OpenAI and AMD have signed a multi year agreement for six gigawatts of AMD Instinct GPUs. The first one gigawatt deployment uses the Instinct MI450 series and is scheduled to start in the second half of 2026. The companies describe a collaboration that spans multiple product generations and deeper codevelopment at the rack and system level. OpenAI
A warrant for up to one hundred sixty million AMD shares is part of the agreement. The warrant vests as OpenAI reaches purchase milestones from one to six gigawatts, with additional vesting tied to AMD share price targets disclosed in securities filings. Several reports based on those filings say the exercise price is one cent per share.
The scale is significant for the industry. Six gigawatts of AI compute is roughly comparable to the power use of four to five million homes in the United States. For AMD, coverage characterizes the deal as a transformative revenue opportunity and a competitive step in data center AI, while for OpenAI it provides a second large scale supplier for frontier training and inference.
Timelines and execution risks will be in focus through 2026. The first phase starts in the second half of that year. Progress depends on chip availability, facility construction, and power procurement. AMD’s forward looking statements highlight these uncertainties for investors and partners. OpenAI+1
What this means for startups and developers
-
Expect more capacity in the market over the next two to three years which can improve access to training and inference for fast growing applications.
-
Watch for software enablement around MI450 such as libraries and optimized frameworks that could narrow performance gaps in common workloads.
-
Multi cloud and multi vendor strategies become easier when hyperscale providers source from more than one GPU line.
