Amazon’s Reported $8B Nvidia Chip SPV: How AI Capex Is Moving Off Big Tech Balance Sheets

Amazon’s AI spending problem is becoming a financing problem—in the neutral sense of the word. Demand for AWS computing remains strong, but the physical infrastructure needed to serve that demand has become so expensive that even the world’s largest technology companies are looking for new ways to fund chips and data centers. On October 2, the Financial Times reported that Amazon has been discussing a special-purpose vehicle that would hold roughly $8 billion of Nvidia Grace Blackwell chips and lease them back to Amazon.

The proposed structure has not been confirmed by Amazon or Nvidia, and the talks could change. That distinction is important. What is confirmed, however, is the scale of Amazon’s AI build-out: the company raised its 2026 cash capital-spending plan to roughly $220 billion in July, and its latest filings show a sharp increase in technology-infrastructure investment and financing activity. The reported SPV therefore fits an already visible trend rather than appearing out of nowhere.

What Happened

According to the Financial Times report cited by Reuters, Amazon has held discussions with outside investors about moving thousands of Nvidia Grace Blackwell chips into a special-purpose vehicle. The chips are already being installed in more than a dozen U.S. data centers across five states, including Nevada and Virginia.

Under the reported structure, the SPV would own the hardware and raise capital from external investors, primarily through debt. Amazon would lease the chips back and continue using them for AWS and AI workloads. The company could reportedly offer outside investors up to a 10% equity stake in the vehicle.

This is effectively a form of asset-backed or sale-and-leaseback financing. Amazon would still receive the economic use of the GPUs but would not necessarily hold the same amount of hardware directly on its own balance sheet. The arrangement could preserve liquidity, reduce upfront ownership requirements and transfer some residual-value risk to outside capital providers.

Neither Amazon nor Nvidia had publicly confirmed the plan when Reuters reported the story. The correct way to read the event is therefore “Amazon is reportedly exploring” rather than “Amazon has completed” an $8 billion transaction.

Why It Matters

The financing method matters because AI infrastructure is becoming one of the largest capital cycles in corporate history. Amazon raised expected 2026 cash capital spending to about $220 billion after AWS growth accelerated. In the second quarter alone, Amazon reported $53.1 billion of cash capital expenditures, with the majority tied to technology infrastructure supporting AWS and generative AI.

That spending is producing revenue. AWS sales rose 37% year over year in Q2 to $42.2 billion, and management said demand still exceeded available capacity. But the build-out is also consuming cash. Amazon reported trailing-12-month free cash flow of negative $7.6 billion at the end of Q2, largely because purchases of property and equipment increased sharply.

The tension is straightforward: the company wants to build faster because customers want more compute, but every new GPU cluster, power connection and data-center shell increases near-term capital intensity. Externalizing ownership of some chip assets can help Amazon keep investing without funding every dollar with internal cash or conventional corporate debt.

Market Impact

For Amazon shareholders, the potential benefit is flexibility. If the company can match long-lived customer demand with long-lived financing, it can reduce the cash-flow volatility created by massive upfront hardware purchases. That may matter more as memory, networking equipment and GPU systems become more expensive.

The cost is complexity. Leaseback structures do not make the economic cost of the hardware disappear. Amazon would still pay to use the chips, and outside investors would require a return. The company could also create long-term contractual obligations that behave like debt even if the accounting presentation differs from direct ownership.

For Nvidia, the development is strategically important. GPU demand is no longer constrained only by chip supply. Financing capacity is becoming part of the demand equation. If hyperscalers, neoclouds and AI laboratories can finance GPUs through asset-backed vehicles, more customers can continue ordering hardware even when traditional balance sheets are stretched.

The broader market implication extends to banks, private credit, infrastructure funds and insurance investors. AI compute is beginning to resemble an infrastructure asset class. Investors are being asked to underwrite the useful life, utilization, residual value and technological obsolescence of chips in much the same way lenders analyze aircraft, data centers or energy projects.

Key Data and Timeline

Amazon’s capital intensity was already rising before the reported SPV. In its Q2 filing, the company said cash capital expenditures reached $53.1 billion in the quarter and $96.3 billion for the first six months of 2026. Financing cash inflows also increased, and Amazon said it expected additional financing activity during the year.

In July, management raised expected 2026 cash capital spending to roughly $220 billion, citing strong AWS demand and higher infrastructure costs. AWS Q2 revenue reached $42.2 billion, up 37% from a year earlier.

The reported chip-financing talks became public on October 2. No closing date, final investor group or definitive financing terms had been announced as of the data cutoff. Amazon has also not formally announced its Q3 earnings date, although late October is the normal reporting window.

Market Debate

The bullish interpretation is that sophisticated financing allows Amazon to keep building scarce AI capacity while preserving balance-sheet flexibility. If the underlying AWS contracts are long duration and utilization remains high, matching those cash flows with dedicated asset financing can be rational.

The bearish interpretation is that off-balance-sheet or leaseback-style structures can obscure the true capital burden of the AI boom. Investors still need to include lease payments and contractual obligations when assessing returns on invested capital. If GPU prices fall quickly or new generations make older chips economically obsolete, residual-value assumptions could also prove too optimistic.

There is another strategic debate around custom silicon. Amazon is investing heavily in Trainium and Graviton while also buying large amounts of Nvidia hardware. The coexistence of both approaches suggests AWS customers still demand Nvidia at scale even as Amazon tries to improve economics through its own chips.

Risks

The first risk is technological obsolescence. Advanced GPUs can be economically valuable for years, but the pace of AI hardware improvement is unusually fast. Investors financing current-generation chips must estimate what those assets are worth after newer Nvidia, AMD or custom silicon enters service.

The second risk is utilization. The model works best if AI demand remains strong enough to keep the hardware busy. A slowdown in model training or cloud demand would make fixed lease obligations less attractive.

The third risk is transparency. Investors should evaluate the full economic obligation, not just whether an SPV sits outside Amazon’s consolidated hardware ownership. Accounting structure does not eliminate cash commitments.

What to Watch Next

Watch for an Amazon filing or official statement confirming whether the SPV is formed, its size, debt structure and ownership. The identity of outside investors and the credit terms would reveal how capital markets are pricing AI-chip risk.

Also watch Amazon’s Q3 cash flow and capex commentary. If AWS growth remains near current rates while free cash flow improves, the financing model will look more like optimization. If free cash flow stays deeply negative and more assets move into external vehicles, investors may start asking whether AI infrastructure is becoming too capital intensive even for hyperscalers.

Conclusion

The reported $8 billion Nvidia-chip SPV is not evidence that Amazon is retreating from AI. It points in the opposite direction: the company is trying to sustain an infrastructure build-out so large that the financing architecture itself is becoming strategic. The key question is no longer simply how many GPUs Amazon can buy. It is how efficiently the company can finance, depreciate and monetize them. If dedicated chip-financing vehicles become standard, the next phase of the AI boom may be shaped as much by credit markets as by semiconductor supply.