Microsoft and OpenAI just rewrote their deal. Here’s what actually changed.

Microsoft and OpenAI have quietly updated their partnership agreement, and while the press release is light on specifics, the signal is pretty clear: both sides wanted to clean up the messiness that had crept into their relationship over the past few years.

Let’s be honest — the original deal was always a bit of a patchwork. Investment rounds, compute commitments, exclusivity clauses, governance seats. It worked when things were simple, but as OpenAI grew from a scrappy research lab into a full-blown product company, the friction points multiplied. Microsoft wanted clearer terms on how much compute it could access and what exclusivity really meant. OpenAI wanted more freedom to shop around for cloud capacity and maybe not have every API call routed through Azure.

This amended agreement addresses some of that directly. The key changes, as I read them:

First, the compute arrangement gets simplified. Instead of negotiating capacity in lumpy chunks, Microsoft has committed to a more predictable, long-term compute allocation. That matters because training and inference costs aren’t going down anytime soon, and OpenAI needs to know it can scale without hitting a ceiling.

Second, the revenue share model got an overhaul. The old structure had multiple tiers and thresholds that made it hard to figure out who owed whom what. The new terms are supposedly cleaner, though neither side is publishing numbers. I’d guess Microsoft gets a more favorable cut on commercial API sales, while OpenAI gets to keep more of its direct consumer revenue from ChatGPT subscriptions.

Third — and this is the part that will get attention — OpenAI is no longer exclusively tied to Azure for all its compute. Microsoft retains a “right of first refusal” on new capacity, but if they can’t deliver, OpenAI can go elsewhere. This is a big shift. It acknowledges that no single cloud provider can keep up with the demand spikes that come with GPT-scale model training.

Does this mean OpenAI is shopping around? Probably not immediately. But it’s a hedge. And honestly, it’s smart. AWS and Google Cloud have been circling for years, and this gives OpenAI leverage without burning the bridge with Microsoft.

What hasn’t changed? Microsoft still has a seat on OpenAI’s board, though it’s now a non-voting observer role instead of a full voting member. That’s a downgrade in formal power, but in practice, the two companies are still deeply intertwined. Satya Nadella and Sam Altman talk regularly. The strategic alignment isn’t going anywhere.

The bigger question nobody is asking: does this partnership still make sense long-term? Microsoft is building its own AI models under the Copilot brand. OpenAI is increasingly competing with Microsoft’s customers by selling directly to enterprises. The tension was always there, but now it’s out in the open. This amended agreement buys them another few years of cooperation, but I wouldn’t be surprised if we see a full separation down the road — maybe with Microsoft taking a minority stake and OpenAI going fully independent.

For now, though, the deal is done. Both sides get more clarity, more flexibility, and a cleaner path forward. That’s about the best you can ask for in a partnership this complex.