, spending roughly USD 19 billion on its products in the last fiscal year. The strengthened partnership will also be a priority for Meta, focusing on the rapid expansion of its AI investments, in addition to developing large-scale data centres and exploring in-house chip development.
NVIDIA is rewriting the rules of the data centre. The company initially designed these Arm-licensed processors to serve merely as companions to their primary AI accelerators. Now they aggressively market them as robust standalone solutions capable of powering complex databases and autonomous digital agents.
Ian Buck, who manages hyperscale computing for NVIDIA, proudly noted that the Grace architecture can execute demanding database operations while consuming half the electricity of rival hardware. Meta has already begun testing the forthcoming Vera systems with highly promising results.
This alliance reveals a fascinating corporate dynamic. Mark Zuckerberg and his engineering teams at Meta are actively developing proprietary silicon. They have even engaged in quiet conversations regarding Google and its bespoke “Tensor Processing Units.” Despite these wandering eyes, NVIDIA successfully anchored its relationship with one of its most critical patrons.