OpenAI taps AMD for massive AI chip deal aimed at boosting compute capacity

OpenAI has signed a multi-year deal with chipmaker AMD to deploy billions of dollars worth of AMD GPUs for building its next-generation AI infrastructure. Under the agreement, OpenAI will deploy six gigawatts worth of AMD GPUs, starting with the first one-gigawatt deployment of AMD Instinct MI450 GPUs in the second half of 2026.

​OpenAI CEO Sam Altman, while reacting to the deal in a joint release, said, "AMD’s leadership in high-performance chips will enable us to accelerate progress and bring the benefits of advanced AI to everyone faster."

​AMD has also issued the ChatGPT maker with a warrant to buy up to 160 million shares of its stock, which is roughly 10% of the chipmaker's share. The warrant will vest based on AMD achieving a certain share price target and OpenAI achieving technical and commercial milestones required to enable AMD deployments at scale. The first tranche vests with the initial one-gigawatt deployment in 2026.

AMD executives believe that the deal with OpenAI wold help bring in tens of billions of dollars in annual revenue and as a ripple effect of the agreement, it expects to receive more than $100 billion in new revenue over four years from OpenAI and other customers, Reuters reported.

AMD shares were up by more than 35% in the pre-market trading after the announcement and if the trend holds it could put them on track to open at their highest level since March 2024.

OpenAI and its chip deals:

The deal with AMD comes shortly after Nvidia announced a strategic partnership and investment of up to $100 billion in OpenAI to deploy 10 gigawatts of AI data centre infrastructure. The partnership involves OpenAI deploying Nvidia's future Vera Rubin chips with first deployment set for 2026.

Meanwhile, previous reports have indicated that OpenAI is also working on its own in-house AI silicon in collaboration with Broadcom. The rise of large language models (LLMs) had led to an increasing demand for specialized AI chips to handle the massive computational needs of training and running these models.