Microsoft and OpenAI Clean Up Their Messy Partnership

The Microsoft-OpenAI relationship has always been a bit of a soap opera. A $13 billion investment, a weirdly structured profit-sharing arrangement, and enough governance drama to fill a season of Succession. Now they’ve finally torn up the old contract and written something that looks a lot more like a normal business deal.

OpenAI announced the amended agreement today, and honestly, it’s about time. The original deal had so many moving parts—exclusivity clauses, complex revenue splits, caps on compute access—that it was a miracle either side could make a move without legal review. This new version strips all that down.

Here’s what actually changes:

Microsoft gets right of first refusal on new OpenAI models. That means when GPT-5 or whatever comes next drops, Microsoft can decide whether to license it before anyone else gets a look. But crucially, it’s no longer exclusive. If Microsoft passes, OpenAI can shop it around. This is a big shift from the old arrangement where Microsoft essentially owned the exclusive distribution rights.

The compute situation is also cleaner now. Microsoft still provides the bulk of OpenAI’s training and inference infrastructure, but OpenAI can now go elsewhere if Microsoft can’t meet demand. Given how often GPU shortages have bottlenecked AI development recently, this flexibility is smart. I’ve heard stories from people inside OpenAI about training runs being delayed because Azure had capacity issues. This change doesn’t fix that overnight, but it gives them an escape hatch.

On the financial side, the profit-sharing cap that limited Microsoft’s returns has been removed. Originally, Microsoft’s share was capped at around $100 billion in total profits—a number that seemed generous until you realize how much capital they’ve poured in. Now it’s a straight revenue share with no ceiling. Microsoft clearly believes OpenAI’s revenue trajectory justifies this. Given that OpenAI is reportedly on pace for $10+ billion in annual revenue, that bet might pay off.

What’s interesting is what didn’t change. Sam Altman remains CEO, the board structure stays the same, and Microsoft still doesn’t have a board seat. That last point has been a sore spot since the November 2023 drama when Altman was briefly fired and reinstated. Microsoft’s CEO Satya Nadella made it clear back then that he wanted a board seat, but he never got one. This agreement doesn’t address that, which tells me Microsoft has accepted they can’t control OpenAI’s governance.

Some people are calling this a downgrade for Microsoft because they lost exclusivity. I think that’s shortsighted. The old deal was creating tension every time OpenAI wanted to partner with someone else—remember when they started working with Oracle? That required a special carve-out. Now both sides can breathe. Microsoft gets priority access and a bigger financial upside. OpenAI gets operational freedom. That’s a win-win.

The real question is whether this signals a cooling of the relationship or a maturation of it. My take: it’s the latter. Microsoft and OpenAI are locked together for the long haul. Microsoft’s entire AI strategy—Copilot, Azure AI, the whole stack—depends on OpenAI’s models. And OpenAI needs Microsoft’s cloud infrastructure and distribution. This agreement just acknowledges that they don’t need to be joined at the hip to make it work.

If anything, this deal positions both companies for what’s coming next. The AI landscape is getting more competitive by the month. Google’s Gemini is catching up, Anthropic is raising billions, and open-source models are improving fast. Microsoft and OpenAI can’t afford to be distracted by partnership friction. This cleanup was necessary.

One thing that bugs me: the announcement is light on specifics about compute pricing. The old deal had Microsoft providing compute at cost, which was a massive subsidy. If that’s gone, OpenAI’s margins take a hit. I’d love to see the actual terms, but as usual, these details stay private.

Overall, this is a sensible move. Less drama, more focus on building. Both companies have bigger problems to solve than their partnership structure.