How OpenAI’s AMD bet raises the stakes in GPU dominance
The OpenAI partnership with AMD gives it the option to invest directly in the chipmaker
Just weeks after it signed a landmark 10 gigawatts (GW) of datacentre capacity agreement with Nvidia, OpenAI has penned a deal with rival chipmaker AMD.
Tied to a commitment to an initial purchase of 1GW AMD Instinct MI450 Series graphics processor unit (GPU) products, an alternative to Nvidia’s Blackwell family, the AMD deal gives OpenAI share options, which could result in the generative artificial intelligence (GenAI) provider opting to take a 10% stake in AMD, if it purchases a full 6 GW of AMD GPU power.
Through the partnership, AMD and OpenAI said they would build out the infrastructure to meet growing AI demands.
AMD CFO and treasurer Jean Hu said: “Our partnership with OpenAI is expected to deliver tens of billions of dollars in revenue for AMD while accelerating OpenAI’s AI infrastructure buildout. This agreement creates significant strategic alignment and shareholder value for both AMD and OpenAI.”
OpenAI recently signed a GPU deal with Nvidia. At the time, Greg Brockman, co-founder and president of OpenAI, said: “We’ve utilised their [Nvidia] platform to create AI systems that hundreds of millions of people use every day. We’re excited to deploy 10 gigawatts of compute with Nvidia to push back the frontier of intelligence and scale the benefits of this technology to everyone.”
Commenting on the AMD deal, Brockman said: “Building the future of AI requires deep collaboration across every layer of the stack. Working alongside AMD will allow us to scale to deliver AI tools that benefit people everywhere.”
OpenAI CEO Sam Altman praised AMD’s technology, adding: “This partnership is a major step in building the compute capacity needed to realise AI’s full potential. AMD’s leadership in high-performance chips will enable us to accelerate progress and bring the benefits of advanced AI to everyone faster.”
Yet at the time of the Nvidia deal, he said: “Compute infrastructure will be the basis for the economy of the future, and we will utilise what we’re building with Nvidia to both create new AI breakthroughs and empower people and businesses with them at scale.”
What is clear from these comments is that OpenAI needs to build out its AI infrastructure and will work both with Nvidia and AMD to achieve this objective.
While the deal with Nvidia represents a significantly larger commitment, what OpenAI’s partnership with AMD shows is that the chipmaker’s big AI splash last year is paying dividends.
AMD cements strategy
In July, Forrester senior analyst Alvin Nguyen looked at how the company was setting up for an AI inference revolution in 2028. During AMD’s Advancing AI 2025 event in July this year, Nguyen said the company publicly benchmarked performance against both its previous generation (MI300A/X) and Nvidia’s GB200 and B200 GPUs.
According to Nguyen, this signals AMD is confident in its AI architecture and execution. He also noted that the release of the company’s latest Rocm 7 open source platform shows AMD’s continued investment in open ecosystems and reflects its commitment to enabling scalable, enterprise-grade AI.
The latest PassMark server processor benchmark is dominated by AMD server chips both from an outright performance perspective and price/performance against rival Intel. The company wants to replicate this success in the GPU market to take advantage of the massive growth potential it can achieve by offering AI acceleration hardware.
However, an analysis from SemiAnalysis shows that while the total cost of ownership of high-end AMD GPUs is lower than similar Nvidia GPUs, IT buyers selecting AMD AI acceleration hardware are likely to see higher rental prices from specialist cloud providers, mainly due to easier availability of Nvidia hardware.
An injection of cash from OpenAI could help to improve the availability of AMD GPUs, which would not only benefit those organisations planning to build-out on-premise AI infrastructure, but also potentially lower AMD hardware costs the specialist AI infrastructures need to pay. This, in turn, could reduce GPU hosting and GPU cloud-based rental prices for IT buyers.
