These days, the OpenAI investment news cycle feels like the most expensive game of musical chairs ever played. And this week, there’s a new player scrambling for a seat.
Amazon is in talks to invest at least $10 billion in OpenAI, according to a report from The Information, in a deal that could push the value of the startup past $500 billion.
Analysts say the deal looks like a marriage of necessity: OpenAI needs help funding its enormous burn rate, while Amazon needs proof that its custom Trainium chips matter in a world still dominated by Nvidia.
According to two veteran industry watchers—Charles Fitzgerald, a cloud infrastructure investor, former Microsoft employee, and self-described “capex obsessive,” and Anshel Sag, a principal analyst at Moor Insights & Strategy—the Amazon talks look less like a partnership and more like a framework. But even the framework shows how OpenAI has the capacity to set the rules of the AI economy, using vendors as financiers and its own scale and urgency as leverage in negotiations.
To understand the deal, follow the money—or, more precisely, the absence of it.
‘This is a fake deal’
Fitzgerald points out that OpenAI does not have the cash to honor the $38 billion cloud-spending commitment it announced with Amazon earlier this year—let alone a “fraction of a percentage” to the more than $1 trillion in aggregate spending commitments it has floated across cloud providers, chipmakers, and infrastructure partners.
“This is a fake deal,” Fitzgerald told Fortune plainly. “Or, more politely, it’s a framework.” That is where the new $10 billion comes in. In Fitzgerald’s view, the investment functions less like traditional venture capital and more like a financing scheme designed to paper over that gap.
“If OpenAI wins the lottery, then they’d have the money to pay for this,” he said. “In practice, the deal will be much, much smaller.”
The mechanics of what some have called the “circular” trade are simple, if quite dizzying: Amazon would move $10 billion from its balance sheet to OpenAI’s bank account. OpenAI would then effectively hand that money right back to Amazon’s cloud division (AWS) to pay for the compute it promised to buy. Amazon, thus, gets to book $10 billion in “new” cloud revenue—juicing its growth numbers for Wall Street—while OpenAI gets $10 billion worth of free computing power without actually burning its own cash.
“It certainly looks like circular financing,” Fitzgerald said. “But they’ve got to do something to come up with the money.”
Sag argues that in 2025, these types of deals have become the standard cost of doing business at the frontier. The sheer capital required to train modern models is so high that traditional revenue models can’t support it yet.
“There’s a lot of circular economics happening right now,” Sag told Fortune. “Companies want to potentially profit from the relationship beyond just a regular business engagement … By making those financial investments, it does inherently increase the risk.”
However, Sag notes, the loop also provides a safety net. By funding OpenAI directly, Amazon is effectively buying itself a guaranteed customer for its massive infrastructure build-out, ensuring its new data centers don’t sit empty.
What does Amazon get, and what does OpenAI get?
Sag sees the deal less as financial maneuvering and more as part of OpenAI’s frantic hunt for computing power. “OpenAI is trying to secure as much compute as it can, from as many places as possible,” he said.
That reflects the reality of the AI economy going into 2026: Demand for AI chips still exceeds supply. Nvidia remains the gold standard for training models, but capacity is constrained. Microsoft’s infrastructure is heavily committed, so if OpenAI wants to keep scaling, it can’t afford to be loyal to any one ecosystem. By pulling Amazon closer, OpenAI gains access not just to capital but to additional pools of hardware—including Amazon’s Trainium and Inferentia chips. They may not match Nvidia’s latest offerings on raw performance, but they’re available and might even be preferable in “commercial contexts,” Sag said.
For Amazon, the appeal is simpler: credibility.
Despite being the world’s largest cloud provider, Amazon has struggled to position itself as a first-tier player in generative AI. While Microsoft locked in OpenAI early on, and Google built Gemini around its own in-house ecosystem, Amazon has spent years pitching its silicon to a market that is skeptical.
Amazon, however, has already committed at least $8 billion to Anthropic, which trains and runs its Claude models on Amazon infrastructure, including its Trainium chips. But Sag suggested Amazon’s deal with Anthropic happened largely because Anthropic couldn’t get the chips from Nvidia, which is known to be “pretty preferential” with the companies it chooses to supply.
Landing OpenAI, even partially, changes that narrative overnight, getting Amazon squarely into a chair before the music (and capital) runs out.
“ChatGPT is still seen as the Kleenex of AI,” Sag said. “If OpenAI uses your hardware at any scale, that’s a huge validation.”
If OpenAI publicly treats Amazon’s chips as “good enough,” it sends a signal to enterprise customers that Trainium is safe, viable, and future-proof.
The bad business problem
Fitzgerald cautions that Amazon may be paying for the wrong kind of exposure.
The structure of the relationship suggests Amazon would primarily handle training workloads, the compute-intensive process of building new models. So far, that’s been a brutal business for cloud providers: expensive to stand up, short-lived in value, and constantly made obsolete by the next refresh of chips.
“Training clusters are the worst business,” Fitzgerald said. “They’re massively expensive, used intensely for a short period, and then Nvidia makes them irrelevant.”
The more stable and lucrative side of the AI economy—inference, distribution, and ongoing customer interaction—remains firmly in Microsoft’s hands. When users interact with ChatGPT, they’re still hitting Microsoft servers, Fitzgerald says. But by pulling Amazon into its orbit, OpenAI weakens its dependence on Microsoft and creates a multiparty standoff. Fitzgerald describes it as a deliberate effort to play suppliers against one another.
“They can go back to Nvidia or Microsoft or Oracle and say, ‘If you don’t give us better terms, we’ll just use Amazon,’” he said.
It’s a powerful strategy for a company that, paradoxically, doesn’t have the cash to pay for what it’s promising. OpenAI is betting its technology is essential enough—and its collapse unthinkable enough—that rivals will keep funding the ecosystem just to stay close, Fitzgerald said.
That doesn’t mean we’re in a bubble, though, he added. There is real demand, real revenue, and real scarcity—alongside very real excess.
There’s an easy way to tell whether this deal mattered, according to Fitzgerald. Two years from now, “look at how much of that headline number actually turned into AWS revenue.”
Until then, OpenAI’s check might be in the mail.
“Now, they just have to find the other $28 billion,” Fitzgerald said, laughing.
