Amazon is preparing to offload roughly $8 billion worth of its artificial intelligence chips and then lease them back, according to a report by SiliconANGLE. The move would convert a growing chunk of the company's AI compute hardware into a financing arrangement rather than a balance-sheet asset it owns outright.

The reported figure — $8 billion — is substantial by any measure of AI infrastructure spending, and it points to the scale of the accelerator fleet Amazon has assembled to train and run machine learning workloads for Amazon Web Services and its own services. Under a sale-and-leaseback structure, Amazon would sell the chips to a financial partner and simultaneously sign leases to keep using them, freeing capital while retaining access to the compute capacity.

The report did not specify which chips would be included in the transaction, the identity of the buyer or lessor, the duration of the leases, or the expected financial terms. Amazon has built its AI infrastructure on a mix of in-house silicon and merchant hardware: its custom accelerators include the Trainium family for training and the Inferentia family for inference, deployed across AWS regions, alongside GPUs supplied by vendors such as Nvidia.

For a company of Amazon's size, an $8 billion sale-leaseback is less about affording the hardware than about how it accounts for it. Leasing shifts massive capital expenditure into operating expense lines, which can flatter free cash flow metrics that investors watch closely as AI capex balloons across the hyperscaler tier. Amazon, Microsoft, Alphabet and Meta have all sharply raised infrastructure spending over the past two years to secure AI compute, and financing structures have become one of the levers available to manage the reported cost of that buildout.

The AI chip market context makes the timing notable. Demand for accelerators has outstripped supply for extended stretches, and hyperscalers have committed tens of billions of dollars to capacity, in-house silicon programs and multi-year supplier agreements. Against that backdrop, treating AI chips as assets that can be sold and leased back signals confidence that the hardware will hold productive value over the life of the leases — or at least that Amazon can structure terms that protect it if the technology cycle turns faster than expected.

SiliconANGLE characterized the plan as a report rather than a confirmed transaction, and Amazon has not publicly detailed the arrangement. The company has not disclosed how many accelerators the $8 billion represents, nor whether the figure covers Trainium and Inferentia chips, third-party GPUs, or both.

If the deal proceeds, it would rank among the larger hardware sale-leaseback financings in the technology sector and could set a template for other cloud providers wrestling with the accounting weight of their AI fleets. How Amazon's rivals — and its chip suppliers — respond to that financing model may shape how the next wave of AI data center capacity gets funded.