Amazon is seeking to offload roughly $8 billion worth of Nvidia chips to investors, the Financial Times reported, in a deal that would turn one of the largest AI accelerator fleets in the world into a vehicle for outside capital.

The report, surfaced via Yahoo Finance on Friday, gives no counterparty names and no closing timeline. What it does establish is scale: $8 billion of Nvidia silicon is a meaningful number even by hyperscaler standards, and it signals that Amazon is actively looking for financial structures that move GPU assets — or the obligations tied to them — off its own balance sheet.

Nvidia does not publish per-customer shipment figures, and neither Amazon nor the FT report specifies which product families are involved. At current list pricing, however, $8 billion corresponds to a fleet measured in the tens of thousands of top-end accelerators, whether those are H100-class parts, the newer Blackwell generation, or a mix across both. The report does not break this down, so any figure on unit counts remains an estimate rather than a confirmed number.

The commercial logic is straightforward. AI accelerators sit at the center of a capital-intensity problem: hyperscalers committed to massive GPU orders through 2023 and 2024, and the depreciation schedules, financing costs and supply commitments tied to those orders now weigh on margins. Selling chips to investors — and effectively leasing capacity back, or contracting compute from entities that own the hardware — would free capital while keeping the compute available to AWS customers. The FT report describes the effort as a search for investors to take the chips; it does not confirm a completed structure or a price.

For Nvidia, the report cuts both ways. Demand from financial buyers would underscore how far AI hardware has moved toward being an asset class of its own — income-producing infrastructure that pension funds, private credit or infrastructure funds want exposure to. At the same time, hyperscalers offloading chips suggests customers are testing ways to cap their own exposure to the hardware cycle, which could reshape how future GPU orders are financed.

The deal, if completed at anything near the reported $8 billion figure, would rank among the largest secondary transactions of AI silicon to date. It would also set a template other cloud providers could copy as they balance buildout commitments against shareholder pressure on capital spending.

No party has commented publicly beyond the FT's reporting, and Amazon has not confirmed the figure. Whether investors accept Nvidia-hardware valations that assume multi-year useful life for chips that refresh roughly every two years will determine whether this becomes a recurring financing channel or a one-off.