The revised Microsoft-OpenAI deal teases far more than it delivers

When Microsoft and OpenAI announced their revised AI deal on Thursday, it reduced Microsoft’s ownership share of OpenAI, but seemed to give it everything that it wanted in exchange, analysts said. Well, everything other than AGI, which Microsoft doesn’t really care about.

The announcement of the revised details indicated a change in the nature of the OpenAI business structure, but was murky on many of the underlying details. It said that Microsoft supports the formation of a public benefit corporation (PBC) and recapitalization by the OpenAI board. That restructuring will dilute Microsoft’s ownership from 32.5% to 27%.

OpenAI, which reportedly lost $7.8 billion in the first half of 2025, said that the new Microsoft shares are worth $135 billion.

Scott Bickley, advisory fellow at the Info-Tech Research Group, scoffed at the OpenAI valuation. “This is continuing on this path of crazy financing that defies the laws of physics. The truth is that 27% of nothing is still nothing,” he said.

The revised agreement also requires OpenAI to purchase “an incremental $250 billion of Azure services.” That prompted Michal Prywata, co-founder of AI infrastructure firm Vertus, to look askance at the underlying metrics.

“As for the $250 billion number, this is venture capital math at its most absurd. OpenAI is burning billions quarterly while promising to spend a quarter trillion over multiple years,” Prywata said. “We know the money won’t come from revenue, let alone profit. It’s coming from the investors betting that someone else will pay even more later. Classic.”

Overall, Bickley said the changes in the agreement almost entirely favor Microsoft. “Microsoft held all the cards in terms of the negotiation,” he said. “This is all wins for Microsoft.”

Microsoft blinked because it wanted to

The only area where Bickley saw Microsoft not win involved the ill-defined artificial general intelligence (AGI), and that was mostly because Microsoft simply does not care that much about AGI.

Although the agreement referred to AGI as a trigger event, the Microsoft statement lacked any specifics that would allow IT buyers to evaluate the timetable.

However, AGI was prominent in the list of changes. It began, “once AGI is declared by OpenAI, that declaration will now be verified by an independent expert panel,” but there were no details about who would appoint those panel members, how the firms would establish that independence, the number of panel members, and from what backgrounds. Computerworld asked both OpenAI and Microsoft to clarify, but neither offered any specifics.

Another part of the agreement said, “if Microsoft uses OpenAI’s IP [intellectual property] to develop AGI, prior to AGI being declared, the models will be subject to compute thresholds. Those thresholds are significantly larger than the size of systems used to train leading models today.” But it did not define those thresholds.

The revised deal also said, “the revenue share agreement [in the original partnership] remains until the expert panel verifies AGI, though payments will be made over a longer period of time” and, “Microsoft’s IP rights for both models and products are extended through 2032 and now include models post-AGI, with appropriate safety guardrails.” 

Another timeline would be triggered two years earlier, the agreement said: “Microsoft’s IP rights to research, defined as the confidential methods used in the development of models and systems, will remain until either the expert panel verifies AGI or through 2030, whichever is first.” However, Microsoft explicitly retains the rights to what it calls “non-research IP,” including model architecture and weights, inference and fine tuning code, and “any IP related to data center hardware and software.”

The AGI definition debate

Mike Gualtieri, a VP/principal analyst for Forrester, was intrigued by how much of the deal revisions focused on AGI.

“It’s fascinating that Microsoft and OpenAI partnership evolution is, in part,  based on when AGI is declared. AGI could be three years away or it could be 10 years away. If OpenAI or Microsoft were to achieve AGI, then every enterprise in the world would be clamoring for it. But neither Microsoft nor OpenAI have telegraphed anything other than hype that AGI is near,” Gualtieri said. “I think it is highly likely that AGI won’t be achieved by just one organization, but near simultaneously by other research and commercial groups like Google, xAI, Meta, and Chinese hyper-scalers. OpenAI certainly doesn’t have a lock on LLMs.”

Info-Tech’s Bickley agreed the AGI calendar targets are very soft. Both companies want to delay dealing with AGI until it happens, assuming that it ever does happen.

“They want to kick the can down the road” on AGI, Bickley said. “They are miles away from what either party claims is AGI.”

The details about AGI “might be the one concession that Microsoft made” because “they either see it as so unlikely or so far out in the future” that they are ambivalent. “I don’t think it’s core to Microsoft’s mission at all,” Bickley said. 

Vertus’ Prywata was even more blunt. He said the AGI definition is farcical. 

“It’s corporate theater. Who defines AGI? I know my definition likely won’t match theirs. I bet the team who left OpenAI for Anthropic likely have their own definition. Who picks the independent experts? What are their qualifications?” Prywata asked. “Zero details suggests they have no idea, otherwise they’d mention it. It’s like saying ‘we’ll know it when we see it,’ but with a committee. Turing test 2.0, I suppose. And who put them in charge to decide this for all of us?”

Gualtieri added that his overall take is that this deal is unlikely to make much of a purchasing strategy change among enterprise IT executives. “I’m not awed by this and I don’t think tech leaders at the world’s largest enterprises will be either.”

Many of the revisions involved restrictions, or softening of restrictions, on how the two firms will cooperate, he pointed out. For example, OpenAI can now jointly develop some products with third parties, and API products developed with third parties will be exclusive to Azure while non-API products may be served on any cloud provider, and Microsoft will no longer have a right of first refusal to be OpenAI’s compute provider. Furthermore, OpenAI can now provide API access to US government national security customers, regardless of the cloud provider.

Beginning of the end?

Jason Wong, a distinguished VP analyst at Gartner, said that he saw the contract revisions as the beginning of separation of the two firms. “This positions Microsoft and OpenAI more clearly on the path to independence,” Wong said. 

Info-Tech’s Bickley added that constant industry rumors about an imminent AI bubble burst provide an important background to the revised agreement and the likely impact it will have on enterprise IT AI strategies. 

“The music keeps playing until the bubble bursts. That’s when things get real,” he said, and AI companies will be forced to dramatically increase the prices they charge for AI services. Those prices today are highly subsidized by VCs and other investors. 

“How much can they raise prices and continue to do business? If it rolled out today, you would have a catastrophic implosion” because enterprise AI usage “won’t support the price increase that would support profitability.”

Prywata agreed. “OpenAI needs to generate massive revenue just to service these commitments, let alone turn a profit. They’re betting they can achieve monopolistic pricing power before competitors catch up,” Prywata said. “I’m sorry, but that’s just not happening with all of the brilliance going after AI right now.”