Here’s how the new Microsoft and OpenAI deal breaks down

Microsoft’s relationship with OpenAI has always been complicated, so I expected the close partnership-turned-situationship to end in tears. After all, executive disagreements, rearranged contracts, and frustrations over AI infrastructure have all regularly been part of the partnership, creating plenty of tension along the way. But against all odds, Microsoft and OpenAI divorced this week in a way that looks strangely amicable.

Microsoft announced the updates to its long-standing OpenAI deal on Monday, with the most important change allowing OpenAI to make its products and services available across all cloud providers. A day later, OpenAI announced it was bringing its latest AI models, Codex, and other tools to AWS — Microsoft’s biggest cloud rival.

It’s a move that seemed likely ever since Amazon announced its $50 billion deal with OpenAI earlier this year, which was originally framed as a deal to make AWS a third-party provider of OpenAI Frontier, together with some collaborative work on custom models for Alexa. Microsoft wasn’t happy about OpenAI cutting a deal with its main rival, and it was weighing legal action ahead of this renegotiation.

Amazon and OpenAI clearly wanted a closer partnership, especially as Amazon is playing catch-up in the AI market and trying to push agents to businesses through its Bedrock service. OpenAI told its employees earlier this month that its deal with Microsoft “has also limited our ability to meet enterprises where they are — for many that’s Bedrock.” Ouch.

As much as Microsoft will hate seeing OpenAI models being available to Amazon, it now stands to benefit from OpenAI cozying up to its rival. Sources tell me that as part of the amended agreement, Microsoft will continue to receive 20 percent of the revenue OpenAI earns for ChatGPT and the AI startup’s API platform. This also includes a cut of any revenue OpenAI earns from putting its products and services on rival cloud platforms like AWS. There’s a revenue cap, but this seems mutually beneficial for both Microsoft and OpenAI even if the Amazon deal originally angered Microsoft.

Microsoft will remain “OpenAI’s primary cloud partner” and ship OpenAI’s latest products first on Azure, but the lack of exclusivity now means OpenAI’s new models can ship minutes later to rival cloud platforms. I wouldn’t be surprised to see OpenAI’s models also appear on Google’s Gemini Enterprise Agent Platform service in the future.

That gives Microsoft less of an advantage over its rivals, particularly for its Azure OpenAI service and its Foundry efforts. Microsoft’s non-exclusive license for OpenAI models and products will now last through 2032, a two-year extension over the previous 2030 cutoff date.

The revenue sharing is also one way now, so Microsoft no longer has to pay 20 percent of its Azure OpenAI revenue back to OpenAI. This new deal also means Microsoft won’t be paying a revenue share to OpenAI if its search and news advertising revenue grows by 15 percent year over year. Microsoft still owns around 27 percent of OpenAI’s for-profit arm, too, making it a major shareholder in OpenAI’s future growth.

This amicable split but not-a-split has only been made possible because Microsoft’s access to OpenAI models is no longer tied to a clause about artificial general intelligence (AGI), something that has long dictated OpenAI and Microsoft’s partnership. The removal of the AGI clause means Microsoft will no longer lose access to OpenAI’s most advanced future models once AGI is theoretically reached. And on the flip side, OpenAI is no longer incentivized to declare AGI to get out of the deal.

The pair had been arguing about the AGI clause for well over a year, and things got particularly heated when OpenAI was considering a $3 billion deal to buy AI coding tool Windsurf last year. OpenAI wanted Windsurf to be exempt from the deal with Microsoft, and even reportedly considered accusing Microsoft of anticompetitive behavior. OpenAI eventually backed down on its Windsurf acquisition plans, but sources at Microsoft tell me it led to months of tense negotiations over OpenAI’s controversial for-profit restructuring last year.

I understand Microsoft first became wary of OpenAI as a partner when Sam Altman was ousted as CEO in 2023 by the nonprofit’s board of directors. This wariness led to Microsoft CEO Satya Nadella recruiting DeepMind cofounder Mustafa Suleyman to lead its Microsoft AI business. Altman was reportedly “angered” that Suleyman had joined Microsoft, especially as Suleyman’s team started dealing directly with OpenAI engineers.

The tension between Suleyman’s teams and OpenAI came to a boiling point when Microsoft’s AI chief yelled at former OpenAI CTO Mira Murati during a meeting in 2024, reportedly demanding access to OpenAI code. While Microsoft cofounder Bill Gates, CTO Kevin Scott, and Nadella had all delicately brought the OpenAI deal to life, Suleyman was hired to shake things up and create Microsoft’s own AI models so that in the future it’s less reliant on OpenAI.

Parts of these deal changes also seem designed to distance Microsoft from the chaos of OpenAI and Sam Altman. OpenAI is increasingly juggling public controversies, strategy shifts, and a slew of executive reshufflings. It certainly feels like the vibes are off, and the contract update comes in the same week that Elon Musk and Altman are in court battling over the future of the AI company. Microsoft has already started getting dragged into this drama.

Away from the courtroom antics of Musk and Altman, Microsoft is still quietly trying to build models good enough to replace OpenAI’s. It’s had some impressive results with models like MAI-Transcribe-1, a speech-to-text model that works across 25 different languages. But most of Microsoft’s own AI models have been very task specific or smaller in size compared to OpenAI’s large language models.

To make up for this gap, Microsoft is increasingly turning its attention to options from Anthropic that are more powerful than OpenAI’s models for certain tasks. Sources at Microsoft tell me the company has even considered using Google’s Gemini models for certain products, just like how it rapidly started trialing Anthropic’s Claude Code.

Microsoft is willing to use the best AI model for the job, in an effort to keep businesses hooked to its enterprise software offerings and fend off competition from AI startups and even OpenAI itself. Some of the tension throughout Microsoft and OpenAI’s partnership has been related to the collision course the pair have both been on, each trying to convince the same mutual customers to purchase their AI services.

All this points toward Microsoft’s AI future as a mix of models developed internally and by OpenAI, Anthropic, and others. It leaves Microsoft and OpenAI with a partnership that’s a lot more about finances than close collaboration on the future of AI. They’ll still be closely linked, but the partnership-turned-situationship is looking a lot less complicated than before.

I’m always keen to hear from readers, so please drop a comment here, or you can reach me at notepad@theverge.com if you want to discuss anything else. If you’ve heard about any of Microsoft’s secret projects, you can reach me via email at notepad@theverge.com or speak to me confidentially on the Signal messaging app, where I’m tomwarren.01. I’m also tomwarren on Telegram, if you’d prefer to chat there.

Thanks for subscribing to Notepad.