- OpenAI can now serve products across any cloud provider, ending Microsoft’s Azure exclusivity.
- The controversial AGI clause is dead—an independent panel will declare AGI, not OpenAI itself.
- Microsoft retains model IP through 2032 but lost right of first refusal and will stop paying OpenAI revenue share.
OpenAI and Microsoft announced a new definitive agreement on April 27 that tears up the exclusive cloud arrangement governing their seven-year partnership. OpenAI can now deploy its products on AWS, Google Cloud, or any other infrastructure provider—not just Azure. In return, OpenAI committed to spending an additional $250 billion on Azure services, according to details reported by CNBC. The deal restructures a relationship that had grown increasingly tense as OpenAI pushed for more independence and Microsoft sought to protect its $13 billion investment.
The changes come the same day jury selection began in Elon Musk’s lawsuit against OpenAI, and one week after Amazon announced a $50 billion OpenAI cloud deal on AWS. The timing is not subtle. OpenAI has been building toward a public benefit corporation conversion and eventual IPO, and the old exclusivity terms—particularly Microsoft’s IP rights extending until OpenAI achieved “AGI”—were a liability for both the restructuring and future partnerships. Microsoft’s stake in the new PBC will be roughly 27%, valued at approximately $135 billion, down from 32.5% under the old structure.
“We’ve signed a new definitive agreement that builds on our foundation, strengthens our partnership, and sets the stage for long-term success for both organizations,” the companies said in a joint statement. Brad Smith, vice chair and president of Microsoft, wrote that the partnership “is entering its most productive and impactful phase yet.” The phrasing papers over months of reported friction over compute pricing, revenue splits, and OpenAI’s desire to work with other cloud giants.
What the New Microsoft-OpenAI Deal Actually Changes
The biggest structural shift is the death of the AGI clause. Under the old agreement, Microsoft held exclusive IP rights to OpenAI’s models and products until OpenAI achieved “artificial general intelligence”—a term OpenAI itself got to define. That created a perverse incentive: declaring AGI would have terminated Microsoft’s IP rights, while never declaring it kept those rights locked up indefinitely. Under the new deal, AGI will be declared by an independent expert panel, and Microsoft’s IP license—now non-exclusive rather than exclusive—runs through 2032 regardless of when or whether AGI arrives. Microsoft also retains IP rights to OpenAI research through 2030, and gets access to post-AGI models with what the companies call “appropriate safety guardrails.” As Simon Willison noted, tracking the history of the AGI clause reveals how contested that definition has been since the partnership’s earliest days.
The revenue share structure has also been overhauled. Microsoft will no longer pay OpenAI a revenue share—a significant shift, given that CNBC previously reported Microsoft was receiving roughly 20% of OpenAI’s revenue. OpenAI’s payments to Microsoft continue through 2030 at the same percentage, but are now subject to a total cap and spread over a longer period. Microsoft also lost its right of first refusal to be OpenAI’s compute provider—meaning OpenAI can shop for infrastructure without offering Microsoft first dibs. The one exception is joint API products developed with third parties, which remain exclusive to Azure.
For Microsoft, the concessions come with real upside: $250 billion in committed Azure spending, which substantially de-risks its infrastructure investment, plus continued access to OpenAI’s frontier models. Microsoft can now also pursue AGI independently, something it was effectively already doing under the banner of “superintelligence.” If Microsoft uses OpenAI IP to develop AGI, those models will be subject to compute thresholds “significantly larger than the size of systems used to train leading models today,” per the agreement. Microsoft also loses IP rights to any consumer hardware OpenAI releases—relevant given analyst Ming-Chi Kuo’s report that OpenAI is working with Qualcomm and MediaTek on custom smartphone chips.
Why OpenAI Needed Out—And What It Cost
OpenAI’s push for flexibility has been building for months. The $50 billion Amazon deal announced last week would have been impossible under the old exclusivity terms. Amazon CEO Andy Jassy confirmed that OpenAI models will be available on Amazon Bedrock “within weeks.” Google Cloud access is likely next—9to5Google reported that the deal explicitly permits OpenAI to use Google as a cloud provider. The multi-cloud shift lets OpenAI negotiate better compute pricing and reduces its dependence on a single infrastructure partner at a time when training runs for frontier models cost hundreds of millions of dollars each.
The restructured deal also clears a path for OpenAI’s conversion to a public benefit corporation and eventual IPO. The old AGI clause—a provision that gave Microsoft indefinite IP control tied to an undefined milestone—was a dealbreaker for potential public market investors who need predictable IP arrangements. With the clause replaced by a fixed 2032 non-exclusive license, OpenAI’s cap table and IP obligations are now legible to institutional investors. Microsoft’s endorsement of the PBC conversion, announced alongside the deal terms, removes another obstacle. The question now is whether the $250 billion Azure commitment—roughly ten times Microsoft’s original $13 billion investment—will look like a bargain or an anchor if OpenAI’s compute costs keep scaling at current rates.
Microsoft stock fell roughly 1% on the day of the announcement. The Musk v. Altman trial continues in Oakland federal court.
