# Amazon Wants $8 Billion of Nvidia Chips to Vanish From Its Balance Sheet

By Abdullah Al Foysal · Finance · Published Mon, 05 Oct 2026 09:54:19 GMT
Source: The Current Tribune — https://currenttribune.com/article/amazon-8-billion-nvidia-chips-balance-sheet-deal

Amazon wants $8 billion worth of Nvidia’s best chips to technically not belong to Amazon anymore, even though Amazon will still be the one plugging them in and running workloads on them. It’s a financial maneuver that sounds like a magic trick, and in a lot of ways, it is one. Welcome to the increasingly strange world of AI infrastructure financing, where the chips are real, the compute is real, but the balance sheet treatment is anything but straightforward.

## What Amazon Is Actually Trying to Do

According to reporting that traces back to the Financial Times, Amazon is exploring a deal structured as a sale-leaseback: the company would sell a chunk of its top-end Nvidia chips, worth roughly $8 billion, to outside investors, then lease that same hardware back to keep using it in its data centers. The chips never move. The servers never go dark. What changes is who technically owns the asset on paper, and crucially, who’s carrying the depreciation and capital expenditure load on their books.

This isn’t Amazon inventing a new trick. Sale-leaseback arrangements have existed in corporate finance for decades, typically used for real estate, aircraft, and industrial equipment. What’s new is applying that structure to AI accelerators that lose relevance every eighteen to twenty-four months as Nvidia ships faster chips. That compressed useful life is exactly why this kind of deal is suddenly interesting to Wall Street, and exactly why it should raise an eyebrow or two.

### Why “Off the Books” Actually Matters Here

Keeping tens of billions of dollars in hardware on a balance sheet isn’t free. It drags down return-on-assets calculations, inflates depreciation schedules, and makes capital expenditure guidance look scarier to investors who are already nervous about how much the hyperscalers are spending chasing AI capacity. Moving that $8 billion in chips to an outside special-purpose vehicle, with Amazon paying a lease fee instead of owning the depreciating asset outright, cleans up those metrics without actually reducing Amazon’s reliance on the hardware for a single day.

## This Is Part of a Much Bigger Pattern

Amazon isn’t out here improvising. Big Tech’s AI capital expenditure is on pace to hit roughly $600 billion in 2026 across the major hyperscalers, a figure that’s forced finance teams at every one of these companies to get creative about how that spending shows up in quarterly filings. Off-balance-sheet financing structures, special-purpose vehicles backed by chip collateral, and leasing arrangements with third-party infrastructure funds have all become more common as the sheer scale of AI buildout spending has outpaced what companies want sitting directly on their own books.

Context matters here too. Amazon has reported hyperscaler earnings alongside Meta and Microsoft that beat expectations, with all three stocks rising on strong results tied directly to AI infrastructure demand. AWS, in particular, has leaned on pricier AI chip rentals as a lever to accelerate revenue growth, meaning the chips at the center of this proposed deal are also central to one of Amazon’s fastest-growing and most profitable business lines. That’s part of why this specific $8 billion tranche matters: it’s not surplus hardware sitting idle. It’s active, revenue-generating infrastructure that Amazon wants to keep using while reshaping how it appears in the company’s financial statements.

### Who Benefits on the Other Side of the Table

For the investors on the other end of a sale-leaseback like this, the appeal is a steady, contracted lease payment backed by one of the most creditworthy companies in the world, secured against hardware that currently has enormous demand. It’s a lower-risk way to get exposure to the AI infrastructure boom without the operational headache of actually running a data center or negotiating power contracts. Infrastructure funds and private credit shops have been circling exactly this kind of opportunity as traditional bank lending to hyperscalers has grown more cautious given how much capital the sector is absorbing all at once.

## The Risk Nobody Wants to Say Out Loud

Here’s the uncomfortable part of this story. Chips depreciate fast, and AI accelerators depreciate faster than almost anything else in a data center. A sale-leaseback works cleanly when the underlying asset holds value over the life of the lease. Nvidia’s top-end chips are historically obsolete, relative to the newest generation, within two to three years. If whoever buys this $8 billion tranche is locked into a long-term lease assuming the hardware retains meaningful value, and Nvidia’s next architecture makes it look dated faster than expected, that’s a mismatch that someone eventually has to eat. Right now, that someone looks like it’s going to be whoever is financing the other side of this deal, not Amazon.

That’s precisely the dynamic that’s made some analysts nervous about the broader wave of off-balance-sheet AI financing spreading across the industry. When the underlying collateral is compute hardware with a rapidly shrinking shelf life, these structures start to resemble the kind of financial engineering that looks fine right up until growth assumptions stop holding, at which point the mismatch between asset life and lease terms becomes everyone’s problem at once.

## What This Means

Amazon moving $8 billion in Nvidia chips off its balance sheet doesn’t change what AWS can actually do with that hardware tomorrow morning. It changes how the spending looks to investors parsing Amazon’s next earnings report, and it signals that even a company with Amazon’s cash position and credit profile would rather spread AI infrastructure risk to outside capital than carry all of it directly. That’s worth paying attention to. If the companies with the deepest pockets in tech are reaching for creative financing to manage the sheer scale of AI capital spending, it says something about just how large these numbers have gotten, even for them.

Watch for whether other hyperscalers follow with similar structures in the coming quarters, and watch even more closely for what happens to these arrangements the first time a major new chip generation makes a leased fleet look obsolete ahead of schedule. That’s the moment that will reveal whether this is smart balance-sheet management or just AI-era financial engineering waiting for its first real stress test.