Microsoft and OpenAI just restructured one of the most consequential deals in AI, and the terms reshape who can sell GPT models, who pays whom, and what happens if anyone declares artificial general intelligence. The Verge AI broke down the new contract this week, and the headline change is simple: OpenAI can now ship its products and services across every major cloud, not just Azure. A day after Microsoft announced the update, OpenAI confirmed it’s bringing its latest models, Codex, and other tools to AWS, Microsoft’s biggest cloud rival.
This is significant because exclusivity was the entire reason Azure had a moat in enterprise AI. That moat is gone.
What changed
- Cloud exclusivity is dead. OpenAI models can now ship to AWS, and likely Google’s Gemini Enterprise Agent Platform next. Microsoft still gets new models “first” on Azure, but rivals can have them minutes later.
- Revenue share flipped. Microsoft keeps a 20 percent cut of OpenAI’s ChatGPT and API revenue, including money OpenAI earns from rival clouds. But Microsoft no longer pays 20 percent of its Azure OpenAI revenue back to OpenAI.
- Search ad clause loosened. Microsoft stops paying a revenue share if its search and news ad revenue grows 15 percent year over year.
- Term extended to 2032. Microsoft’s non-exclusive license for OpenAI models now runs two extra years past the original 2030 cutoff.
- Microsoft still owns ~27 percent of OpenAI’s for-profit arm, keeping it a major shareholder regardless of how the partnership evolves.
The AGI clause is gone
The biggest unlock here isn’t commercial, it’s contractual. The Verge AI reports that Microsoft’s access to OpenAI models was previously tied to a clause about artificial general intelligence. Once AGI was “theoretically reached,” Microsoft would lose access to OpenAI’s most advanced models. That clause is now removed.
What stands out is the incentive flip. OpenAI no longer has any reason to declare AGI as a contract escape hatch, and Microsoft no longer has to fear getting locked out of frontier models. Both sides had been fighting over that clause for more than a year, with tensions spiking around OpenAI’s reported $3 billion Windsurf acquisition attempt and its for-profit restructuring.
Why this matters for the industry
For enterprises, OpenAI on AWS Bedrock means a lot of Fortune 500 buyers who already standardized on Amazon’s cloud no longer have to spin up Azure to use GPT. OpenAI itself told employees, per The Verge AI, that the old Microsoft deal “limited our ability to meet enterprises where they are, for many that’s Bedrock.”
For Microsoft, the calculus is mixed. Azure OpenAI Service loses its exclusivity advantage, which hurts Foundry and the broader Microsoft AI strategy. But Microsoft now collects revenue from OpenAI’s deals on every cloud, including AWS. It’s a hedge: if OpenAI wins on Amazon, Microsoft still wins.
For Amazon, this validates the $50 billion OpenAI deal announced earlier this year. AWS finally has frontier models on its native platform, which closes a real gap in its enterprise AI pitch.
What Microsoft is doing in parallel
The Verge AI notes that Microsoft is still quietly building models to reduce its OpenAI dependency. MAI-Transcribe-1, a 25-language speech-to-text model, is one early result. Most Microsoft AI models so far have been task-specific or smaller than OpenAI’s frontier LLMs, and to fill the gap Microsoft is leaning more on Anthropic.
Mustafa Suleyman, the DeepMind cofounder Satya Nadella hired in 2024 to run Microsoft AI, is the visible face of that strategy. Tensions between Suleyman’s team and OpenAI got loud, including a reported incident where the Microsoft AI chief yelled at former OpenAI CTO Mira Murati demanding access to code.
What to watch next
Expect OpenAI models to land on more clouds quickly. Watch Google Cloud, watch how Microsoft markets Azure OpenAI without the exclusivity hook, and watch whether Microsoft’s own model lineup gets aggressive about replacing GPT inside Copilot.
Full breakdown is over at The Verge AI.