AMD has signed a comprehensive, multi‑year and multi‑generation agreement with OpenAI to deploy six gigawatts of Instinct accelerators, starting with one gigawatt of Instinct MI450 series capacity in the second half of 2026. The deal makes OpenAI a lead customer for both MI450 and AMD’s Helios rack‑scale systems, marking one of the first public commitments by a major Artificial Intelligence player to adopt AMD’s next‑gen platform. OpenAI’s parallel embrace of NVIDIA’s Vera Rubin platform highlights its multi‑vendor strategy, but AMD is now firmly in the conversation for flagship training and inference infrastructure.
On a conference call, AMD CEO Lisa Su framed the partnership as a catalyst for the company’s data center ambitions, expecting revenue to begin in the second half of 2026 and to add double‑digit billions in annual incremental data center Artificial Intelligence revenue once the program ramps. Internally, AMD characterizes MI450 as pivotal, comparable in strategic importance to the company’s EPYC Milan breakthrough in server CPUs. The MI450 platform is described as leveraging the latest process technology, high‑bandwidth memory, architectural enhancements and higher TGPs, with a ramp that extends to rack‑scale deployments via the Helios system built around MI400‑series chips and next‑gen EPYC Venice CPUs.
Su emphasized detailed supply chain preparations for MI450 and Helios, noting the use of two nanometer technology and confirming broader “strategic” customer interest beyond OpenAI. While AMD acknowledges it is late to a robust rack‑scale portfolio, the company expects strong demand as it scales production through 2026. The agreement positions AMD as the second major Artificial Intelligence compute supplier after NVIDIA, and signals intensifying competition that could benefit the broader computing ecosystem by increasing choice and accelerating innovation at cloud scale.
