A seven-year exclusive agreement was formally terminated on the afternoon of April 27.
Microsoft's official blog and OpenAI simultaneously published statements that completely rewrote the partnership agreement — a deal that had been both opaque to outsiders and a constraint on Sam Altman for an entire year.
The key points are stark: From today, OpenAI can sell all of its products on any cloud — AWS is allowed, Google Cloud is allowed. Microsoft has gone from being the "sole exclusive distributor" for seven years to a "preferred partner." The one-word difference in title represents hundreds of billions of dollars in leverage.
What the core terms changed
This is not a minor tweak. It replaces several main pillars of the original 2019 agreement. The four most critical changes:
First, the exclusivity clause is completely gone. Previously, a significant portion of OpenAI's technology — including the Frontier agent-building tool and stateful runtime capabilities — was contractually required to run only on Azure. That restriction is now lifted. OpenAI can choose any cloud. Azure retains a "first ship" right, meaning products default to Azure first unless Microsoft itself chooses not to support them.
Second, the direction of money has reversed. The original contract included a revenue-sharing component where Microsoft paid OpenAI a share — little-known but real — and that is now void. The reverse flow — OpenAI's 20% revenue share paid to Microsoft — remains in place until 2030, but a total cap has been added. The specific figure was not disclosed, but analysts estimate that given OpenAI's revenue run rate of roughly $7 billion per quarter in recent quarters, the cap likely falls in the range of tens of billions to hundreds of billions of dollars.
Third, the AGI clause has been replaced by a time lock. The original contract's most surreal provision stated: "Until the day OpenAI achieves AGI, Microsoft's exclusive rights remain in effect." Putting such language into a commercial contract was a ticking time bomb — what constitutes AGI? Who decides? If OpenAI's board were to declare "we've reached it," Microsoft's billions in investment would instantly depreciate. The new agreement replaces this with a termination date of 2032. A concrete calendar date replaces a philosophical question.
Fourth, IP licensing has been changed to non-exclusive. Microsoft's IP license for OpenAI models and products is retained until 2032, but it has been changed from exclusive to non-exclusive. Microsoft can still use, integrate, and wrap the technology, but OpenAI can now license the same technology to others as well.
Why now
The timing is not coincidental.
In early April, OpenAI finalized a $50 billion AWS compute deal. Under the original contract framework, this deal directly violated the exclusivity clause — Microsoft could theoretically have pursued legal action to block it. VentureBeat's coverage directly used the term "legal peril" to describe the situation.
Simply put: OpenAI was already in effective breach of contract, but Microsoft did not want a public legal battle, and OpenAI needed the money and AWS compute capacity. Both sides had incentives to sit down and rewrite the deal.
The deeper reason is that OpenAI is burning through cash faster than it can raise it. Anthropic closed a $30 billion Series G at a $380 billion valuation, with AWS adding another $5 billion and committing to $100 billion in compute consumption over ten years. Google's Gemini 3.1 Pro took first place in 12 out of 18 benchmarks. If OpenAI remained locked into Azure alone, it would be handing all its compute pricing leverage to Satya Nadella.
Microsoft also wanted relief. Microsoft 365 Copilot's in-house model strategy has become increasingly clear over the past two years, already using a hybrid stack of Anthropic and its own MAI-1. Continuing to hold OpenAI as an exclusive partner was actually constraining Azure's customer base — enterprise clients that prefer Claude or Gemini over OpenAI saw Azure as "the OpenAI cloud" and were reluctant to sign on.
Did Microsoft lose or gain?
The initial public reaction was that "Microsoft took a loss." Seeking Alpha reported that Microsoft's stock rose slightly on the day — the market voted with its feet that the rewrite was positive for Microsoft.
Why? Three reasons:
1. The 27% equity stake remains unchanged. Microsoft's shareholder position in OpenAI is fully preserved. Every time OpenAI's valuation rises, this paper gain doubles.
2. The 20% revenue share now has an expiration date and a cap, and no longer depends on AGI. The original contract's "termination upon AGI" clause was a hidden landmine for Microsoft — if OpenAI internally declared "we've done it," the revenue share would stop. Now it has a hard 2030 deadline plus a total cap, making it far more predictable.
3. Future litigation risk is eliminated. The $50 billion AWS deal would have been a legal fight under the old contract, with uncertain odds and enormous time costs. Now it is settled in one stroke.
But there are also concerns: analysts note that 45% of Microsoft's AI business orders are tied to OpenAI. Once OpenAI begins steering customers toward AWS or Google Cloud, this backlog could begin to shrink within a year.
What OpenAI gained
The most immediate benefit: compute pricing leverage.
OpenAI's biggest bottleneck over the past year has not been models or talent — it has been GPUs. Being locked into Azure meant GPU supply was entirely at Microsoft's discretion. Now, with three clouds competing on price, OpenAI can deploy different models on different clouds to maximize the output of every H100.
The second gain is enterprise customer coverage. Many large enterprises follow a cloud-first IT procurement strategy — once they choose AWS, they purchase exclusively within the AWS ecosystem. Previously, these customers had to open an Azure account to use ChatGPT Enterprise. Now they can call OpenAI's latest models directly from AWS Bedrock. Salesforce and Snowflake, both deeply tied to AWS, represent markets OpenAI has been unable to penetrate for the past two years.
The third gain is paving the way for an IPO. OpenAI is targeting a Q4 listing. A trillion-dollar valuation story is hard to sell if the company runs on only one cloud. Now OpenAI can tell the secondary market: "Our compute is distributed across three clouds, with no single-supplier risk." That message resonates with both retail and institutional investors.
The three-cloud landscape is shifting
| Before the rewrite | After | |
|---|---|---|
| Azure | OpenAI exclusive | Default first ship, but no longer exclusive |
| AWS | Primarily Anthropic | Now also has native OpenAI Bedrock access |
| Google Cloud | Promotes Gemini in-house | Will also get OpenAI access |
The most interesting dynamic going forward is Claude and GPT coexisting in AWS Bedrock. Anthropic just signed a 10-year, $100 billion commitment with AWS, and now AWS Bedrock also has OpenAI — two of the largest closed-source AI competitors sitting on the same shelf. Adam Selipsky has played this hand very well.
As for whether Google Cloud will get full "complete product" access to OpenAI or just an API proxy, the announcement did not provide details. Sundar Pichai will certainly want the former, but that still needs to be negotiated.
With this hurdle cleared, OpenAI's next task is Stargate — the $500 billion data center plan with Oracle and SoftBank. If that money actually starts flowing, today's cloud unbundling is just an appetizer.
Sources: The next phase of the Microsoft-OpenAI partnership (Microsoft Official Blog), CocoLoop, OpenAI ends Microsoft legal peril over its $50B Amazon deal (TechCrunch), Microsoft and OpenAI gut their exclusive deal, freeing OpenAI to sell on AWS and Google Cloud (VentureBeat), Microsoft to Stop Sharing Revenue With Main AI Partner OpenAI (Bloomberg), OpenAI shakes up partnership with Microsoft (CNBC)