Monday’s announcement that OpenAI and AMD have struck a deal, albeit an unusual one without cash commitments, could mean that AMD chips may become a viable enterprise IT option. That is good news, not because of AMD quality, which is seen as suboptimal by some, but because of the limits of Nvidia chip availability.
The Monday announcement simply said that the two companies would work together and that they had crafted “a 6 gigawatt agreement to power OpenAI’s next-generation AI infrastructure across multiple generations of AMD Instinct GPUs. The first one gigawatt deployment of AMD Instinct MI450 GPUs is set to begin in the second half of 2026.”
That likely means anywhere from 3.5 million to 5 million chips, according to Moor Insights & Strategy. “AMD is now able to seed the market with a lot of its GPUs,” said Matt Kimball, a Moor VP and principal analyst.
NVIDIA supply limits a big factor
Under other circumstances, that OpenAI endorsement might not mean much, but enterprise IT executives are finding it increasingly difficult to purchase GPUs from Nvidia, so this gives them a critically needed second source for those chips.
An AMD spokesperson, Drew Symonds of AMD corporate communications, told Network World in an email that “OpenAI is purchasing the GPUs” but couldn’t specify the amount or whether there was a direct payment in cash.
“Best I can refer you to about revenue expectations is a quote in our press release from Jean Hu, CFO, AMD. ‘Our partnership with OpenAI is expected to deliver tens of billions of dollars in revenue for AMD while accelerating OpenAI’s AI infrastructure buildout,’” Symonds wrote.
But that doesn’t specify that the dollars referenced would come from OpenAI. Others have interpreted the remark as referring to potential increased revenue from companies buying from AMD because of the OpenAI endorsement, and possibly because of OpenAI improvements to AMD’s software. Symonds did not respond to that question.
Rodolfo Rosini, CEO of Vaire Computing, said the supply problems with Nvidia are absolutely a critical background factor for the AMD-OpenAI deal.
“There is unbound demand for Nvidia hardware, but a limited supply, and upstream there is a limited supply of wafers from TSMC to Nvidia,” Rosini said. “So now the demand is overspilling into competing offerings, as AI companies can’t stand still while they wait for an allocation.”
But Rosini also saw some product weaknesses at AMD playing an outsized role.
“AMD’s software stack is bad, but that is a bigger issue for training than for inference. They were always viable for enterprise use. They just could not command premium pricing like Nvidia does, and developers preferred [Nvidia’s] CUDA,” Rosini said. “[OpenAI] directing AMD’s software roadmap instead of the management of AMD will be great. Labs like OpenAI know exactly what they want and will be very vocal about it.”
Chip supplier diversity needed
Another analyst, Jack Gold, principal analyst for J. Gold Associates, agreed.
“This is an indication that OpenAI recognizes a need to diversify its processor suppliers, as it continues to expand its data centers. The most advanced Nvidia GPUs are on allocation, with buildouts outpacing supplies,” Gold said. “By solidifying AMD chip supplies through this commitment and investment, OpenAI can continue its massive build out campaign.”
Gold said that he also expects this to fuel more AMD purchases from enterprises.
“It’s highly likely that other major AI players will follow suit and deploy AMD-powered datacenters, even more so than with the current movement, with AMD GPUs seen as the secondary supplier,” Gold said.
However, Gold disagreed with Rosini’s poor assessment of AMD software. “AMD software is not all that terrible,” he said, noting that the problem is the popularity of CUDA. He said, “you can’t just take CUDA and put it on an AMD chip,” and that means that OpenAI will have to write “some level of abstraction.” He added, “if I am writing that level of abstraction, do I really care what the underlying chip is?”
Gold estimated that the chips being delivered are roughly worth $180 billion, assuming that the deal will likely need about 6 million chips to reach the six gigawatts mentioned in the news release and he sees those chips typically selling for about $30,000 each.
“These guys [enterprise IT executives] are going to look at it seriously, especially if they can’t get Nvidia chips,” Gold said. “An endorsement by OpenAI is worth a lot.”
Moor’s Kimball’s estimate of the number of chips needed for this deal is lower, suggesting it will be “anywhere from 3.5 million to 5 million GPUs.”
He said that this deal might help AMD dig itself out of the AI hole it has found itself in for years.
“AMD has been struggling to capture market share relative to Nvidia, despite a very good architecture. Hardware-wise it is superior,” Kimball said. “It’s been an issue forever. ROCm is its Achilles heel. It is not used and it is not compatible with CUDA.”
“AMD needs to seed the market and it needs to get some adoption out there, and OpenAI is a great vehicle for that,” Kimball said.
OpenAI has the leverage today
What is odd about the agreement is that nothing in the copious amount of detail published — the SEC 8K filing alone has multiple attachments — indicates that OpenAI will be paying any money for these chips, or at least not a specific agreed amount.
Abhishek Singh, a partner at the Everest Group, sees this deal as not being about money. And Singh also sees it as a very smart move for both OpenAI and AMD.
“It is asymmetric, isn’t it? And why wouldn’t it be? OpenAI has the leverage right now. Every chipmaker wants to be part of its supply chain because OpenAI effectively defines the reference workload for AI,” Singh said. “For AMD, this isn’t just about selling chips. It’s about proximity. Getting in early means access to OpenAI’s models, data patterns, and performance feedback that will shape AMD’s roadmap for years. That’s worth more than immediate cash flow.”
Singh added that, in this instance, the revenue wouldn’t be nearly as attractive as the long-term potential benefits.
“OpenAI doesn’t need to part with cash to get that value. The warrant structure is clever: it gives them a financial upside if AMD executes well, and no exposure if it doesn’t. So the money isn’t flowing, because this isn’t a cash-for-silicon deal,” Singh said. “It’s a trade of influence for opportunity. AMD is buying relevance in AI compute, and OpenAI is buying flexibility and optionality for its next growth phase.”
Singh also addressed one of the biggest quiet truths in the GenAI space, which is that OpenAI is publicly committing to spending a lot of money that it doesn’t appear to have. The company is reported to have current annualized revenue of only $8.6 billion and it has already lost $7.8 billion in the first half of 2025.
“It’s fair to say that questions about OpenAI’s cash capacity have become a recurring theme. The warrant structure gives OpenAI optionality rather than obligation. If its cash position and priorities allow, it can exercise those rights. If not, there’s no exposure or liability. It’s a smart construct for both sides,” Singh said. “AMD secures strategic alignment with one of the world’s largest AI compute consumers, while OpenAI gains upside participation without committing cash today.”
Even though OpenAI is receiving the chips, the only potential cashflow goes to OpenAI. OpenAI was given the right to purchase AMD shares, trading at more than $200 on Monday night, for one penny a share. There is a variety of restrictions based on the performance of both companies, but that agreement has the potential to deliver a lot of cash to OpenAI.
Most observers said that AMD was clearly entering into the negotiations with OpenAI as the weaker party. Cashflow notwithstanding, OpenAI has massive momentum within AI, and AMD has relatively little.
However, AMD gets two tremendous benefits from the deal: OpenAI will be improving AMD’s software, mostly its ROCm open software stack, and AMD will benefit from the OpenAI endorsement.
That improved software would also go back to AMD, which could help its current and future customers. More to the point, AMD could use that improved software to try and make a lot more sales.