Meta and AMD Sign Long-Term Deal for Up to 6GW of AI Infrastructure

Meta said on February 24, 2026 that it had signed a long-term agreement with AMD to power its AI infrastructure with up to 6GW of AMD Instinct GPUs. The company framed the deal as more than a component purchase: Meta and AMD say they will align their roadmaps across hardware, software, and systems so the partnership can support large-scale AI deployment over multiple generations.

AMD described the scope as spanning Instinct GPUs, EPYC CPUs, and rack-scale AI systems. Meta said shipments supporting the first GPU deployments will begin in the second half of 2026 and will be built on the Helios rack-scale architecture that the two companies previously discussed at the Open Compute Project Global Summit. That detail matters because it suggests the agreement is meant to shape how clusters are assembled and operated, not just which accelerators Meta buys.

Why Meta Is Doing This

• Meta wants more flexible, resilient AI infrastructure as model demand grows.
• The company is explicitly diversifying beyond a single-vendor compute strategy.
• The agreement fits Meta Compute, Meta’s broader effort to scale infrastructure for what it calls the era of personal superintelligence.
• Meta also continues to invest in its in-house MTIA silicon program alongside external partners.

For Meta, the logic is straightforward. AI products now depend on sustained access to training and inference capacity, and the supply chain for that capacity remains strategically constrained. A multi-year agreement with AMD gives Meta leverage, supply diversity, and a tighter feedback loop between workload requirements and future hardware design. For AMD, the announcement is equally significant because it places the company deeper inside one of the industry’s largest AI buildouts.

The immediate outcome will still depend on whether the second-half 2026 deployments arrive on schedule and deliver the performance Meta wants. Even so, the announcement is a clear marker of how hyperscalers are re-architecting their AI stacks: not around one chip win, but around portfolios of vendors, custom silicon, and rack-scale system design that can keep pace with rising demand.