Amazon Reportedly Explores $8B Nvidia-Chip Financing Deal

Amazon's AI buildout has gotten so expensive that the company is reportedly looking for a creative way to keep the chips without keeping them on its books. The Financial Times reported Friday, October 2, 2026, citing people familiar with the matter, that Amazon has been in talks with outside investors about moving roughly $8 billion worth of Nvidia chips off its balance sheet, then leasing the same hardware straight back.

It's worth being clear upfront: this is not a done deal. The FT describes preliminary discussions, Amazon has been gauging investor interest "in recent weeks," and no transaction has actually been announced. Neither Amazon nor Nvidia responded to Reuters when asked to comment on the report.

Here's the structure being discussed, in plain terms. Amazon would transfer thousands of Nvidia Grace Blackwell chips, hardware that's already installed and actively running in its data centers, into a newly created special-purpose vehicle, essentially a standalone entity built just to hold these assets. That vehicle would then lease the chips back to Amazon, so AWS keeps using the exact same hardware it's using today. To fund itself, the vehicle would raise money partly through debt and partly by selling outside investors an equity stake of up to 10 percent. The net effect, if it happens, is that Amazon gets to keep operating the chips while someone else technically owns them, a classic sale-leaseback move, just applied to AI infrastructure instead of real estate or airplanes.

The chips in question aren't hypothetical future orders. According to the FT's reporting, they're already installed across more than a dozen U.S. data centers spanning five states, including Nevada and Virginia. Some of that hardware was purchased outright by Amazon, and some was already leased, which means the proposed restructuring would touch a mixed pool of owned and borrowed equipment rather than one clean category.

The timing lines up with just how much AWS has committed to spending on Nvidia silicon. On August 26, 2026, AWS and Nvidia jointly announced plans to deploy an additional 2 million Nvidia GPUs across AWS infrastructure through 2027 and 2028. That's an enormous forward commitment, and it helps explain why Amazon might want a financing structure that keeps some of that spending from weighing directly on its own balance sheet as the rollout continues.

It also doesn't hurt that Grace Blackwell chips are, by Nvidia's own description, in a category of their own right now. Nvidia CEO Jensen Huang said in the company's fiscal 2026 fourth-quarter earnings release that "Grace Blackwell with NVLink is the king of inference today." Nvidia backed that up with the numbers: fiscal 2026 revenue hit a record $215.9 billion, up 65 percent year over year, with data center revenue alone reaching $62.3 billion in the fourth quarter. When the chips powering that growth are this sought-after, finding smarter ways to finance them, rather than simply buying and depreciating them outright, becomes a much more attractive option for a buyer spending at Amazon's scale.

A few specifics remain genuinely unclear. The FT's reporting doesn't specify exactly how many chips would move into the vehicle or what the final terms might look like, only that the broader figure under discussion is around $8 billion. There's no confirmed timeline for when, or whether, this moves from talks to an actual signed deal, and it's possible the structure changes significantly, or falls apart entirely, before anything is finalized.

For now, what's confirmed is narrower than the headline number might suggest: a report of exploratory talks, a rough dollar figure, a description of how the vehicle would work, and the locations where the underlying hardware already sits. Everything else, including whether investors actually bite at the proposed terms, is still an open question.

If Amazon does go through with something like this, it would be a notable example of how hyperscalers are starting to rethink the financial plumbing behind the AI buildout, not just how much compute to buy, but how to structure the balance sheet so that buying it doesn't become its own liability. Given how many other cloud providers are making similarly massive GPU commitments right now, it wouldn't be surprising if this kind of leaseback arrangement, assuming it works for Amazon, starts showing up elsewhere in the industry too.