Microsoft emails show cloud fragility

Cloud deals are supposed to look sticky. Big commitments, custom infrastructure, years of integration — the whole point is to make switching hard. But the Microsoft-OpenAI emails that surfaced this week show something simpler was driving the relationship early on: fear. Microsoft wanted OpenAI badly, and not just because of the models. It worried that losing OpenAI to Amazon would damage Azure’s credibility right when AI compute was becoming strategic. ### What came out this week? Emails and internal documents entered into evidence in the Musk v. Altman trial opened a rare window into 2018, when Microsoft was still deciding how seriously to take OpenAI. Some executives were skeptical of OpenAI’s timelines and claims. But they were also plainly worried that if Microsoft didn’t move, OpenAI could end up on AWS instead. (wired.com) ### Why was Amazon the real fear? Because this was never just a startup investment. Azure was chasing AWS, and OpenAI’s workloads were exactly the kind that could signal which cloud was best for frontier AI. One Microsoft executive warned that OpenAI might “storm off to Amazon in a huff and shit-talk us for years.” Crude wording, but the logic was clear — if elite AI researchers picked AWS, that choice would market Amazon’s cloud for them. (wired.com) ### Was Microsoft actually sold on OpenAI? Not fully. That’s what makes the emails interesting. The company seems to have had two ideas in its head at once: OpenAI might be overpromising, and OpenAI might still be too dangerous to lose. That is a very normal big-tech calculation, basically — you don’t need total belief in a partner’s vision if you believe the downside of letting a rival win is worse. (theverge.com) ### So why push exclusivity? Because exclusivity solved two problems at once. It gave Microsoft a shot at upside if OpenAI’s models turned into a platform, and it denied that upside to Amazon and Google. In cloud, defensive strategy often matters as much as product love. Locking in the hottest AI lab was a way to protect Azure’s position, not just monetize OpenAI. That’s the subtext running through the newly public messages. (wired.com) ### What changed after that? The twist is that the thing Microsoft feared eventually happened in softer form. Recent reporting says OpenAI’s arrangement with Microsoft has loosened enough for OpenAI models to appear on AWS Bedrock, ending the old clean story where Azure was the exclusive home for OpenAI’s best models. Even if the partnership remains deep, the exclusivity wall is not what it used to be. (wired.com) ### Why does that matter beyond these two companies? Because it shows how fragile “cloud loyalty” really is in the AI era. Customers want optionality. Model companies want leverage. Clouds want marquee workloads, but they also don’t want to become hostage to a single supplier. The result is a market that keeps drifting toward portability, even after years of expensive lock-in rhetoric. (forbes.com) ### Is this just an OpenAI story? Not really. It’s a preview of how future AI infrastructure deals will probably work. The hottest model builders will try to avoid being trapped on one cloud. The biggest clouds will still chase exclusives, but those exclusives may expire, narrow, or get renegotiated once bargaining power shifts. Turns out the hardest thing to lock in is not compute — it’s the relationship. (theverge.com) ### Bottom line? The leaked emails don’t just add gossip to the Microsoft-OpenAI saga. They expose the real pressure underneath cloud partnerships: everyone talks like infrastructure is permanent, but everyone negotiates like it can move. And now, in at least one of the industry’s most important alliances, it has. (wired.com)