OpenAI just made AMD a viable counterweight to Nvidia for GPUs. OpenAI said it has inked a 6 gigawatt deal with AMD to provide AMD Instinct GPUs for its AI buildout.
Under the terms of the deal, OpenAI's first gigawatt deployment of AMD Instinct MI450 GPUs starts in the second half of 2026. OpenAI will also get a warrant to acquire up to 160 million shares of AMD that will vest as milestones are reached. The first tranche vests when the first gigawatt is deployed with additional vesting as OpenAI builds to 6 gigawatts.
OpenAI also said the there is more vesting tied to AMD stock price targets and the ability to deploy Instinct GPUs at scale.
The OpenAI-AMD deal is different than the OpenAI-Nvidia deal. Nvidia invested in OpenAI to help fund purchases of GPUs and AI infrastructure. The OpenAI-AMD deal doesn't provide cash up front, but aligns interest.
AMD CEO Dr. Lisa Su said the partnership will create "a true win-win enabling the world’s most ambitious AI buildout and advancing the entire AI ecosystem." OpenAI CEO Sam Altman said the AMD deal gives it the ability to accelerate its plans.
Constellation Research analyst Holger Mueller said:
"OpenAI is signing deals left and right. The AMD deal is different, as it is the first explicit and only inference deal, as well as equity deal the AI vendor has stuck. It's a big win for AMD that could not get to scale in the data center for AI yet, the big question is now which data center vendor will get the workload. The equity aspect is also interesting. AMD is giving up a lot here for an initial deal. It is also clear that the OpenAI leadership is scared from compute capacity challenges and wants to avoid them at all cost. The concern is that it's unclear how will OpenAI be able to pay for all the performance obligations. We'll worry about tomorrow when it's tomorrow."
Mueller isn't kidding about the questions about payment. OpenAI's deals of late via Goldman Sachs include:
- OpenAI to buy 6GW of AMD chips.
- OpenAI and CoreWeave expand partnership to $22.5 billion.
- OpenAI will spend $85 billion a year over the next five years on server rentals and $100 billion total on backup servers in 5 years.
- OpenAI plans to spend $115 billion on AI capital expenditures through 2029.
- OpenAI is leading the $500 billion Stargate project. Stargate ramps as OpenAI, Oracle, Softbank outline 5 new data centers | Oracle Q1 misses, but sees OCI revenue surging over next 4 years
That spending is against OpenAI's 2025 revenue target of $13 billion, according to The Information.
Clearly, the OpenAI deal is huge for AMD, which now has solidified itself as a viable second option for GPUs. The OpenAI-AMD deal is also big for anyone procuring AI compute.
Here's why:
- Nvidia has had a lock on the AI infrastructure market and those nice profit margins are being funded by IT buyers. Enterprises have been waiting for two years to see Nvidia competition.
- As more AI infrastructure is deployed on-premises and at the edge, AMD will be a natural option for enterprises.
- AMD will land more tier-1 customers and cloud instances.
- The constrained GPU market will be less constrained with AMD and cloud hyperscalers' custom chips adding competition.
- AMD is likely to land more deals with AI-centric cloud providers.
- AMD's ROCm platform will be more viable against the Nvidia software stack, which is where the lock-in will really occur.
- Nvidia still has the installed base and dominance, but will arguably face its first real competition in the market.
