Meta Signs Long-Term NVIDIA Deal for Millions of AI Chips

Meta is expanding its data-center computing capacity to train and deploy generative AI while reducing its reliance on NVIDIA GPUs. The order, which outside estimates say could be worth as much as $100 billion, gives AMD’s Instinct GPUs and ROCm platform the backing of a major customer and bolsters its prospects of challenging NVIDIA’s market dominance.

AMD and Meta announced a five-year partnership on February 24, 2026, under which Meta will deploy up to 6 GW of Instinct GPUs. The first 1 GW is expected to ship in the second half of 2026, using customized MI450 chips, EPYC “Venice” processors and the Helios architecture. AMD also granted Meta warrants for up to 160 million shares, which could give it a stake approaching 10% if specified milestones are met.

Meta has long worked with Broadcom to develop its in-house MTIA AI accelerator chips. Custom designs for recommendation, advertising and generative AI workloads improve data-center computing efficiency. The initiative is central to securing the computing power Meta needs to build “personal superintelligence” and helps reduce its reliance on external GPU suppliers such as NVIDIA, along with the associated supply risks.

Meta and Broadcom have extended their MTIA partnership agreement through 2029 and plan to expand chip deployment. The collaboration will focus on the first MTIA chip built using a 2-nanometer process, which is expected to become a key foundation of Meta’s in-house AI computing infrastructure. The contract value and precise mass-production date have not been disclosed, but the agreement explicitly runs through 2029.

Meta is accelerating the expansion of computing capacity for AI model training and cloud services, while outsourcing some of its infrastructure needs. AI cloud provider Nebius can supply dedicated GPU computing resources. The partnership reflects how major technology companies are using outsourcing to shorten data-center deployment times and reduce the risk of computing capacity shortages.

Meta and Nebius signed AI infrastructure agreements worth a combined $27 billion in July 2026, including $12 billion in dedicated computing capacity to be supplied by Nebius. The partnership also includes the first large-scale deployments of the NVIDIA Vera Rubin platform, helping Meta expand its AI cloud operations and model-computing capacity.

As competition for generative AI computing capacity intensifies, Meta Platforms is expanding its data centers and reducing its reliance on NVIDIA as a single supplier. AMD is challenging the AI chip market with its Instinct accelerators and EPYC processors. The total value of the multiyear deal was not disclosed, but its 6GW scale is large enough to reshape the companies' supply and capital ties.

AMD and Meta announced on February 24, 2026, that Meta will deploy up to 6GW of AMD Instinct GPUs across multiple generations. The first 1GW will use customized products based on the MI450 architecture, with shipments scheduled to begin in the second half of 2026. AMD will also issue performance-based warrants covering up to 160 million shares at $0.01 per share. If the purchase and share-price thresholds are met and the warrants are exercised in full, Meta will own about 9% of AMD.