Amazon Is Moving $8 Billion of Nvidia Chips Off Its Balance Sheet — and It's Part of a $3 Trillion AI Tab
In 60 seconds: Amazon wants to move about $8 billion worth of Nvidia AI chips into an investor-backed vehicle and lease them back — keeping the debt off its own balance sheet. It's the latest sign of a much bigger phenomenon: Big Tech is carrying roughly $3 trillion in AI commitments that never show up in the headline numbers.
- The Financial Times reports Amazon has been talking to investors about shifting ~$8B of Nvidia Grace Blackwell chips into a special-purpose vehicle (SPV), then leasing them back.
- The SPV would raise debt from outside investors; Amazon would offer up to a 10% equity stake and keep no ownership — an "asset-light" way to fund its AI buildout.
- The chips sit in 12+ US data centers across five states, including Nevada and Virginia. Amazon and Nvidia declined to comment to Reuters.
- Context: a Wall Street Journal analysis found nine big tech firms carry ~$3 trillion in off-balance-sheet AI commitments — five times their reported capex.
- Why it matters: the structure protects Amazon's balance sheet if chip values fall — but makes its true AI spending harder for investors to track.
Eight billion dollars' worth of the world's most coveted computer chips — and Amazon would rather someone else owned them. The Financial Times reports the cloud giant has spent recent weeks quietly sounding out investors about moving thousands of Nvidia's Grace Blackwell AI processors into a special-purpose vehicle, then renting them straight back. You keep using the chips. Someone else carries the debt. Welcome to the new math of the AI boom.
What happened
The FT reported Friday, citing people familiar with the matter, that Amazon wants to offload roughly $8 billion of advanced Nvidia chips through a financing vehicle designed to strengthen its balance sheet. The plan: transfer thousands of Grace Blackwell chips into an SPV that raises debt from outside investors — Amazon would offer up to a 10 percent equity stake and retain no ownership — then lease the chips back and keep running them as before. The hardware was bought or leased by Amazon and sits in more than a dozen US data centers across five states, including Nevada and Virginia. Neither Amazon nor Nvidia responded to Reuters' requests for comment.
Latest verified facts
This is part of the biggest open secret in corporate finance. A Wall Street Journal analysis of securities filings in August found nine major tech companies — Alphabet, Amazon, Meta, Microsoft, Oracle, Nvidia, Broadcom, SpaceX, AMD — sitting on roughly $3 trillion in off-balance-sheet commitments, mostly tied to AI infrastructure. About $1.2 trillion is leases that haven't started yet; roughly $1.9 trillion is purchase commitments for chips, construction, and power. Under accounting rules these stay off the books until rent flows or products deliver — legal, disclosed in footnotes, and easy to underweight.
Amazon wouldn't be the first to play this game. Meta routed the debt for its giant Louisiana data center through a Blue Owl joint venture where it holds only a minority interest. As Dow Jones analysts noted, the SPV lets Amazon raise cash to pay down debt and shifts the risk of falling chip values to outside investors. The catch: the SPV's higher borrowing costs come back as higher lease payments — and the arrangement makes it harder for shareholders to track what the company has truly committed to spend.
Why it matters
On one level, it's smart treasury management — AI chips are brutally expensive and depreciate fast. On another, it's a signal about where the AI boom's finances are headed. When the biggest cloud company on earth starts engineering chip purchases to keep them out of the headline numbers — the same week it criticized proposed restrictions on building AI data centers ahead of the midterms — the spending has reached a scale where even Amazon feels the need to dress it up. The question investors are starting to ask isn't whether AI demand is real. It's whether anyone can still see the full bill.
Related coverage: 'Boom Phase': Nvidia-Backed Firmus Prices a $5 Billion IPO — After Its Valuation Tripled in Two Months · 'These Chips Are the Product of American Ingenuity': DOJ Just Busted a $300 Million Nvidia Server Smuggling Pipeline to China
Impact on the USA and the world
For American investors, the SPV cuts both ways. Done right, Amazon funds its AI infrastructure more cheaply and protects its credit rating. But every dollar of AI spending that migrates to the footnotes is a dollar harder to price into risk models — and with the 10-year Treasury above 5 percent, the cost of hidden leverage is climbing. For the world, it's a preview: American hyperscalers are writing the playbook for financing the AI buildout, and every cloud provider on the planet will copy it. The AI revolution isn't just being built in data centers. It's being financed in the footnotes.
What happens next
First, the investor talks: Amazon has been gauging interest "in recent weeks," so a deal could land soon — or fizzle. Second, the copycats: if Amazon pulls it off, expect Microsoft, Alphabet, and Meta to follow within quarters. Third, the politics, with midterm candidates pushing curbs on data-center construction. The crisp takeaway: Amazon isn't selling its AI future — it's just asking someone else to hold the receipt. In a $3 trillion footnote economy, that's the most rational move on the board.