Amazon has asked investors to fund roughly $8 billion of Nvidia AI chips and then lease the hardware back to the company, according to a report by Moneycontrol. The structure, if executed, would shift a substantial block of accelerator capital expenditure off Amazon's balance sheet while keeping the compute capacity available to its cloud operations.

The reported arrangement targets the most expensive asset class in modern data center construction: Nvidia's AI accelerators, which now dominate spending on AI infrastructure among hyperscalers. Under a sale-and-leaseback model, an outside investor — or a consortium of investors — would purchase the chips outright. Amazon would then pay to use them over a fixed contract term, converting an upfront capital outlay into a recurring operating expense.

That distinction matters for how the market reads hyperscaler finances. Chip purchases today are treated as capital expenditure, a line that has swollen across the industry as cloud providers race to build out AI training and inference capacity. Moving $8 billion of Nvidia silicon onto someone else's balance sheet would flatter Amazon's reported capex intensity and free balance-sheet capacity for other investments, while the lease payments would flow through operating results instead.

For the investors, the pitch is the same one that underpins the broader AI financing boom: long-term contracted cash flows from a creditworthy hyperscaler tenant, secured against scarce compute hardware. Sale-and-leaseback structures are well established in real estate, aircraft, and telecom towers. Applying them at this scale to GPUs is newer territory, and it carries a specific risk profile: accelerators depreciate quickly as Nvidia ships successive product generations, so the residual value of the collateral depends heavily on contract length and how the parties treat end-of-term disposition of the chips.

The reported plan also illustrates how tightly hyperscaler AI buildouts remain coupled to Nvidia's supply. Amazon designs its own AI accelerators — the Trainium and Inferentia families — and has been deploying them across AWS, but its most advanced customer-facing AI offerings still depend heavily on Nvidia's GPUs. An $8 billion commitment of this size would rank among the larger single-vendor hardware financings in the sector and signals that demand for Nvidia silicon at AWS continues to outpace what Amazon's in-house silicon can cover alone.

The financing could also set a template for peers. Microsoft, Google, and Meta face the same accounting pressure: AI infrastructure spending has grown faster than revenue from AI services, and investor scrutiny of capex discipline has intensified across the group. If Amazon closes the deal on workable terms, sale-and-leaseback structures for GPUs could become a standard tool for managing the sector's hardware bills.

Moneycontrol's report did not specify which investors Amazon has approached, the proposed lease terms, the timeline for a deal, or which Nvidia chip families the $8 billion would cover. Amazon and Nvidia have not publicly confirmed the plan.

If the structure proceeds, it will offer the first large-scale test of whether institutional capital will accept GPUs — a fast-depreciating asset — as leaseback collateral, and at what implied cost of capital that appetite comes.