Amazon.com Inc. plans to sell roughly $8 billion worth of artificial intelligence chips and then lease the same hardware back, SiliconANGLE reported, citing people familiar with the matter.

The transaction would take the form of a sale-leaseback, a financing structure that lets a company convert owned assets into cash while retaining operational use of them. Under the reported arrangement, Amazon would offload AI accelerator chips to a buyer or consortium, then lease that capacity back on a long-term basis to keep its data center operations running without interruption.

The move addresses a straightforward capital question. AI chips — the GPUs and custom accelerators that power large-scale model training and inference — rank among the most expensive line items in data center construction. An $8 billion sale would return that capital to Amazon's balance sheet while spreading the cost of using the hardware over the term of the lease.

Amazon has invested heavily in custom silicon for AI workloads. Its Trainium chips handle model training and its Inferentia line targets inference, and the company deploys both alongside processors purchased from third-party suppliers such as Nvidia Corp. The reported $8 billion portfolio covers chips deployed across Amazon's infrastructure, which backs both its AWS cloud business and the AI services built on top of it.

The structure also shifts certain risks. Chip values depreciate quickly as newer generations reach the market, and a leaseback can transfer some of that depreciation risk to the buyer, who takes formal ownership of the assets. For the lessor, the deal offers long-term contracted revenue from a creditworthy tenant.

Amazon is not alone in examining such arrangements. Sale-leaseback financing has a long history in real estate and heavy industrial equipment, and hyperscale operators have applied versions of it to data center property. Applying it directly to AI silicon at this scale would mark a notable extension of the practice, given how rapidly the underlying assets lose value compared with buildings or networking gear.

If completed, the deal would rank among the largest financing transactions tied specifically to AI hardware. It would also free capital Amazon could redirect toward additional compute purchases, data center expansion, or other investments in its AI buildout.

Neither Amazon nor potential buyers have publicly confirmed the terms. The report did not name the counterparties involved, the expected lease duration, or the pricing of the lease payments. Those details will determine the actual cost of the financing and, by extension, whether the structure proves cheaper than holding the chips on balance sheet and depreciating them over time.

The figure itself signals the scale of Amazon's AI infrastructure commitment. An $8 billion chip portfolio represents a substantial share of the compute capacity behind AWS's AI offerings, including the Trainium-based instances the company sells to customers training and running large models.

For the broader market, a transaction of this size would establish a reference point for how AI hardware can be financed. Other operators facing similar capital pressures — building out GPU fleets that cost billions per installation — could follow with comparable structures if Amazon's deal proceeds on favorable terms.

The report remains unconfirmed, and Amazon has not filed any disclosure indicating a completed agreement. Any deal of this magnitude would likely surface in future financial reporting, either as financing obligations on the balance sheet or in the company's capital expenditure commentary.