AMD basically copy-pasted its OpenAI deal for Meta, six gigawatts and ten percent equity included

Meta and AMD have agreed on a multi-year partnership covering up to six gigawatts of AMD GPUs, with a focus on inference. The deal includes an unusual equity component.

AMD will supply up to six gigawatts of Instinct GPUs for Meta's AI infrastructure, spanning several product generations. According to AMD, the first deliveries for an initial one-gigawatt rollout are set to begin in the second half of 2026.

Both companies highlight inference as a key use case, running already-trained AI models rather than training new ones. With billions of users to serve, Meta needs massive inference capacity to power its AI features at scale.

"We're excited to form a long-term partnership with AMD to deploy efficient inference compute and deliver personal superintelligence," said Meta CEO Mark Zuckerberg.

The deal also comes with an equity component: it includes performance-based warrants for up to 160 million AMD shares, roughly ten percent of outstanding shares. Vesting is tied to AMD achieving certain stock price thresholds and to specific shipment milestones, while exercise is tied to Meta achieving key technical and commercial milestones. The first tranche kicks in with the initial one-gigawatt delivery, with additional tranches following as the deployment scales to six gigawatts.

OpenAI has already struck a very similar deal with AMD covering up to six gigawatts of compute and ten percent of shares.

Custom chip built on MI450 architecture

According to AMD, the technical foundation is an Instinct GPU custom-built for Meta based on the MI450 architecture. It's paired with sixth-generation EPYC processors codenamed "Venice" and a future EPYC chip called "Verano" with workload-specific optimizations. The entire infrastructure runs on AMD's ROCm software stack.

The platform is the Helios rack-scale architecture, which AMD and Meta co-developed as part of the Open Compute Project and unveiled at the OCP Global Summit 2025. According to the announcement, both companies plan to closely align their roadmaps for silicon, systems, and software, effectively building vertical integration across the entire infrastructure stack.

Meta already runs millions of EPYC CPUs and significant quantities of Instinct MI300 and MI350 GPUs across its global infrastructure. The new agreement builds on that existing footprint. Meta will also become the lead customer for the upcoming "Venice" and "Verano" EPYC generations.

AMD makes its strongest case yet as an Nvidia alternative

For AMD, this deal is first and foremost a market signal. Nvidia has dominated the AI accelerator market almost unchallenged, and landing a customer like Meta—one willing to deploy six gigawatts of AMD hardware—gives AMD another major reference alongside OpenAI. The deal is strategic for Meta too: OpenAI has reportedly already been able to negotiate 30 percent lower Nvidia GPU prices simply by having a TPU deal with Google on the table.

AMD faces plenty of competition, though. Meta itself is developing its own AI inference chips that could eventually handle training workloads too. Microsoft and AWS are building custom silicon as well. And startups like Cerebras recently signed a billion-dollar deal with OpenAI to provide fast AI inference.