AMD and OpenAI strike multi-year GPU partnership targeting 6 gigawatts of AI compute

AMD and OpenAI said they have signed a multi-year, multi-generation partnership that aims to deploy 6 gigawatts of AMD GPU compute to power OpenAI’s next wave of AI infrastructure. The companies framed the agreement as a deepening of their existing collaboration and as a major step toward scaling the “rack-level” systems needed for frontier AI workloads.

Under the announced timeline, OpenAI’s initial deployment is slated to be a 1-gigawatt buildout using AMD Instinct MI450 series GPUs starting in the second half of 2026. The deal is designed to extend beyond that first phase into future GPU generations, suggesting a long-run roadmap alignment rather than a single product cycle purchase.

AMD said the partnership includes close technical coordination intended to optimize both product roadmaps and large-scale deployments. In practical terms, that means co-design and tuning efforts that span not only chips, but also systems integration and the software layers required to keep massive clusters efficient and reliable.

A key feature of the agreement is an equity warrant for OpenAI tied to specified milestones. The warrant structure is intended to increase alignment as deployments scale, with vesting linked to the initial 1-gigawatt rollout and then additional tranches as the partnership moves toward the 6-gigawatt target.

The announcement is significant for the competitive landscape in AI infrastructure. Nvidia remains the dominant supplier of AI accelerators, but AMD has been pressing to expand share as AI clusters grow larger and as buyers look to diversify supply, manage costs, and reduce dependence on a single vendor.