Amazon Tied to $8B Nvidia AI Chip Sale-Leaseback Deal

A roughly $8 billion sale-leaseback deal involving Nvidia AI chips and Amazon has drawn attention, according to reports. Under the typical structure, Amazon would sell the Nvidia chips it uses for AI workloads to an investor and then lease those same chips back, keeping its compute capacity running while reshaping its asset base.

Sale-leaseback is a well-established financial maneuver. A company sells a heavy asset it owns to a third party and then rents it back for continued operation. The buyer gains an asset that produces steady cash flow, while the original owner frees up capital, moves the asset off its balance sheet, and still keeps operational use of it.

In the AI context, the move stands out because Nvidia GPUs are both extremely expensive and fast-depreciating. Cloud providers racing to serve training and inference demand must commit enormous capital expenditure to secure chips well ahead of the revenue those chips generate, and they carry the depreciation and obsolescence risk themselves. A sale-leaseback lets a company lock in compute while shifting part of that capital burden elsewhere.

That makes this more a sign of the times than an isolated accounting trick. As AI infrastructure spending by leading cloud providers and AI labs keeps setting records, financing these heavy assets and managing their depreciation and payback cycles has become a more pressing problem than the models themselves.

The takeaway is that AI compute is moving from simply buying hardware toward becoming a structured, tradeable financial asset. Once compute can be packaged, leased, and refinanced, competition among cloud providers is no longer only about who buys the most chips; it is also about who can turn compute into a sustainable business at a lower cost of capital and with more flexible structures.

What to watch next is the terms and the follow-through: the lease duration, the pricing, who the counterparty is, and whether the deal ultimately closes. If similar structures spread to more players, how AI chips are financed and accounted for could be redefined, and that would shape how the true cost of cloud providers' AI buildout is judged.