Microsoft-OpenAI Exclusivity Is Over — AWS and Google Can Now Sell GPT

In January 2023, Satya Nadella signed a check worth roughly $10 billion and called Microsoft's partnership with OpenAI "the most important technology collaboration of our generation." He stood on a stage in Redmond, flanked by Sam Altman, and painted a picture of a future where Azure would be the backbone of artificial intelligence — the one cloud to rule them all. Three years later, on a quiet Sunday afternoon in late April 2026, the core terms of that partnership were rewritten in ways that would have been unthinkable at the time of that photo op.

Cloud exclusivity? Dead. The infamous AGI clause? Deleted. The revenue-sharing formula that bound the two companies together like financial Siamese twins? Completely restructured. And the punchline: OpenAI is now free to sell GPT-5.5, Sora, and every other model it builds on Amazon Web Services and Google Cloud Platform.

Microsoft's consolation prize isn't exactly small — a roughly 27% equity stake worth an estimated $135 billion at the latest tender prices. But the nature of the relationship has fundamentally shifted. What was once an exclusive marriage is now, at best, a very expensive open relationship.

Here's the story of how it happened, why both sides agreed to it, and what it means for everyone from cloud engineers to Wall Street analysts.

The origin story: how a billion dollars bought a monopoly

To understand why April 27 matters so much, you have to go back to July 2019. That's when Microsoft first invested $1 billion in OpenAI — a company that, at the time, most people in tech had barely heard of. GPT-2 had just been released a few months earlier, and it was impressive enough to make headlines, but nobody was calling it a world-changing technology. Not yet.

The deal Nadella struck was elegant in its simplicity. Microsoft would provide the compute — the thousands of GPUs needed to train ever-larger models — and in return, it would get two things. First, an exclusive license to all of OpenAI's commercial intellectual property. Second, the guarantee that every OpenAI product would run on Azure and only Azure. No AWS. No Google Cloud. Azure or nothing.

For a while, this looked like the smartest bet in tech history. When ChatGPT launched in November 2022 and became the fastest-growing consumer application ever, every single API call flowed through Azure data centers. Microsoft didn't just have a partnership with the hottest AI company in the world — it had a monopoly on distributing that company's products.

By 2025, AI-related workloads had pushed Azure's revenue to new heights. Roughly 16% of Microsoft's total cloud business came from AI services, and a huge chunk of that was powered by OpenAI models. Enterprise customers who wanted GPT had exactly one option: sign an Azure contract. And they did, by the tens of thousands.

But here's the thing you need to understand about monopolies in tech: they tend to work brilliantly until they don't. And by late 2025, cracks in the Microsoft-OpenAI structure had started forming in ways that neither company could ignore.

The cracks: how Anthropic changed the game

The first real warning sign came from an unlikely competitor. Anthropic — the AI safety company founded by former OpenAI researchers Dario and Daniela Amodei — had been quietly eating OpenAI's lunch in the enterprise market. While OpenAI was locked into Azure-only distribution, Anthropic's Claude models were available everywhere. AWS Bedrock. Google Vertex. Azure Marketplace (which was particularly embarrassing for OpenAI). Direct API.

The numbers told a brutal story. By early 2026, Anthropic had captured roughly 40% of the enterprise LLM API market. OpenAI, despite having better brand recognition and arguably stronger models, had fallen to 27%. The reason wasn't technical — it was structural. Enterprise buyers overwhelmingly prefer to consume AI services through their existing cloud provider. If you're an AWS shop running 80% of your infrastructure on Amazon's cloud, spinning up a separate Azure account just to access GPT feels like an unnecessary headache. And increasingly, enterprises were deciding it simply wasn't worth the hassle.

Sam Altman saw this happening in real time. Internal sales reports showed deal after deal slipping away — not because GPT wasn't good enough, but because the prospect's CTO would say something like, "We love the model, but we're not moving our cloud infrastructure to Azure for one vendor." OpenAI's leadership held internal reviews in late 2025, and the conclusion was unanimous: Azure-only distribution was costing them enterprise deals at an accelerating rate.

There was a deeper irony at play, too. Anthropic's success was partly built on a strategy that OpenAI had pioneered — making the best possible model and letting developers build on top of it. But Anthropic had added a crucial ingredient: platform agnosticism. Claude didn't care where you hosted your infrastructure. That flexibility turned out to be worth more than raw model performance in the enterprise market.

Source: unsplash.com · Unsplash License

The $50 billion trigger

If the Anthropic competition was the slow-burning fuse, the bomb went off on April 22, 2026 — five days before the exclusivity ended. That's when OpenAI and Amazon announced a jaw-dropping $50 billion compute deal. Under the agreement, OpenAI would gain access to massive GPU clusters in AWS data centers, and in return, OpenAI's models would become available through AWS Bedrock.

Think about that number for a second. Fifty billion dollars. That's more than the GDP of most countries. It's roughly what Disney paid for Fox. And it came with a very obvious prerequisite: the Azure exclusivity clause had to die. You can't sign a $50 billion deal with Amazon while being contractually obligated to run everything on Microsoft's cloud. The math simply doesn't work.

An internal OpenAI memo, later reported by TechCrunch, captured the mood: "Demand on AWS has been staggering." Enterprise customers who had been waiting for years to access GPT through their preferred cloud provider were lining up before the ink was dry.

Andy Jassy, Amazon's CEO, had been courting OpenAI for over a year. Amazon had already invested heavily in Anthropic — reportedly $4 billion — but Jassy understood that in the enterprise AI market, offering more models is always better than offering fewer. If AWS could offer both Claude and GPT through Bedrock, it would become the undisputed one-stop shop for enterprise AI.

The AWS deal also solved a practical problem for OpenAI. Training frontier models requires obscene amounts of compute, and Azure's GPU capacity, while massive, wasn't growing fast enough to keep up with OpenAI's ambitions. AWS offered access to a second massive pool of GPUs — diversifying OpenAI's infrastructure dependency in the same way the deal diversified its distribution.

What Nadella was thinking

So why did Microsoft agree to give up what was arguably its most valuable competitive asset? From the outside, dropping exclusivity looks like a concession — maybe even a surrender. But Satya Nadella's calculus was more nuanced than it appears, and it came down to three interrelated factors.

The first was regulatory pressure. And not just a little pressure — a full-court press from multiple jurisdictions simultaneously. The UK's Competition and Markets Authority (CMA) had been investigating the Microsoft-OpenAI relationship since 2023. The European Commission had opened its own probe. And in the United States, the Federal Trade Commission under Lina Khan's successor was asking increasingly pointed questions about whether Microsoft's exclusive control over OpenAI's distribution constituted a de facto acquisition without the regulatory scrutiny that would normally accompany one.

Dropping exclusivity was the cleanest way to defuse all three investigations at once. It allowed Microsoft to credibly argue: "This is an investment, not an acquisition. We don't control their distribution. They can sell to anyone." It's the kind of move that makes antitrust lawyers smile and regulators quietly close their files.

The second factor was Microsoft's own model ambitions. Here's something that didn't get enough attention in the initial coverage: Microsoft has been steadily reducing its dependence on OpenAI for months. The company's own AI models — Phi-5 and MAI-3 — have been taking over an increasing share of the workload inside Copilot, Microsoft's AI assistant that's embedded across Office, Windows, and Azure.

Nadella has always been a portfolio thinker. He doesn't want Microsoft's AI future to depend entirely on one external company, no matter how good that company is. Loosening the OpenAI dependency gives Microsoft more room to shift workloads to its own models when they're good enough, while still having access to OpenAI's frontier models when nothing else will do.

The third factor — and this is the one that Wall Street initially missed — was margin improvement. Under the old deal, Microsoft paid OpenAI a revenue share on resold products. Every time Microsoft sold a ChatGPT Enterprise subscription through Azure, it had to kick a percentage back to OpenAI. Under the new structure, that payment is eliminated. Microsoft keeps 100% of the margin on every OpenAI product it resells through Azure.

Let that sink in. Microsoft gave up exclusivity, but in exchange, every dollar of OpenAI revenue flowing through Azure is now pure Microsoft margin. For a company doing hundreds of billions in annual revenue, that's not a rounding error — it's a meaningful improvement to the bottom line.

The three changes that rewrote the rules

Let's get specific about what actually changed on April 27, because the details matter enormously.

Term	Before (2019-2026)	After (April 27, 2026)
Cloud exclusivity	Azure only	Non-exclusive (AWS, GCP allowed)
IP license	Microsoft exclusive	Non-exclusive, through 2032
AGI clause	Auto-terminates on AGI	Deleted
MS reselling revenue share	MS pays OpenAI a cut	Eliminated
OpenAI revenue share to MS	Uncapped, indefinite	Capped, through 2030
MS ownership	Revenue-share structure	~27% equity (~$135B)
Azure priority	Exclusive	"Primary partner" + first-ship

The first change — dropping cloud exclusivity — is the most visible, but arguably not the most consequential. OpenAI can now distribute its models on any cloud platform, but Microsoft retains "primary partner" status and a first-ship privilege. In practice, this means new OpenAI models will still launch on Azure before they're available anywhere else. It's not exclusivity, but it's a meaningful head start.

The second change is the deletion of the AGI clause, and this one deserves a much closer look, because it was the single biggest obstacle to OpenAI's corporate future.

Here's how the old clause worked. The original 2019 agreement gave OpenAI's nonprofit board the unilateral power to declare that OpenAI had achieved artificial general intelligence — AI that matches or exceeds human-level cognitive ability across virtually all domains. If the board made that declaration, Microsoft's IP license would automatically terminate. Overnight. No negotiation, no appeal, no transition period. Microsoft would simply lose access to the technology it had spent tens of billions of dollars helping to create.

The clause was originally designed as a safety mechanism. The idea was that if OpenAI built something truly transformative — something that could reshape civilization — it shouldn't be controlled by a single corporation's commercial interests. The nonprofit board would act as a check on commercialization run amok.

In practice, the clause had become a corporate governance nightmare. Institutional investors looked at it and saw an unquantifiable risk. How do you value a company whose core technology license could evaporate based on a philosophical judgment call by a nonprofit board? The answer is: you don't invest. Or if you do, you demand a massive risk premium.

This mattered because OpenAI has been openly preparing for an IPO. The company's valuation has ballooned to roughly $850 billion in private markets, and taking it public is the logical next step. But no investment bank would underwrite an IPO with the AGI clause hanging over it. Deleting it didn't just clean up a contract — it cleared the runway for what could be the largest technology IPO in history.

The third change — the revenue-share restructuring — is the subtlest, and it reveals who actually got the better deal in the short term. Under the new terms, Microsoft's take from OpenAI's revenue is capped and expires in 2030. But here's the asymmetry: Microsoft simultaneously eliminated the payment it used to make to OpenAI on resold products. So Microsoft collects from OpenAI (albeit with a cap), while OpenAI gets nothing from products sold through Azure.

Brad Smith, Microsoft's Vice Chair, described it as "a better structure for both companies." That's diplomatically phrased. In reality, Microsoft got the better short-term economics while OpenAI got the better long-term strategic position. OpenAI traded Azure revenue for the freedom to sell on AWS and GCP — betting that the new distribution channels will more than compensate for what it lost on the Microsoft channel.

The collapse of the cloud-model moat

To appreciate the full significance of April 27, you need to zoom out and see it as part of a larger trend that's been reshaping the cloud industry for the past two years.

Until about 2024, the cloud AI market had a neat, almost feudal structure. Your cloud provider dictated your AI model choice. If you were on Azure, you used GPT. If you were on AWS, you used Claude (via Bedrock). If you were on Google Cloud, you used Gemini (via Vertex). Each cloud had its champion model, and that model was either exclusive or so deeply integrated that switching wasn't worth the effort.

This structure was comfortable for the cloud providers — it created lock-in on top of lock-in. But it was terrible for customers, and it was increasingly untenable as the models themselves became more commoditized. When GPT-4 was the only model that could handle complex reasoning tasks, being forced onto Azure to access it was annoying but tolerable. Once Claude 3.5 and Gemini Ultra proved they could match GPT on most benchmarks, the Azure requirement started feeling like an arbitrary tax.

Anthropic fired the first shot across the bow in 2025 by putting Claude on Azure Marketplace. That was a deliberate provocation — Anthropic's best model, available on Microsoft's own cloud platform, competing directly with OpenAI for Azure customers. It proved that the old feudal structure was already crumbling.

OpenAI's multi-cloud move finishes the job. We're now entering an era where every major cloud offers every major model. AWS will have GPT, Claude, and eventually Gemini. Azure will have GPT, Claude, and its own Phi/MAI models. Google Cloud will have Gemini, Claude, and soon GPT. The model is no longer the moat. Distribution, pricing, and integration quality are the new battlegrounds.

Source: unsplash.com · Unsplash License

For developers, this is unambiguously good news. If your company runs on AWS, you'll soon be able to call GPT-5.5 through Bedrock without setting up a single Azure resource. Multi-model architectures — where you route different tasks to different models based on cost, speed, and capability — are about to become dramatically easier to build. The sales pitch "we use the best model for each task" just went from aspirational to practical.

For startups and product managers, the implications are equally significant. Cloud vendor lock-in no longer dictates your AI model choice. You can pick your cloud based on pricing, regional availability, and existing infrastructure, then layer whatever models you want on top. The days of choosing between "the cloud I want" and "the model I want" are ending.

The regulatory shadow

There's another dimension to this story that's easy to overlook but may end up being the most consequential in the long run: regulation.

The EU AI Act — the world's most comprehensive AI regulation — begins enforcing its General-Purpose AI (GPAI) provisions on August 2, 2026, barely three months after the Microsoft-OpenAI restructuring. Under the GPAI rules, companies that develop or distribute powerful AI models face new transparency, testing, and reporting requirements. A structure where Microsoft exclusively controlled OpenAI's distribution was a regulatory liability — it blurred the lines between developer and distributor in ways that would have complicated compliance.

In the United States, the FTC has been scrutinizing vertical integration in AI markets with increasing intensity. The question regulators keep asking is straightforward: when one company invests billions in another company, gets exclusive distribution rights, and controls the infrastructure that company depends on, is that really just an "investment"? Or is it an acquisition by another name?

By dropping exclusivity, Microsoft and OpenAI have essentially answered that question preemptively. "Look," they can now tell regulators, "OpenAI sells on AWS and Google Cloud. We don't control their distribution. This is a financial investment, not a structural one." It's a powerful argument, even if skeptics will point out that Microsoft's 27% stake and first-ship privileges still give it enormous practical influence.

The timing here is almost certainly not coincidental. Getting the restructuring done before the EU AI Act's GPAI provisions take effect gives both companies a clean slate to build their compliance frameworks around.

OpenAI's road to IPO

Pull back far enough, and you can see April 27 as one step in a much longer journey — one that started with OpenAI's founding as a nonprofit in 2015 and is heading toward what will likely be one of the biggest IPOs in history.

The arc goes like this. OpenAI started as a nonprofit research lab with a mission to build safe artificial general intelligence for the benefit of humanity. In 2019, it created a "capped profit" subsidiary to attract the investment it needed to compete with Google's DeepMind. Microsoft invested $1 billion and got exclusivity. In 2023, Microsoft invested another $10 billion. In October 2025, OpenAI completed its conversion to a fully for-profit corporation, shedding the capped-profit structure entirely.

Each step along this path removed an obstacle to going public. The for-profit conversion eliminated the awkward governance structure. The AGI clause deletion removed the single biggest risk factor that would have scared off institutional investors. The exclusivity removal widened the revenue growth story — instead of being limited to Azure's customer base, OpenAI can now pitch investors on total addressable market across all three major clouds.

Sarah Friar, OpenAI's CFO — who was specifically hired to prepare the company for an IPO — has been orchestrating much of this behind the scenes. The market already values OpenAI at roughly $850 billion in private transactions. A public listing could push that north of $1 trillion, making it one of the most valuable companies on Earth within a decade of launching its first consumer product.

The skeptics aren't wrong to note that this trajectory represents a stunning departure from OpenAI's original mission. A nonprofit founded to ensure AI benefits all of humanity has become a trillion-dollar corporation optimizing for shareholder value. Whether that's pragmatic evolution or mission abandonment depends on your perspective — and probably on whether you own shares.

Who wins, who loses, who watches

Every restructuring creates winners and losers, and this one is no exception.

OpenAI walks away with the biggest strategic prize: multi-cloud distribution that unlocks the entire enterprise market, a cleared path to IPO, and the $50 billion AWS deal that gives it a second major infrastructure partner. Sam Altman has been playing a long game, and April 27 represents the culmination of at least two years of careful maneuvering.

Microsoft's wins are more defensive but equally real. The regulatory shield alone may be worth more than the exclusivity it gave up — avoiding a hostile CMA or FTC action could have cost far more in legal fees, management distraction, and forced divestitures. The 27% equity stake gives Microsoft enormous upside if OpenAI's IPO succeeds. And the elimination of the reselling revenue share means better margins on every OpenAI product sold through Azure.

The clearest losers are Azure's sales teams. For years, they've walked into enterprise pitches with a killer argument: "You want GPT? You need Azure." That argument is gone. Azure will still have first-ship privileges, but "we get it first" is a much weaker pitch than "we're the only place you can get it." Expect some painful quarters as enterprise customers who were locked into Azure purely for GPT access start evaluating whether to move those workloads to their preferred cloud.

Small LLM API intermediaries — the companies that built businesses routing API calls between different model providers — are also in trouble. When all three major clouds offer GPT, Claude, and Gemini directly, the value of a third-party routing layer shrinks considerably.

Then there's Anthropic. Claude currently holds that 40% enterprise API market share, built largely on the advantage of being available everywhere while GPT was locked to Azure. That advantage just disappeared. OpenAI's models on AWS Bedrock will compete directly with Claude on Anthropic's home turf. Anthropic still has strong relationships with AWS customers and a reputation for safety-focused development, but the competitive landscape just got significantly harder.

Regulators in the EU, UK, and US will be watching closely. The exclusivity is gone, but Microsoft's 27% stake combined with first-ship privileges may still constitute what competition lawyers call "practical integration." Don't be surprised if at least one regulatory body decides that the restructuring didn't go far enough.

The view from the cheap seats

Not everyone is buying the narrative that this restructuring is a win-win.

Gary Marcus, the NYU Professor Emeritus who has become perhaps the most prominent AI skeptic in public discourse, put it bluntly: "The structure changed but the practical dependency hasn't. As long as Microsoft holds Azure first-ship rights, OpenAI's 'freedom' is limited." He has a point. First-ship means that every new OpenAI model launches on Azure before it's available anywhere else. In a market where being first matters enormously — where enterprise customers sign multi-year contracts based on who has the newest capabilities — that's a significant practical advantage that survives the end of formal exclusivity.

Matt Shay, a closely followed analyst at Bernstein, offered a more investor-focused critique. "Short-term, this is bad for Microsoft shareholders," he told CNBC. "They gave up exclusivity — their core asset — in exchange for a 27% stake in a company that may or may not IPO." Shay's concern is that Microsoft traded a concrete, revenue-generating advantage (being the only place to get GPT) for a paper gain (equity in a company that hasn't gone public yet and whose valuation could fluctuate dramatically).

Microsoft's stock did dip on the news. The market's initial read was that losing exclusivity was a net negative, regardless of the margin improvements and regulatory benefits. Whether that reaction is temporary or structural should become clearer in the coming weeks as investors digest the full implications.

What happens next

For everyday users, the most immediate effect will be improved availability and potentially lower prices. More clouds running OpenAI models means more competition for your business. Regional availability should improve too — if you're in a geography where Azure has limited presence but AWS has strong data center coverage, you'll finally be able to access GPT through a local endpoint.

For developers, the practical advice is straightforward: keep an eye on AWS Bedrock for GPT-5.5 early access. Reports suggest a public preview is targeting mid-May 2026. If your infrastructure is on AWS, you'll soon be able to call GPT models without spinning up Azure resources. Multi-model routing — running Claude, Gemini, and GPT side by side and dispatching each request to the best model for the task — is becoming the standard architecture, and the tooling to support it is maturing rapidly.

For product managers and startup founders, the calculus just simplified. Cloud vendor lock-in no longer dictates your AI model choice. Pick your cloud based on pricing and infrastructure, then layer whatever models you want on top. The multi-model future that everyone's been talking about for years just arrived.

For investors, the key metrics to watch are Microsoft's post-announcement stock trajectory and any signals about OpenAI's IPO timeline. The AGI clause deletion removed the biggest obstacle to going public, and Sarah Friar's hiring was always about preparing for that moment. If the IPO happens in 2027 — which multiple sources now consider likely — Microsoft's 27% stake could be worth significantly more than $135 billion.

The bigger picture

Step back far enough and April 27, 2026 looks like one of those dates that gets remembered not for what happened on that specific day, but for what it signaled about the direction of an entire industry.

The era of exclusive cloud-model partnerships is ending. The idea that you'd pick your cloud provider and accept whatever AI model came bundled with it — that's over. We're moving into a world where models are interchangeable commodities distributed across every major platform, and the competition shifts to infrastructure quality, pricing, developer experience, and integration depth.

That's good for customers. It's good for competition. It might even be good for AI safety, since it reduces the concentration of power in any single company's hands. But it also means that the competitive advantages in AI are shifting from "who has the best model" to "who has the best distribution" — and in that game, the cloud hyperscalers hold all the cards.

The most expensive alliance in AI history just went non-exclusive. The free market for frontier models has arrived. And the next chapter — where AWS, Azure, and Google Cloud compete on equal footing to sell the same models to the same customers — promises to be even more interesting than the last.

References

• Microsoft Official Blog — The next phase
• OpenAI Blog — Next phase of Microsoft partnership
• Bloomberg — OpenAI Breaks Free
• CNBC — OpenAI shakes up partnership
• TechCrunch — OpenAI ends Microsoft legal peril
• Engadget — OpenAI breaks out of exclusivity
• Yahoo Finance — Microsoft Loses Its OpenAI Exclusivity