Meta Platforms has agreed to buy millions of Nvidia’s current and next-generation artificial intelligence chips in a multiyear strategic partnership announced on Tuesday, marking a significant expansion of a relationship that stretches back more than a decade.
The deal covers Nvidia’s existing Blackwell graphics processing units and its forthcoming Rubin chips, as well as standalone deployments of Nvidia’s Grace and Vera central processing units. Financial terms were not disclosed, though Ben Bajarin, chief executive of tech consultancy Creative Strategies, told the Financial Times the agreement was worth billions of dollars, and told CNBC that “the deal is certainly in the tens of billions of dollars.”
The partnership makes Meta the first large-scale buyer of Nvidia’s Grace CPUs as standalone chips, rather than bundled alongside GPUs in a single server unit. Ian Buck, general manager of Nvidia’s hyperscale and high-performance computing unit, told Reuters that Grace processors can use half the power for common tasks such as running databases, with further gains expected from the next-generation Vera chips. Next-generation Vera CPU deployment is planned for 2027.
Bajarin told the Financial Times that the CPU element was the most significant aspect of the announcement. “We were in the ‘training’ era, and now we are moving more to the ‘inference era’, which demands a completely different approach,” he said, adding that the shift toward inference workloads – running AI models rather than training them – opens a new competitive front for Nvidia against specialised chip start-ups.
The agreement also includes Nvidia’s Spectrum-X Ethernet networking switches, which link GPUs within large-scale data centres, and Nvidia’s Confidential Computing security technology for WhatsApp’s private processing features. Engineering teams from both companies will collaborate on optimising AI models across Meta’s core workloads.
Mark Zuckerberg, Meta’s chief executive, said the company was “excited to expand our partnership with NVIDIA to build leading-edge clusters using their Vera Rubin platform to deliver personal superintelligence to everyone in the world.” The deal forms part of Meta’s broader commitment to spend up to $135 billion on AI infrastructure in 2026 and $600 billion on US data centres by 2028.
Despite the expanded deal, Meta continues to develop in-house silicon and has held discussions with Google about using its Tensor Processing Unit chips. The Financial Times reported that Meta’s own chip strategy had encountered technical challenges and delays, citing a person familiar with the matter. Shares in Advanced Micro Devices fell around four per cent on Tuesday following the announcement.
