Advertisement
Business

Microsoft and OpenAI’s Uneasy Partnership Keeps Getting Renegotiated

A partnership built on a single contract usually goes quiet after the signing. Microsoft and OpenAI’s has done the opposite. Since Microsoft first put money into the startup in 2019, the terms governing cloud access, intellectual property, revenue sharing and control have been rewritten repeatedly, each time after one side’s ambitions outgrew the last deal. The most recent overhaul, announced in October 2025, gave Microsoft a stake of roughly 27 percent in OpenAI’s new for-profit entity and extended its rights to OpenAI’s models into the early 2030s. It also stripped away some of the exclusivity that once made the relationship feel like a lock.

That pattern is the real story. Neither company can afford to walk away, and neither can afford to stay on the old terms.

Rows of servers inside a large data center powering cloud computing
Photo by Brett Sayles / Pexels

How a Research Bet Became a Dependency

The original logic was clean. OpenAI needed enormous amounts of computing power to train larger models, and Microsoft wanted a frontier research lab it did not have to build itself. Microsoft’s initial billion-dollar investment in 2019 made Azure the lab’s cloud home, and a much larger commitment in early 2023, widely reported at around $10 billion, deepened it. Much of that money circled back to Microsoft as Azure spending, which made the arrangement as much a customer relationship as an investment.

The weakness in that design surfaced in November 2023, when OpenAI’s nonprofit board abruptly fired Sam Altman and Microsoft learned how little formal say it had over the organization it was funding. Altman returned within days, but the episode left a mark. A company with tens of billions of dollars of exposure to a startup had discovered that the startup’s governance could change overnight, and it had no vote on the matter.

Where the Contract Started to Pinch

Compute was the first pressure point. Training and serving frontier models takes more capacity than a single partner can reliably supply, and OpenAI’s appetite grew faster than Azure could comfortably feed it. When OpenAI announced its Stargate data center project with Oracle and SoftBank in January 2025, Microsoft’s exclusive cloud status gave way to something weaker: a right of first refusal on new capacity. In practice that meant OpenAI could go elsewhere if Microsoft declined to build what it needed. The October 2025 agreement went further and removed even that preference.

Microsoft, for its part, has been quietly reducing its own dependence. It built its Copilot products on OpenAI models, but it has also been developing in-house models and, by multiple accounts, adding models from other labs into some of its products. A company that sells an assistant to millions of workers does not want its core feature controlled by a partner that might change terms, change leadership or change allegiance. Hedging is the rational response, even when the partner is also a major shareholding target.

Executives sitting around a conference table during a negotiation
Photo by Yan Krukau / Pexels

Structure was the third source of friction. OpenAI began as a nonprofit and later added a capped-profit arm so it could raise money from investors, but a cap on returns is an awkward fit for the sums required to keep up in this race. Converting to a more conventional public benefit corporation required sign-off from regulators and, just as importantly, from Microsoft, whose rights were written around the old setup. That gave Microsoft real leverage in the talks. The nonprofit foundation kept control and a large equity position, while Microsoft traded some of its exclusivity for a defined slice of the new company.

Then there is the AGI clause, probably the strangest term in modern corporate contracting. Under the earlier arrangement, Microsoft’s rights to OpenAI’s technology were tied to the moment OpenAI’s board declared it had achieved artificial general intelligence, a term with no settled definition. That gave one party the power to end the other’s access by announcing a milestone it alone defined. The revised deal routes that declaration through an independent panel of experts, which sounds more orderly, though it still hangs a commercial relationship on a concept nobody can measure.

What Each Side Actually Wants

OpenAI wants freedom of movement. It needs capital on a scale that only a handful of companies can supply, compute from whichever providers can deliver it, and the ability to sell its products to enterprise customers wherever they already do business. Reports of its growing agreements with other cloud and chip providers fit that goal. Every restriction written into the Microsoft contract is, from OpenAI’s perspective, a cost on its ability to grow.

Microsoft wants three things at once: durable access to the best models, an equity stake that rises with OpenAI’s value, and room to build alternatives in case the relationship sours. The October agreement gave it a reported commitment from OpenAI to purchase a further $250 billion of Azure services, which keeps OpenAI’s growth tied to Microsoft’s cloud revenue even without exclusivity. But the stake cuts both ways. It is an asset if OpenAI thrives and a concentrated exposure if the company stumbles, and Microsoft has to manage both possibilities in the same boardroom conversations.

Why This Won’t Be the Last Renegotiation

Every version of the deal has been priced on assumptions that keep moving: what the next generation of models will cost to train, how quickly customers will pay for them, how regulators will treat the corporate structures behind them. A contract that runs into the 2030s is trying to forecast an industry that has rewritten its own economics several times in five years. Amendments are not a sign of sloppy drafting. They are what happens when the thing being contracted over keeps changing shape.

Modern glass office building exterior representing a corporate technology campus
Photo by Lars H Knudsen / Pexels

The competitive overlap makes matters harder. OpenAI sells directly to businesses and consumers, which puts its products in the same procurement conversations as Microsoft’s Copilot offerings. Each new workplace feature OpenAI ships is both a reason for a company to spend more on Azure and a reason to question whether it needs Microsoft’s own assistant at all. Few partnerships have to survive their members competing for the same customer, and fewer still have to do it while one holds a large stake in the other.

The clause most likely to trigger the next round is the AGI panel. Microsoft’s rights to OpenAI’s models run through 2032, but its revenue-sharing arrangement and certain protections hinge on whether that expert panel ever confirms the milestone. Somebody will eventually have to convene it, and neither company has an obvious incentive to be the one asking.

Related Articles