What happened
Amazon is talking to investors about moving roughly $8 billion worth of Nvidia AI chips into a new investment vehicle — and then leasing them right back. The Financial Times broke the story on Friday, citing people familiar with the matter; Reuters and MarketWatch have since confirmed the reporting.
Here’s the key detail most headlines gloss over: Amazon isn’t giving up the chips. It would keep using them. This is a sale-leaseback — a financing structure, not a fire sale.
The details
The chips in question are Nvidia’s Grace Blackwell processors — the current top tier of AI silicon — spread across more than a dozen Amazon data centers in five US states, including Nevada and Virginia. Under the proposed deal, Amazon would transfer them into a special-purpose vehicle (SPV), a standalone legal entity created to hold the assets. The SPV would then raise money from outside investors through debt issuance, while Amazon leases the chips back for its own use. Amazon may also offer investors an equity stake of up to 10% in the vehicle.
The deal logic is straightforward. Amazon plans to spend about $220 billion on capital expenditure this year, almost double last year’s spend, and CEO Andy Jassy has said demand is still outrunning supply. Moving $8 billion of depreciating hardware off the balance sheet would let Amazon keep building compute capacity while protecting its margins and credit profile. The debt investors are reportedly expecting an investment-grade rating backed by Amazon’s own AA credit standing rather than the hardware itself.
The same day, Amazon announced it is raising prices by about 15% for EC2 Capacity Blocks — the rental slots that give customers access to Nvidia chips ranging from the older A100 to the newer B300 — starting next week. Amazon also raised some rental prices in July. Amazon did not immediately respond to requests for comment on either move.
Why it matters
This is the clearest signal yet of how the AI infrastructure boom is forcing big tech into Wall Street-style financial engineering. Hyperscalers are spending hundreds of billions on data centers, and the creative deals are starting to look like the property market: sell the building, rent it back, keep using it. The Wall Street Journal reported that the nine largest US tech companies carried roughly $3 trillion in off-balance-sheet commitments as of August — their real spending plans may be far larger than they appear.
For the AI-curious subscriber, the takeaway is simpler: when even Amazon starts asking investors to carry the cost of its chips while it raises rental prices on the same hardware, compute is becoming scarce enough that everyone pays more. The era of cheap, plentiful GPU capacity is not coming back anytime soon.
Frequently asked questions
Is Amazon dumping Nvidia chips?
No. Under the proposed sale-leaseback, Amazon would keep using the same chips — they’d just be owned by outside investors through a special-purpose vehicle while Amazon leases them back.
Why would Amazon do this?
To get $8 billion of depreciating hardware off its balance sheet while continuing to use it, preserving its margins and credit rating as it ramps toward ~$220 billion in annual capital spending.
Did Amazon also raise GPU rental prices?
Yes — on the same day, Amazon announced ~15% price increases for EC2 Capacity Blocks, which rent access to Nvidia chips from the A100 through the B300, effective next week.
Is this confirmed?
Not yet — the FT report cites people familiar with the talks, and Amazon has not commented. Treat it as a proposed deal in progress.
Sources: Financial Times, Reuters, MarketWatch, Investopedia.