Amazon is seeking to sell roughly $8 billion worth of Nvidia chips to investors, according to a report carried by Yahoo Finance — a figure that, if confirmed in detail, would mark one of the largest attempted transfers of AI accelerator hardware from a hyperscaler's balance sheet to outside financial ownership.

The reported plan centers on Nvidia's data-center GPUs, the chips that power the bulk of Amazon Web Services' artificial intelligence training and inference offerings. Amazon, like Microsoft, Google and Meta, has committed tens of billions of dollars to acquiring such hardware from Nvidia over the past two years, and the reported $8 billion disposition effort would represent a meaningful share of that accumulated position.

The mechanism under discussion, per the report, is a sale to investors rather than a disposal through the used-equipment channel. That structure matters commercially. A direct sale to financial parties could move the chips — and potentially the depreciation burden that comes with them — off Amazon's books while keeping the hardware inside its cloud ecosystem through resale or leasing arrangements. The report does not specify which Nvidia product families are involved, whether the deal covers Hopper-generation or Blackwell-generation parts, or at what discount to original purchase price the chips would change hands.

The reported move comes as hyperscalers face mounting pressure over capital intensity. Building and equipping AI data centers has pushed combined annual capex at the largest cloud providers toward several hundred billion dollars, and investors have begun questioning how quickly that spending converts into revenue. Offloading accelerator hardware to third-party investors would let a cloud provider recoup part of its outlay while retaining operational use of the silicon — a structure similar in spirit to sale-leaseback arrangements used in real estate and aircraft finance.

It also reflects the peculiar economics of the current AI buildout. Nvidia's top-line accelerators carry list prices in the tens of thousands of dollars per unit for Hopper-class parts, with Blackwell-class chips reportedly commanding more, and delivery slots for the newest parts remain allocated rather than freely available. That scarcity gives even previously deployed hardware residual value, and a pool of $8 billion in chips implies tens of thousands of physical units if priced near current flagship levels — though the report does not break the figure down by product, count or vintage.

Any such transaction would also carry second-order effects for Nvidia. A liquid secondary market for its accelerators could, over time, give cloud customers an alternative procurement channel and create pricing visibility for used silicon that does not exist today. Nvidia has historically tightly controlled its distribution, selling primarily through direct OEM and hyperscaler relationships, and the company has not commented on the reported Amazon effort.

For buyers on the investor side, the appeal would rest on the gap between Nvidia's constrained new-unit supply and surging demand for AI compute. Renting accelerator capacity at premium hourly rates has become a core revenue line for AWS through services built on Nvidia hardware, and investors taking ownership of the underlying chips would in effect be underwriting that rental stream.

Much remains unconfirmed. The report does not identify prospective investors, say whether talks have produced term sheets, or indicate a timeline for any transaction. Amazon and Nvidia have not publicly confirmed the $8 billion figure, and hyperscalers routinely explore financing structures that never reach completion.

If Amazon proceeds, the deal would set a precedent for how the industry's largest AI hardware owners manage their silicon positions — and how quickly a secondary market for data-center GPUs emerges alongside Nvidia's primary sales channel.