According to Fortune, Amazon is in talks to invest at least $10 billion into OpenAI, a deal reported by The Information that could value the AI startup north of $500 billion. Analysts Charles Fitzgerald and Anshel Sag characterize this not as a partnership, but as a “framework” or even a “fake deal.” The core issue is OpenAI’s massive, unfunded cloud spending commitments, including a $38 billion promise to Amazon Web Services (AWS) it can’t pay for. The proposed $10 billion investment would effectively cycle from Amazon to OpenAI and back to AWS as payment for compute. For Amazon, the goal is to validate its custom AI chips, while OpenAI gets the vital computing power it needs to keep scaling.
The Circular Financing Game
Here’s the thing: this isn’t a normal investment. It’s a financing scheme. Fitzgerald calls it out plainly. OpenAI has promised to spend insane amounts on cloud compute—over a trillion dollars in total commitments—with no clear way to pay for it. So this deal creates a neat, if dizzying, loop. Amazon gives OpenAI $10 billion. OpenAI then uses that exact money to pay Amazon for cloud services. Amazon books $10 billion in new revenue, OpenAI gets $10 billion in compute, and no actual cash leaves OpenAI’s pocket. It’s circular. But in the capital-intensive hellscape of frontier AI, this has somehow become a standard play. Companies need to secure compute at any cost, and vendors are willing to fund their own customers to make their growth numbers work and fill their new data centers.
What Amazon Really Buys
For Amazon, this is about one word: credibility. AWS is the cloud leader, but in generative AI, it’s been playing catch-up to Microsoft and Google. Its custom Trainium and Inferentia chips are met with market skepticism. Landing OpenAI as a user, even partially, is a massive validation. As Sag puts it, “ChatGPT is still seen as the Kleenex of AI.” If OpenAI uses Amazon’s hardware, it signals to every enterprise customer that it’s safe and viable. That’s priceless. It’s the stamp of approval Amazon’s AI silicon has desperately needed. They’re buying a marketing win and a guaranteed tenant for their infrastructure build-out, ensuring those expensive data centers aren’t empty.
OpenAI’s Desperate Power Play
From OpenAI’s side, this is a frantic scramble for compute and leverage. They’re supply-constrained. Microsoft’s capacity is heavily committed, and Nvidia chips are scarce. They can’t be loyal to one ecosystem if they want to keep growing. So they’re playing the field, using their “Kleenex” brand status as ultimate leverage. By pulling in Amazon, they gain access to another huge pool of hardware and, just as importantly, weaken their dependence on Microsoft. It creates a multi-vendor standoff. They can now go back to Microsoft or Nvidia and say, “Give us better terms, or we’ll just use more of Amazon’s stuff.” It’s a brilliant power move for a company that, paradoxically, can’t afford its own bills. They’re betting the entire industry is so terrified of them failing that rivals will fund their survival.
The Brutal Business Reality
But there’s a catch, and Fitzgerald nails it. This deal likely ropes Amazon into the worst part of the AI business: training. Building those massive training clusters is a brutal, low-margin game. They’re obscenely expensive, used intensely for a short time, and then made obsolete by the next chip cycle. “Training clusters are the worst business,” he says. The lucrative, steady work of inference—running the models for customers—still largely flows through Microsoft. So Amazon might be paying billions for the privilege of handling OpenAI’s most expensive, least profitable workloads. The real test, according to Fitzgerald, will be in two years. Look at how much of that headline $10 billion actually turned into lasting AWS revenue. Until then, it’s a high-stakes game of musical chairs. And OpenAI is still looking for about 28 more billions.
