The news
Amazon (AMZN) has held talks with investors about a deal to move roughly $8 billion of Nvidia (NVDA) chips into a separate financing vehicle and then lease them back, the Financial Times reported on October 2, 2026, citing people familiar with the matter. The Amazon Nvidia chip sale-leaseback would take expensive AI hardware off Amazon's own books while letting it keep using the chips.
According to the report, as relayed by Reuters and published by The Star, the chips are Nvidia Grace Blackwell accelerators installed across more than a dozen US data centers in five states, including Nevada and Virginia. Amazon would move them into a special-purpose vehicle (SPV), a separate legal entity set up to hold specific assets, and lease them back from it.
The SPV would raise money from outside investors by issuing debt, and Amazon would offer investors an equity stake of up to 10% in the vehicle, according to the reports. SiliconANGLE said the talks had been going on for several weeks and that investor names were not disclosed.
The talks are not a finished deal. Amazon has not announced the plan, and Amazon and Nvidia did not immediately respond to Reuters for comment outside regular business hours. Lease length, pricing and the investors involved were not reported.
SiliconANGLE estimated that $8 billion would cover roughly 114,000 to 133,000 Grace Blackwell units, based on per-unit prices of $60,000 to $70,000. That is the outlet's calculation, not a figure from Amazon or the FT.
The numbers
- Value of chips in reported talks
- About $8 billion
- Equity stake offered to investors
- Up to 10% of the vehicle
- Data centers holding the chips
- More than a dozen, in five US states
- SiliconANGLE's unit estimate
- 114,000 to 133,000 chips
Why CEOs should care
For CFOs, this is a sign of where AI financing is heading. A sale-leaseback turns a large upfront purchase into ongoing lease payments and moves the asset onto someone else's books. It frees cash and can make a balance sheet look lighter, but the company still owes the lease payments. Finance teams evaluating their own AI infrastructure should ask whether leasing GPUs, or buying capacity from providers that lease them, changes their cost per unit of compute and their exposure if the hardware ages faster than expected.
For investors and boards, the key question is visibility. When AI spending moves into special-purpose vehicles and long leases, headline capital-expenditure figures can understate how much a company has committed to. SiliconANGLE cited Wall Street Journal reporting that Amazon, Google and seven other large tech companies hold about $3 trillion in off-balance-sheet obligations. Analysts and audit committees should read lease footnotes as closely as capex lines.
For AWS customers, the deal would not change the service they buy. But it does show that the economics behind cloud AI capacity increasingly depend on outside lenders, whose appetite and interest rates can affect how fast new capacity gets built and what it costs.
The bigger picture
Large tech companies have been finding new ways to fund AI data centers without carrying the full cost themselves. SiliconANGLE pointed to Meta selling an 80% stake in its Hyperion data center project to an investment firm. Using GPUs themselves as collateral or lease assets extends that approach from buildings and land to the chips inside, which lose value faster as newer generations arrive.
What’s next
Watch for whether Amazon confirms or completes the transaction, who the investors are, what lease term and rate it carries, and how Amazon describes such arrangements in its next quarterly filing. Similar moves by Microsoft, Google or Meta would confirm a broader shift.
What “Fact-checked” means
Fact-checking means testing a story’s facts against the evidence before it is published. This story went through at least two separate checks before this version was published.
- What we checked
- Its names, figures, dates, job titles, quotes and who said what were checked against the story’s sources, including its main source where it could be opened. The headline was checked for accuracy and overstatement.
- How
- A first check reviewed the whole story. If it passed, a second, skeptical check went back to the sources to look for mistakes in the most important facts. If a check flagged the story, it was edited to fix the problems found, and a separate re-check then reviewed the whole story again.
- Who
- The checks are made by our newsroom, as steps kept separate from the writing, under rules set by our editor, Hussein Mukhtar. A story the checks still flag is held for the editor, who decides whether it is fixed, published or dropped.
- If something is wrong
- “Fact-checked” does not mean error-free. If a material error is found after publication, we correct the story and add a note saying what changed. Report an error
