Amazon is exploring a deal to sell roughly $8 billion worth of Nvidia AI chips to a group of investors, according to a Benzinga report. The move signals how hyperscalers are rethinking the balance-sheet math behind AI buildouts as capital costs for compute keep climbing.

The reported structure would shift the economics of Amazon's AI hardware fleet. Rather than carrying the full cost of Nvidia accelerators on its own books, Amazon would sell the chips to outside investors and effectively lease back the capacity. That converts a massive capital expenditure into an operating arrangement spread over time, freeing capital for further data center expansion while giving investors exposure to hardware that underpins rented AI compute.

The figure at the center of the report is large even by AI-infrastructure standards. An $8 billion basket of Nvidia silicon would represent one of the biggest third-party financing deals yet seen around merchant AI accelerators, a category that already dominates hyperscaler capex. Nvidia's data center GPUs and networking have become the single largest line item in most large-scale AI budgets, and financing vehicles that move those purchases off balance sheet have drawn growing interest across the industry.

For Amazon, the calculus is straightforward. AI infrastructure spending has surged faster than most revenue lines can absorb, and Wall Street has begun scrutinizing the gap between capex and monetized AI demand. Selling chips to investors and leasing them back would let Amazon keep deploying the compute it needs for AWS AI services and its own models without reporting the full hardware outlay as traditional capital expense. Benzinga's report frames the plan as a direct response to those surging AI infrastructure costs.

For investors, the appeal lies in the scarcity and utilization of the hardware. Top-tier AI accelerators remain supply-constrained, and cloud providers continue to rent out that capacity at premium rates. A chip-leasing arrangement offers a claim on assets with demonstrated demand — an increasingly common pitch as private capital hunts for yield tied to AI compute.

The deal also lands amid intensifying pressure on Nvidia's biggest customers. Microsoft, Google, Meta and Amazon have all committed tens of billions of dollars to AI data centers, and several have explored alternative structures — including joint ventures, special-purpose vehicles and vendor financing — to stretch those budgets. Amazon's reported $8 billion plan would be among the most explicit attempts yet to bring outside investor money directly into the AI hardware layer.

Nvidia, for its part, benefits either way. Whether a hyperscaler buys chips outright or finances them through third parties, the accelerators ship and the revenue books. Financing structures like the one Amazon is reportedly considering could even expand the addressable market by letting capital-constrained buyers join the queue.

Benzinga characterizes the plan as under consideration rather than finalized, and neither Amazon nor Nvidia has publicly confirmed the structure or the $8 billion figure. If Amazon proceeds, expect other hyperscalers to move quickly toward similar arrangements, deepening the pool of private capital funding the AI buildout and reshaping how chip purchases are financed across the industry.