Amazon is exploring a deal that could shift about $8 billion of Nvidia AI chips to outside investors while allowing the company to continue using the hardware through a leaseback arrangement. The proposed structure would move thousands of Grace Blackwell processors into a special-purpose vehicle as Amazon looks for ways to finance an unprecedented wave of spending on artificial intelligence infrastructure.
The plan, first reported by the Financial Times and summarized by Reuters, reflects how expensive the AI buildout has become even for the largest technology companies. Amazon is still investing aggressively through Amazon Web Services, but the potential transaction would let it reduce the amount of high-cost hardware sitting directly on its balance sheet while preserving access to the computing capacity.
Amazon Nvidia Chips Deal Could Shift $8 Billion Off Balance Sheet
Amazon has held talks with investors about transferring roughly $8 billion of advanced Nvidia chips into a special-purpose vehicle. The proposed entity would raise outside financing and then lease the processors back to Amazon, according to the report.
The company could offer investors an equity stake of up to 10% in the vehicle while debt issued by the SPV would provide most of the funding. The chips involved were either purchased or leased by Amazon and are already being installed in more than a dozen U.S. data centers across five states, including Nevada and Virginia.
Amazon and Nvidia did not immediately comment on the reported transaction. Because the structure is still under discussion, the final amount, ownership terms and financing costs could change before any agreement is completed.
Why Amazon Wants a More Asset-Light AI Strategy
AI infrastructure is forcing cloud companies to commit extraordinary sums to chips, data centers, power and networking equipment. Amazon raised its 2026 capital-spending forecast to about $220 billion after AWS growth accelerated, with much of that money directed toward computing capacity and artificial intelligence.
The company has also said demand continues to exceed available capacity in parts of its cloud business. That gives Amazon a reason to keep expanding aggressively, but it also creates pressure on free cash flow and increases the amount of long-lived hardware whose value can decline quickly as newer AI accelerators arrive.
A leaseback structure can address part of that problem. Instead of owning every processor directly, Amazon could pay investors for continued use of the chips while outside capital absorbs some of the upfront cost and depreciation risk.
AI Financing Is Becoming a Bigger Wall Street Business
The proposed Amazon structure fits a broader shift in the artificial-intelligence boom. Technology companies and infrastructure operators are increasingly using debt, special-purpose vehicles and private capital to fund data centers and computing equipment that previously would have been financed more directly on corporate balance sheets.
American News Brief recently examined a related trend in AI financing tied to Broadcom and Anthropic, where chip suppliers and technology companies are taking on a larger role in financing the customers buying their infrastructure. That pattern blurs the traditional line between vendor, lender, landlord and strategic investor.
The financing boom has attracted interest from banks, private-credit firms, insurers and asset managers looking for exposure to contracts backed by large technology companies. Investors can earn yields tied to infrastructure that is expected to remain heavily utilized, while hyperscalers gain another source of capital outside ordinary corporate bond issuance.
Nvidia Grace Blackwell Chips Are Central to the Deal
The processors involved are part of Nvidia’s Grace Blackwell family, among the most valuable components in current AI data centers. Blackwell systems are designed for training and running large artificial-intelligence models, and demand for them has remained strong across cloud companies, AI laboratories and enterprise customers.
American News Brief has followed the larger Nvidia spending cycle, including how Nvidia earnings have become a key test of whether the AI infrastructure boom can sustain its current pace. Amazon’s proposed financing arrangement underscores the same issue from the buyer’s side: demand may be strong, but acquiring enough hardware requires enormous amounts of capital.
The transaction would also deepen the practical relationship between Amazon and Nvidia even though AWS develops its own Trainium AI processors. Amazon uses Nvidia hardware to serve customers that want the industry-standard CUDA ecosystem, while promoting its own chips as a lower-cost alternative for some workloads.
Data Centers in Virginia and Other States Carry the Hardware
The reported chips are installed across more than a dozen data centers in five states. Virginia is especially important because Northern Virginia hosts the world’s largest concentration of data-center capacity and remains a core AWS region.
That physical footprint has become a political and economic issue in its own right. Data centers require large amounts of electricity, cooling infrastructure, land and transmission capacity, and state and local governments are increasingly debating how much of those costs should be borne by technology companies rather than ordinary utility customers.
Amazon announced separately Friday that AWS plans to invest $1 billion over five years in communities that host its U.S. data centers. The company said the money would support education, workforce development, energy affordability and resource-preservation projects as public scrutiny of data-center expansion increases.
The Bigger Question Is Who Finances the AI Boom
The proposed $8 billion chip transaction does not indicate that Amazon is pulling back from artificial intelligence. On the contrary, the company is trying to find additional ways to sustain a spending program that has grown large enough to influence bond markets, power demand and the balance sheets of some of the world’s biggest companies.
The important question for investors is whether off-balance-sheet or leaseback structures genuinely improve capital efficiency or simply move obligations into less obvious forms. Amazon would still need to make lease payments and depend on the hardware to generate enough cloud revenue to justify the cost.
If the deal is completed, it could provide a template for other hyperscalers facing the same problem: how to keep buying expensive AI chips without allowing infrastructure spending to overwhelm cash flow. The outcome will therefore matter not only for Amazon and Nvidia, but also for the broader financing model behind the U.S. AI investment boom.
