Amazon plans to sell roughly $8 billion worth of Nvidia AI chips and lease them back — a financing move designed to shore up its balance sheet while keeping the compute capacity running for its cloud business.

The transaction is a sale-leaseback: Amazon would transfer ownership of the Nvidia accelerators to a buyer, typically a financial institution or leasing specialist, and then rent the same hardware back under a long-term contract. The company converts a capital-intensive asset into cash upfront. The chips stay in the datacenters. The AI workloads do not pause.

The $8 billion figure anchors what is otherwise a straightforward piece of financial engineering. Amazon, like every hyperscaler, has been spending heavily on Nvidia GPUs to build out AI compute capacity for Amazon Web Services. Those chips sit on the balance sheet as depreciating assets. Selling them and leasing them back pulls that value off the books and back onto the balance sheet as liquidity — while the lease payments spread the cost of using the hardware over time instead of concentrating it in capital expenditure.

The timing matters as much as the mechanism. AI accelerators are the fastest-appreciating line item in cloud capex, and they depreciate fast in an accounting sense even as demand for them climbs. A sale-leaseback lets Amazon keep provisioning Nvidia compute for AWS customers without carrying the full asset weight of the fleet it has already bought. In effect, the company is monetizing chips it owns today to fund the compute it needs tomorrow.

For Nvidia, the deal is a signal rather than a direct revenue event. The chips in question were presumably purchased long ago; the sale-leaseback moves them between balance sheets, not off the market. But it demonstrates that top-tier AI silicon has become collateral-grade infrastructure — assets substantial enough, at $8 billion, for institutional buyers to underwrite. That is a new category of financial utility for GPUs, and one that other cloud operators watching their own capex lines can copy.

The move also reflects the strain that AI buildouts place on hyperscaler financials. Buying tens of thousands of Nvidia accelerators ties up capital in hardware with a short useful-life window. Leasing structures, resale, and now sale-leasebacks are the industry's emerging toolkit for reconciling the scale of AI infrastructure spending with the return profile investors expect. Amazon's approach keeps the compute in place — and the revenue it generates for AWS — while lightening the asset side of the ledger.

Details of the transaction — the buyer, the lease terms, the payment schedule — were not disclosed in the report, and Amazon has not publicly broken out how much of its AI fleet the $8 billion represents. What is confirmed is the plan itself: sell the chips, lease them back, and convert a chunk of GPU inventory into balance-sheet strength.

If the structure works as intended, expect other cloud providers with large Nvidia fleets to explore the same arithmetic — turning their AI chips from a capex burden into a financing source.