Court filings: Microsoft feared OpenAI would defect to Amazon

The interesting part of this story is not that Microsoft worried about OpenAI. Of course it did. The real reveal is how early that anxiety started — and how specific it was. Internal emails shown in the Musk v. OpenAI trial lay out a partnership that was never built on pure trust. It was built on fear, leverage, and the belief that letting a rival win would be worse. ### What came out in court? New exhibits in Elon Musk’s case against OpenAI, Sam Altman, and Microsoft surfaced internal Microsoft messages from 2018, when the companies were still feeling each other out. The documents showed senior Microsoft leaders questioning OpenAI’s plans, Altman’s asks, and the economics of a deal — while also worrying that pushing too hard could send OpenAI straight to Amazon Web Services. (wired.com) ### Why was Amazon the nightmare scenario? Because this was never just about one startup. In 2018, Azure was still chasing AWS, and OpenAI looked like a potentially strategic customer and showcase account. One Microsoft executive warned that OpenAI could “storm off to AWS in a huff and shit-talk us publicly,” which is unusually blunt language but also very clear — the fear was commercial loss plus reputational damage. If OpenAI became a loud AWS customer, that could make Azure look second-rate in the most important computing race in tech. (wired.com) ### Why was Microsoft skeptical in the first place? Because OpenAI was asking for a lot, and Microsoft was not yet convinced it would get enough back. The filings describe concern about the scale of compute support OpenAI wanted for projects like Dota 2, plus doubt about whether the partnership would produce meaningful Azure revenue. There was also ideological discomfort inside Microsoft once OpenAI started moving toward a commercial structure. (theverge.com) Kevin Scott, Microsoft’s CTO, questioned whether OpenAI’s donors understood that an “open” effort might become a closed, for-profit one. ### So why do the deal anyway? Because strategic markets work like that. You do not need to love the counterparty. You need to fear the alternative more. Microsoft could see that advanced AI models might become a cloud magnet — the kind of workload that pulls developers, startups, and enterprise customers onto a platform. Losing OpenAI to Amazon or Google would have meant giving a rival both the compute demand and the prestige. (theverge.com) ### Why does this matter now? Because the old fear basically came true in softened form. On April 27, 2026, Microsoft and OpenAI rewrote their partnership so OpenAI could offer products on other clouds and Microsoft’s license became non-exclusive through 2032. On April 28, Amazon said Bedrock would offer OpenAI models, Codex, and OpenAI-powered agent tooling. What Microsoft feared in 2018 stopped being hypothetical last week. (wired.com) ### Does this mean the alliance is breaking? Not exactly. Microsoft is still OpenAI’s major shareholder and primary cloud partner, and OpenAI products still ship first on Azure in many cases. But the relationship is now visibly more transactional. Basically, exclusivity gave way to managed interdependence. OpenAI wants distribution and compute freedom. Microsoft wants continued access to models, product integration, and a giant customer that still spends heavily on Azure. (geekwire.com) ### What does the trial add? It strips away the mythology. The public version of the Microsoft-OpenAI partnership often sounded inevitable — visionary founders, bold capital, shared mission. The private version looks more like a tense corporate alliance where each side worried the other might defect, overreach, or change the rules. Turns out that was true almost from the start. (geekwire.com) ### Bottom line? The filings matter because they show the AI boom’s defining partnership was never a clean marriage. It was a high-stakes hedge — and now that OpenAI can sell on AWS too, the original Microsoft fear looks less paranoid than prescient. (theverge.com) (wired.com)