Amazon is in talks to move roughly $8 billion worth of Nvidia Grace Blackwell chips into a special purpose vehicle that would sell debt to outside investors, buy the hardware outright, and lease it back to the company, according to a report by the Financial Times. The arrangement, known as a sale-leaseback, would leave the chips exactly where they are today — installed in Amazon data centers across at least five US states — while shifting ownership, and much of the financial risk, to outside investors.
Amazon would retain up to 10 percent of the equity in the vehicle, according to the report. The hardware stays where it is; only the ownership changes. For more context on this story, see our ongoing AI industry coverage.
How the Sale-Leaseback Structure Works
Under the proposed structure, a special purpose vehicle would issue debt to institutional investors, use the proceeds to buy the Nvidia Grace Blackwell systems from Amazon, and then lease the same hardware back to the company. The servers never move. The computing capacity Amazon rents to AWS customers stays online, and the operational relationship between Amazon and its chips does not change.
What changes is who owns the hardware and who bears the risk of its depreciation. Sale-leaseback deals are a familiar tool in real estate and aviation, where airlines have long sold aircraft to investors and leased them back over decades of useful life. Applying the structure to AI accelerators is far less settled, because the useful life and resale value of a GPU generation is a matter of intense debate right now.
Why Amazon Wants the Chips Off Its Books
The move comes as Amazon's AI spending has escalated sharply. The company spent $54.2 billion on capital projects in the second quarter of 2026, roughly 68 percent more than a year earlier, and CEO Andy Jassy has committed to around $200 billion in spending for the full year.
That pace has left marks on the financial statements. Amazon's free cash flow over the trailing twelve months turned negative at -$7.6 billion, and long-term debt climbed to $119.1 billion, up from $65.6 billion a year earlier. A sale-leaseback lets Amazon keep running its AI infrastructure while pulling billions in asset value off its balance sheet and replacing ownership with lease obligations — a shift that helps protect the company's credit profile as it continues borrowing to fund data center construction.
Rapidly Depreciating Collateral
The investors buying the debt would be lending against hardware that loses value quickly by the standards of infrastructure assets. Grace Blackwell is already one generation behind Nvidia's newer Vera Rubin platform, which Nvidia says delivers 35 times lower token costs than Grace Blackwell Ultra. The company has also estimated roughly $25 billion in revenue per gigawatt for Grace Blackwell deployments and about $40 billion per gigawatt for Vera Rubin — a gap that illustrates how quickly each new generation changes the economics of the previous one.
Amazon CEO Andy Jassy addressed the lifespan question on the company's July earnings call, saying servers take "a little less than three years" to break even and last "at least five to six years." Under the leaseback structure, Amazon would keep the early, most profitable years of each server's life, while lenders would hold claims on the later years, when the chips are one or two generations old and must be refinanced or resold.
Nvidia's Own Circular Financing Debate
The reported Amazon deal lands amid a broader discussion about circular financing in the AI build-out. Nvidia itself has embraced similar structures: the company reported $279 billion in supply commitments alongside guarantee obligations capped at $108.5 billion for AI cloud and data center partners, and it has signed up six asset managers and banks to raise more than $500 billion of outside capital for AI infrastructure.
Management has been candid about how this looks. "We know some will call this circular financing," the company said. Nvidia expects the AI labs it supports to account for roughly a quarter of its business next year, and says it can currently supply only about 70 percent of the demand customers are requesting.
What to Watch Next
Three signals will show whether the Amazon transaction becomes a template for the industry. First, Amazon's third-quarter earnings report — for which the company has guided to $22.5 billion to $26.5 billion in operating income — will reveal whether the vehicle appears in its disclosures and how much lease obligations grow. Second, the interest rate investors demand on the debt will show whether lenders are pricing in the depreciation risk that equity investors have so far largely ignored. A wide spread would suggest lenders see aging-chip risk clearly; a narrow one would suggest the market views AI hardware as durable collateral.
Third, whether other cloud giants announce similar structures within the next two quarters. If they do, an increasing share of the AI hardware bill will sit off the balance sheets that investors read carefully, and lenders will be left holding chips that each new Nvidia launch makes progressively less valuable.
For now, the deal captures the strange moment the AI industry occupies: demand for computing power is intense enough that Amazon can raise billions against year-old chips, yet the chips age fast enough that whoever holds them is making a bold bet on how long relevance lasts.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
