AMD & Meta sign vast AI GPU & CPU infrastructure pact
AMD and Meta have agreed to a multi-year, multi-generation deal focused on deploying up to 6 gigawatts of AMD Instinct GPUs for Meta's next-generation AI infrastructure.
The arrangement spans multiple hardware generations and includes joint work on a custom GPU design based on AMD's MI450 architecture. The companies also plan closer alignment across CPU, system and software roadmaps.
Shipments for the first 1 gigawatt are expected to begin in the second half of 2026. That initial phase will use AMD's Helios rack-scale architecture, a custom MI450-based Instinct GPU, and AMD's 6th Gen EPYC CPU, codenamed "Venice".
The agreement also expands Meta's use of AMD server processors. Meta will be a lead customer for 6th Gen EPYC "Venice" and for "Verano", a next-generation EPYC processor that AMD says includes workload-specific optimisations.
Compute diversification
Meta has been expanding its AI infrastructure across its global data centre estate as it trains and runs AI models for consumer products and developer services. It has also sought to diversify computing capacity and silicon suppliers as demand for AI hardware tightens across the sector.
Meta CEO Mark Zuckerberg framed the AMD agreement as part of that approach. "We're excited to form a long-term partnership with AMD to deploy efficient inference compute and deliver personal superintelligence," he said. "This is an important step for Meta as we diversify our compute. I expect AMD to be an important partner for many years to come."
The scale cited in the deal is unusually large for a bilateral supplier relationship in the AI chip market. Gigawatt-level language ties the procurement programme to the power consumption of full data centre deployments rather than a single product order. The companies did not disclose financial terms or a unit count for the GPUs covered by the 6-gigawatt figure.
Custom MI450
Under the agreement, the companies will co-engineer a custom GPU based on AMD's MI450 architecture, optimised for Meta's AI workloads, according to AMD.
The first deployments will run ROCm software and use Helios rack-scale architecture. AMD introduced Helios at the 2025 Open Compute Project Global Summit and says it was developed jointly with Meta through the Open Compute Project.
AMD Chair and CEO Dr. Lisa Su said the deal aligns product roadmaps across GPUs, CPUs and rack-scale systems. "We are proud to expand our strategic partnership with Meta as they push the boundaries of AI at unprecedented scale," Su said. "This multi-year, multi-generation collaboration across Instinct GPUs, EPYC CPUs and rack-scale AI systems aligns our roadmaps to deliver high-performance, energy-efficient infrastructure optimized for Meta's workloads, accelerating one of the industry's largest AI deployments and placing AMD at the center of the global AI buildout."
Warrant structure
AMD says it has issued Meta a performance-based warrant for up to 160 million shares of AMD common stock. The vesting schedule is tied to milestones associated with Instinct GPU shipments, starting with the initial 1 gigawatt and scaling through additional tranches up to the 6-gigawatt level.
Vesting is also tied to AMD meeting certain stock price thresholds. Exercising the warrant depends on Meta meeting technical and commercial milestones.
The structure links hardware procurement and delivery to an equity incentive, potentially aligning buyer and supplier interests over long deployment cycles in which delivery schedules, product ramp timing and system qualification can span years.
AMD Chief Financial Officer Jean Hu described the financial impact in terms of growth and earnings.
"We expect this partnership to drive substantial multi-year revenue growth and be accretive to our non-GAAP earnings per share, marking another significant step forward in delivering on our ambitious long-term financial model," said Jean Hu, EVP, CFO and Treasurer, AMD. "The performance-based structure also tightly aligns AMD and Meta around execution and long-term value creation."
Meta has already deployed AMD EPYC CPUs at scale and has used Instinct MI300 and MI350 series GPUs across its infrastructure, according to AMD. Under the expanded partnership, the companies plan to align work across silicon, systems and software as Meta prepares for the first 1 gigawatt phase beginning in the second half of 2026.