As part of the OpenAI-AMD deal, OpenAI will receive the first GW worth of AMD's Instinct MI450 chip in H2 2026, when the new chip is scheduled for deployment
AMD has signed a multi-year chip supply deal with OpenAI that could generate tens of billions in revenue for the chipmaker, helping accelerate its momentum in the AI industry.
Context & Ripple Effects
This delivery schedule turns the broader OpenAI-AMD agreement for up to 6GW of Instinct GPUs into a concrete initial capacity commitment, with MI450 as the first deployment milestone.
It also sits alongside OpenAI’s reported plan for a Broadcom co-designed chip for internal use, indicating that the company is pursuing more than one route to secure AI compute rather than relying on a single supplier.
First-order effects
- OpenAI gains a defined first 1GW allocation of MI450 capacity in H2 2026, tying part of its infrastructure planning to AMD’s new-chip deployment schedule.
- AMD obtains an early flagship deployment for MI450 and a clearer initial tranche of demand within a multi-year supply agreement that could generate tens of billions in revenue.
Second-order effects
- AMD must execute on MI450 availability and the supporting software stack at a scale that makes the initial allocation usable for OpenAI; delivery performance will shape the credibility of later capacity commitments.
- OpenAI’s mix of AMD supply and a reported in-house Broadcom design strengthens its purchasing alternatives, adding pressure on incumbent GPU suppliers to compete on availability, economics, and ecosystem support.
Third-order effects
- If large AI labs continue to reserve capacity across merchant GPUs and custom silicon, AI compute procurement may shift from opportunistic chip buying toward long-duration, portfolio-style infrastructure contracts.
- Large committed allocations could make credible software compatibility and system-level deployment support more important to accelerator competition, not just peak chip performance.
The trend: AI developers are diversifying compute supply through long-term accelerator contracts and custom-chip programs to secure capacity and reduce dependence on any one platform.