AMD, Meta ink multi‑year pact to deploy up to 6GW of AMD Instinct GPUs

Advanced Micro Devices and Meta have announced an expanded strategic partnership to deploy up to 6 gigawatts of AMD Instinct GPUs across Meta’s next generation AI infrastructure, formalised on 24 February 2026. The companies said the agreement is multi‑year and multi‑generation, signalling a long runway for capacity additions and product updates.

Inside the agreement

According to AMD, the first wave will be powered by a custom AMD Instinct GPU based on the MI450 architecture, integrated into Helios, AMD’s rack‑scale platform that was showcased with Meta at the 2025 Open Compute Project Global Summit. Shipments to support the initial 1 gigawatt deployment are scheduled to begin in the second half of 2026, the company said.

The two companies are also aligning silicon, systems and software roadmaps under the partnership. That alignment spans AMD’s GPUs and its server processors, with the first deployments running ROCm software and 6th Gen AMD EPYC CPUs, codenamed Venice, as per AMD.

Broader compute strategy

Meta framed the deal as part of a portfolio approach to a resilient infrastructure stack that mixes partners and its own silicon programmes for different workloads. The company said shipments supporting the first GPU deployments will begin in the second half of 2026 and will use the Helios rack‑scale architecture that Meta and AMD developed through the Open Compute Project.

“We are proud to expand our strategic partnership with Meta as they push the boundaries of AI at unprecedented scale,” said Dr Lisa Su, chair and CEO of AMD. Mark Zuckerberg, founder and CEO of Meta, called it a long‑term partnership to “deploy efficient inference compute” as the company diversifies its compute options.

Financial terms and structure

To further align incentives, AMD disclosed a performance‑based warrant for up to 160 million shares of its common stock issued to Meta. The vesting is linked to milestones tied to Instinct GPU shipments, beginning with the first gigawatt and scaling up as purchases reach 6 gigawatts, with additional conditions tied to stock price thresholds and Meta’s technical and commercial milestones.

How will the first deployments roll out

AMD said the first gigawatt tranche will ship in the second half of 2026, built around a custom MI450‑based Instinct GPU and 6th Gen EPYC CPUs on the Helios rack‑scale design that AMD and Meta co‑developed through the Open Compute Project. Meta has positioned this as foundational capacity for scaling AI services, including personal superintelligence.

Why this matters

The companies have framed the scale in gigawatts, a measure of power capacity, rather than listing an accelerator count. Both sides emphasised roadmap alignment across silicon, systems and software to hasten model development and deployment, and to diversify compute supply. AMD also said the partnership is expected to drive multi‑year revenue growth and be accretive to its non‑GAAP earnings per share.

• Scope, up to 6 gigawatts of AMD Instinct GPUs for Meta’s AI infrastructure under a multi‑year, multi‑generation pact.
• Technology, custom MI450‑based Instinct GPUs on Helios rack‑scale architecture, paired with 6th Gen AMD EPYC CPUs running ROCm.
• Timing, shipments for the first 1 gigawatt begin in the second half of 2026, according to both companies.
• Structure, Meta receives a performance‑based warrant for up to 160 million AMD shares, vesting as GPU shipment milestones and other conditions are met.

AMD scheduled a conference call at 8.30 a.m. ET on 24 February 2026 to discuss the announcement, with a live audio webcast on its investor relations page, the company said.