OpenAI to Deploy 6 GW with AMD, Could Gain Nearly 10%

OpenAI has chosen AMD as a strategic supplier to deploy six gigawatts of GPU capacity for artificial intelligence over the coming years. The deal starts with one gigawatt of Instinct MI450 accelerators in the second half of 2026 and includes an extraordinary incentive: OpenAI will be able to acquire up to 160 million AMD shares—nearly 10% of the company—if it hits the agreed milestones.

The deal opens up enormous growth potential for AMD and reduces OpenAI’s reliance on Nvidia, but it does not represent an immediate replacement for its main supplier. It also reinforces an increasingly visible financing model in the AI race: chipmakers and customers are combining chip purchases, investments and equity stakes to fund infrastructure on an unprecedented scale.

Six gigawatts, but not all at once

The contract spans multiple years and several generations of AMD Instinct accelerators. The first phase covers one gigawatt of MI450 systems beginning in the second half of 2026. The agreement calls for deployment of up to six gigawatts as the partnership expands to future product generations.

Gigawatts indicate the power scale associated with the announced GPU deployment, not an exact processor count. AMD also refers to rack-scale AI solutions, but it has not publicly detailed how capacity will be distributed among the components. A single gigawatt already puts the project in the category of the largest AI infrastructure deployments announced to date.

AMD expects the deal to generate tens of billions of dollars in revenue, according to the company. That figure is a deployment-dependent forecast, not the firm value of a purchase completed today. No specific date has been set for completing the six gigawatts, either.

The work between the two companies is not starting from scratch. AMD says its hardware and software collaboration with OpenAI began with the MI300X and continued with the MI350X series, and the companies will now work together to adapt both hardware and software for future generations. That part is crucial: Nvidia’s advantage comes not only from its chips but also from CUDA, its widely adopted programming platform. AMD needs ROCm, its alternative, to prove reliable and easy to use at scale.

A right to acquire 160 million shares

To align their interests, AMD has granted OpenAI a warrant, a contract that allows it to buy shares in the future under specified conditions. The warrant covers up to 160 million AMD shares, equivalent to roughly 10% of the company’s outstanding stock if fully exercised, at a nominal exercise price of one cent per share.

OpenAI is not receiving that stake immediately. The first tranche will vest with the initial one-gigawatt deployment. Subsequent tranches will depend on purchases scaling up to six gigawatts, OpenAI meeting technical and commercial targets, and AMD’s share price reaching certain levels. The final tranche is tied to the stock reaching $600.

The incentive could be worth an enormous amount if the plan succeeds, but it would also dilute existing shareholders as new shares are issued. The market initially rewarded the prospect of AMD landing one of the world’s biggest chip customers: the stock surged by around 34% on Monday.

AMD gains a reference customer against Nvidia’s dominance

Nvidia remains the dominant player in the accelerators used to train and run large models. OpenAI, Microsoft, Meta and other operators have built much of their infrastructure around its GPUs and CUDA. The AMD deal does not, by itself, break that dominance, but it gives the second-largest chipmaker a reference customer that will be difficult to match.

For OpenAI, adding another supplier helps expand its supply, negotiate prices and reduce the risk of depending on a single architecture. The company needs capacity to train new models and handle growing ChatGPT usage—a level of demand that can no longer be met simply by buying more units from the same chip family.

The deal also coexists with the plan Nvidia and OpenAI announced in September: deploying at least 10 gigawatts of Nvidia systems, alongside a potential investment of up to $100 billion by the chipmaker in OpenAI as installation progresses. The AMD relationship takes the opposite approach: here, the supplier is giving the customer rights to its own equity as an incentive to buy and deploy its products.

A race financed through cross-linked relationships

These deals raise an important question for investors and regulators. Some of the growth chipmakers expect comes from customers receiving financing or equity stakes tied to their purchases. That does not automatically make the revenue artificial, because the data centers and chips still have to be built and brought online, but it makes it harder to determine who is actually financing demand and how much risk each participant is assuming.

The first tangible test will come in the second half of 2026. AMD will have to deliver the MI450 chips and its systems at gigawatt scale; OpenAI will have to integrate them without losing performance relative to Nvidia’s ecosystem. Only then will it become clear whether the deal creates a second AI infrastructure hub or whether its main effect has been to raise expectations for AMD.