By Assaf Biton
For seven years, Microsoft owned OpenAI's cloud. One sentence about "stateful" agents nearly blew that up. On February 27, 2026, Amazon agreed to put up to $50 billion into OpenAI and became the exclusive third-party cloud for OpenAI Frontier, and the whole thing hinged on a distinction most people had never heard of.
I went back through the filings, the joint statements, and the lawsuit threats so you don't have to. The short version: OpenAI found a gap in its Microsoft contract big enough to drive a $50 billion truck through. Then, two months later, both sides quietly rewrote the contract so the gap didn't matter anymore.
So was the "loophole" a real legal weapon, or a negotiating bluff that worked? That's the question I kept circling, and the answer tells you a lot about how the AI industry actually moves money. Let me walk you through it.
TL;DR: Amazon committed up to $50B to OpenAI ($15B up front, $35B conditional) as part of OpenAI's $122 billion funding round that closed March 31, 2026 at an $852 billion valuation. The deal made AWS the exclusive third-party cloud for OpenAI Frontier and put a new "Stateful Runtime Environment" on Amazon Bedrock. OpenAI argued stateful agents fell outside Microsoft's exclusivity on "stateless" ChatGPT and API calls. Microsoft weighed suing in March 2026. Instead, on April 27, 2026 both companies amended their deal: Microsoft's license went non-exclusive, the AGI clause died, and revenue share was capped. The loophole stopped mattering because the contract changed.
What Actually Got Signed on February 27
Let's start with the numbers, because they're the easy part. Amazon agreed to invest up to $50 billion in OpenAI. That breaks into $15 billion paid early (processed around March 15, 2026) and $35 billion that only lands if certain triggers fire.
Those triggers are unusual. The $35 billion is tied to either an OpenAI IPO, or a "Mandatory Funding Event" that the research suggests is linked to artificial general intelligence. Amazon can also choose to buy in at any time. If none of that happens by December 31, 2028, the $35 billion obligation simply expires.
Here's the part that gets ignored: the equity and the cloud deal are legally welded together. If the cloud partnership (the Joint Collaboration Agreement) ends, the $35 billion equity commitment dies with it. Amazon wasn't buying a stake. It was buying a customer.
Pro Tip: When you read "$50 billion investment," mentally split it into "$15B real cash today" and "$35B option Amazon may never exercise." Only about $25B in actual cash hit OpenAI in early 2026 (Amazon's $15B plus SoftBank's first $10B). The rest is conditional or, in Nvidia's case, paid in chips instead of dollars.
The full $122 billion round, in plain numbers
The Amazon money was one slice of the largest private funding round in history. OpenAI closed it on March 31, 2026.
| Investor | Amount | Form |
|---|---|---|
| Amazon | Up to $50B | $15B cash + $35B conditional |
| Nvidia | $30B | Mostly GPU capacity, not cash |
| SoftBank | $30B | Operational funding, tranches in 2026 |
| Others | ~$12B+ | Microsoft, a16z, T. Rowe Price, bank channels |
For context, that round valued OpenAI at $852 billion post-money. The company was making roughly $2 billion a month in revenue and pulled in about $13.1 billion for the prior year. Microsoft, the partner this whole story is about, quietly participated in the round too. If you want the governance angle on who controls all that money, I broke it down in how the OpenAI Foundation controls the company.
The Loophole: Stateful vs Stateless
Now the interesting part. To understand why this deal was even legal, you have to understand one word: stateless.
Microsoft's 2019 and 2023 deals (worth more than $13 billion total) gave it exclusive cloud rights to OpenAI's "stateless" APIs. A stateless call is a single request and a single response, with no memory between them. You ask, the model answers, the conversation forgets you. That's the bulk of how the OpenAI API worked.
OpenAI and Amazon built something they argued was a different category entirely.
| Type | What It Means | Who Claimed It |
|---|---|---|
| Stateless | One request, one response, no memory | Microsoft Azure (exclusive) |
| Stateful | Persistent memory, tools, continuity across sessions | AWS Bedrock (new) |
The Stateful Runtime Environment (SRE) on Amazon Bedrock lets AI agents keep context, remember prior work, use tools, and run multi-step jobs at production scale. OpenAI's argument was clean: agents that maintain state are not "stateless API calls," so Microsoft's exclusivity never covered them.
Microsoft's counter was just as clean. Strip away the fancy framing, they said, and a stateful agent still fires plenty of stateless calls underneath. To them, this was a creative end-run around a contract, not a new product category.
"We will sue them if they breach it. If Amazon and OpenAI want to take a bet on the creativity of their contractual lawyers, I would back us, not them."
That quote, from a Microsoft insider reported by the Financial Times on March 18, 2026, is where the AI wars almost turned into a courtroom drama. For a few weeks, three-way negotiations ran in the background while the public watched for a lawsuit.
OpenAI's flagship product at the center of the Microsoft cloud-exclusivity question
Best for: General knowledge workers needing a capable all-in-one assistant, Developers wanting quick code generation with GPT Image and Codex integration
Why Microsoft Didn't Sue (and What It Got Instead)
Here's the thing about that lawsuit threat: it never happened. And in hindsight, suing was always the weaker move for Microsoft.
Microsoft was already facing antitrust probes in the U.S., UK, and EU over its Azure licensing practices. Dragging OpenAI into court over cloud exclusivity would have invited exactly the kind of scrutiny it was trying to avoid. A judge picking apart "stateful vs stateless" in open court is a regulator's dream and Microsoft's nightmare.
So on April 27, 2026, the two companies did something smarter. They rewrote the deal.
What changed on April 27: Microsoft's IP license to OpenAI's models became non-exclusive. OpenAI can now run all of its products on any cloud, not just Azure. The old AGI clause (which tied Microsoft's rights to OpenAI not reaching AGI) was killed. And OpenAI's 20% revenue share to Microsoft was kept through 2030 but capped at a total dollar limit. The next day, April 28, OpenAI's models showed up on Amazon Bedrock in limited preview.
Read that timeline again. The amended contract landed on April 27. OpenAI models hit Bedrock on April 28. That's not a coincidence, that's a settlement with a ship date.
So the loophole didn't "win" in any courtroom sense. It worked as pressure. OpenAI used the stateful argument to pry open a negotiation, and the result was a broader, cleaner deal that made the original argument moot. The clever legal theory was the crowbar, not the prize.
The Cloud Loophole: What Worked and What Didn't
What Worked
- It gave OpenAI a credible, novel argument to bring Amazon in without instantly breaching the Microsoft contract.
- It pushed Microsoft toward renegotiation instead of a public fight neither side wanted during active antitrust probes.
- It got products shipping fast: amended deal April 27, Bedrock preview April 28, general availability of GPT-5.5 and Codex on Bedrock by June 1, 2026.
- It let Amazon lock in OpenAI as a long-term cloud customer (~$138B in AWS spend over 8 years) on top of the equity.
What Didn't
- The stateful vs stateless distinction was never tested in court, so it sets no real legal precedent. We still don't know if it would have held.
- Microsoft kept a 20% revenue share through 2030 and lost only exclusivity, so OpenAI didn't escape its obligations, it just bought flexibility at a price.
- The "loophole" only mattered for about two months before the contract rewrite made it irrelevant.
- The whole episode showed OpenAI's deals are vague enough that a single ambiguous word can trigger billion-dollar disputes.
Be careful repeating the "OpenAI found a loophole and beat Microsoft" headline. The cleaner reading is that both sides faced more downside from a fight than from a deal, so they settled. The stateful argument gave OpenAI a bargaining chip, not a legal victory. Nobody won this in court because it never reached one.
The Trainium Angle: Why Amazon Cared
Money and cloud aside, Amazon wanted OpenAI running on its own chips. That's the strategic heart of this for Andy Jassy.
OpenAI agreed to deploy 2 gigawatts of capacity on AWS Trainium silicon (Trainium3 now, Trainium4 from 2027). Amazon claims Trainium offers 30 to 40% better price-performance than Nvidia GPUs, and Trainium3 delivers 2.52 PFLOPs of FP8 compute. For OpenAI, it's a way to avoid depending on any single chip vendor.
This is where the story gets awkward. Amazon already backs Anthropic, OpenAI's biggest rival. Its earlier funding totaled $8 billion through 2024, and in April 2026 Amazon committed up to $25 billion more, pushing its potential backing to roughly $33 billion. Anthropic's Claude runs on over a million Trainium2 chips. So Amazon is now selling chips and cloud to both of the leading AI labs at once. If that rivalry interests you, see my breakdown of OpenAI workspace agents vs Anthropic.
Pro Tip: Watch the chips, not the headlines. Amazon's real bet isn't "OpenAI beats Anthropic," it's "everyone runs on Trainium." By taking equity in OpenAI and Anthropic plus selling both companies silicon, Amazon profits no matter which lab wins. That's the hedge, and it's why the $50B number is almost a sideshow next to the ~$138B cloud commitment.
The Circular Money Problem
Now for the part that should make you a little uneasy. Critics flagged the deal's structure as circular, and they have a point.
Amazon invests $50 billion in OpenAI. OpenAI then commits to spending roughly $138 billion on AWS over 8 years. Amazon profits from the equity upside and the cloud bill. One analyst called it "funneling money into companies you hope will give you business."
Nvidia's $30 billion works the same way: it's mostly GPU capacity, which guarantees Nvidia hardware orders. The investor pays the company, the company pays the investor back as a customer, and everyone's valuation goes up. Sound circular? It is.
And the scale is genuinely hard to defend. OpenAI has committed to something like $1.4 trillion in total spending across clouds, chips, and infrastructure, against roughly $25 billion in annualized revenue as of early 2026. Profitability isn't expected before 2030. The company's own cash burn is steep and still climbing. Its late-2025 plan pegged it at roughly $17B in 2026, but by early 2026 OpenAI had revised that sharply upward to about $25B to $27B in 2026 and $57B to $63B in 2027, with cumulative burn through 2030 now projected near $665B.
"The OpenAI and Amazon partnership may have just won consumer agentic commerce."
That's Forrester's bull case: pair the most popular answer engine (ChatGPT) with the biggest marketplace (Amazon) and you threaten Google. Maybe. But it only pays off if enterprise AI agents actually get adopted at scale, and right now Frontier has only a handful of early customers like HP, Intuit, and Uber.
What Happens Next
The deal is signed, the lawsuit is dead, and the products are shipping. So where does this go?
The Stateful Runtime Environment evolved into "Amazon Bedrock Managed Agents, powered by OpenAI." It hit limited preview on April 28, 2026, and OpenAI's GPT-5.5, GPT-5.4, and Codex reached general availability on Amazon Bedrock around June 1, 2026. The agent piece is the real product now, not the legal argument that birthed it.
On the money side, OpenAI confidentially filed for an IPO in late May 2026, with a draft reportedly targeting a valuation above $1 trillion. The company has signaled it's in no rush, and a Q4 2026 listing window keeps getting floated. If and when that IPO happens, Amazon's $35 billion conditional commitment likely triggers.
Amazon's stock told its own story. Citi raised its price target to $285 (from $265) on March 25, 2026, citing AWS revenue from both OpenAI and Anthropic. A separate, more bullish $320 target came from 24/7 Wall St., not Citi. Wall Street, for now, likes the hedge.
The Bottom Line
What this deal really was: not a clever loophole that beat Microsoft, but a $50B lever that forced a contract rewrite. The stateful argument got OpenAI to the negotiating table; the April 27 amendment is what actually freed it.
Who won: Amazon, most clearly. It locked in OpenAI as a cloud and chip customer, kept its Anthropic hedge, and watched its stock targets climb.
What to watch: whether OpenAI's $1.4T in commitments survives contact with reality. The financing is settled. The economics are not.
If you only track one thing from here, track enterprise agent adoption on AWS Bedrock. The legal drama is over. The actual test (will companies pay for OpenAI agents at scale?) has barely started. Watch the customer count on Frontier over the next two quarters, because that number, not the $50B headline, decides whether this deal ages well.
Frequently Asked Questions
How much did Amazon actually invest in OpenAI?
Amazon committed up to $50 billion: $15 billion paid early in 2026 and $35 billion conditional on an OpenAI IPO or an AGI-linked "Mandatory Funding Event." If neither triggers by December 31, 2028, the $35 billion obligation expires. So the guaranteed cash was $15 billion, not the full $50 billion.
What is the stateful vs stateless loophole in the OpenAI Amazon deal?
Microsoft held exclusive cloud rights to OpenAI's "stateless" API calls (single request, single response, no memory). OpenAI argued that "stateful" agents, which keep memory and run multi-step tasks on Amazon Bedrock, were a different product category not covered by that exclusivity. Microsoft disagreed. The dispute was settled out of court when the companies amended their deal on April 27, 2026.
Did Microsoft sue OpenAI over the Amazon deal?
No. Microsoft was reported to be weighing a lawsuit in March 2026, but it never filed. On April 27, 2026, Microsoft and OpenAI instead amended their partnership: Microsoft's license became non-exclusive, OpenAI gained the right to run products on any cloud, the AGI clause was removed, and the 20% revenue share was capped through 2030.
Is OpenAI moving off Microsoft Azure?
Not entirely. OpenAI can now run its products on any cloud, including AWS, but products generally still ship first on Azure unless Microsoft chooses not to support them. OpenAI's models became generally available on Amazon Bedrock around June 1, 2026, so it now runs across multiple clouds rather than abandoning Azure.
What is OpenAI worth after this deal?
OpenAI closed its $122 billion funding round on March 31, 2026 at an $852 billion post-money valuation. Its confidential IPO filing in late May 2026 reportedly targets a valuation above $1 trillion, though the company has said it is in no hurry to list publicly.
