AMD signed a long-term agreement with OpenAI to supply six gigawatts of AI chips over multiple years, with the first delivery of its Instinct MI450 GPUs slated for the second half of 2026. In tandem, OpenAI will have the right to acquire up to 160 million AMD shares-roughly 10 percent of the company-subject to performance and stock-price milestones.

Under the pact, AMD will act as a core compute partner for OpenAI's next-generation infrastructure. The deal is structured in tranches: the initial one gigawatt deployment will trigger the first warrant vesting, followed by further tranches as deployment scales. Vesting is also tied to AMD achieving share-price thresholds that climb up to $600 per share.

This marks a significant shift in OpenAI's hardware strategy, which until now has leaned heavily on Nvidia. Earlier this year, OpenAI had unveiled a $100 billion strategic cooperation with Nvidia to deploy up to ten gigawatts of processing systems. That arrangement is still in effect, but the AMD agreement offers diversification and direct governance incentives.

AMD expects the collaboration with OpenAI to drive“tens of billions” in incremental revenue annually. The firm projects the ripple effect from this and related clients could generate over $100 billion in new revenue across four years. Following the announcement, AMD shares rose sharply-up over 20 percent in pre-market trading. Nvidia's stock dipped modestly.

Lisa Su, AMD's President and CEO, called OpenAI a partner from the early design phase of MI450, citing OpenAI's technical feedback on memory architecture and inference requirements. OpenAI CEO Sam Altman described AMD's role as critical to accelerating AI development, noting that compute capacity remains a bottleneck.

Industry analysts observe that the deal elevates AMD's position in the AI chip market and directly challenges Nvidia's long-standing dominance. AMD's strategy of coupling powerful hardware with an open software stack via ROCm has gained traction with major AI players, and with this agreement, the company aims to deepen its foothold.

Power and energy infrastructure loom as key challenges. The scale of compute deployment implies enormous electricity demand, and experts are warning that data centre operators may need to rely increasingly on self-generation or off-grid solutions to avoid overburdening public grids. One estimate suggests utilities may need to add around 60 gigawatts of new capacity by the end of the decade to meet AI growth.

OpenAI's parallel efforts to produce its own custom chips add complexity to the story. The organisation has been reported to collaborate with Broadcom on internally deployed silicon slated for mass production in 2026. Those chips are meant for OpenAI's internal use and not for external sale.

Competitive dynamics in the chip industry are intensifying. Nvidia is racing ahead with its Vera Rubin architecture, while AMD is aligning its roadmap around rack-scale systems built to compete directly at scale. Other players, including Broadcom, Intel, and custom silicon startups, are increasingly seen as critical to the broader AI infrastructure race.

MENAFN12102025000152002308ID1110183673