Meta signs long-term AMD deal for up to 6GW of AI GPU capacity

A very large infrastructure commitment

Meta announced on February 24, 2026 that it has signed a long-term AI infrastructure agreement with AMD covering up to 6GW of AMD Instinct GPUs. The company described the deal as a multi-year arrangement that also aligns Meta and AMD roadmaps across hardware, software, and systems. At this scale, the announcement is not a routine supplier expansion. It is a strategic statement about how Meta intends to secure enough compute for its next AI phase.

Meta framed the agreement as part of its push to build what it calls personal superintelligence. That framing is ambitious, but the concrete message is simpler: the company expects AI workloads to keep rising fast enough that it needs massive, durable access to accelerator capacity rather than opportunistic purchasing.

What AMD is supplying

Meta said the partnership covers more than chips alone. Lisa Su described it as a multi-year, multi-generation collaboration spanning Instinct GPUs, EPYC CPUs, and rack-scale AI systems. Meta also said the first GPU deployments supported by this agreement will begin shipping in the second half of 2026 and will be built on the Helios rack-scale architecture that Meta and AMD previously discussed through the Open Compute Project ecosystem.

That matters because hyperscale AI buying is shifting from standalone accelerator orders to co-designed systems. Once companies optimize at rack level, they can tune networking, power, memory balance, and software together. That usually produces more durable performance gains than chip procurement alone.

Why Meta is diversifying now

Meta explicitly tied the deal to a portfolio-based infrastructure strategy. The company said it wants a more flexible and resilient technology stack built with diverse partners for different workloads. In the same announcement, Meta pointed to its own Meta Training and Inference Accelerator (MTIA) program, making clear that this is diversification rather than replacement. In other words, Meta is trying to avoid single-vendor dependence while still locking in large external supply.

From AMD's perspective, the agreement is also meaningful. Lisa Su called it one of the industry's largest AI deployments and positioned AMD near the center of the global AI buildout. If shipments and deployment milestones hold, the deal would materially expand AMD's role in hyperscale inference and training infrastructure.

Why the market should pay attention

Deals measured in gigawatts say something important about where AI competition is headed. Frontier progress still depends on models, but the rate of deployment increasingly depends on power, cooling, racks, and multi-year silicon commitments. Meta's AMD agreement is a strong example of that shift from model announcements to industrial-scale compute procurement.

Source: Meta official announcement.