AI BANKRUPTING The US

The video argues that big tech's AI-driven borrowing spree has outgrown the companies' own cash and is now straining the global bond market — pushing up yields and squeezing every other borrower, from governments to homebuyers.

1. The core shift: from cash kings to major borrowers

Big tech's relationship with debt used to be "optional" — borrowing was done for tax efficiency or buybacks, not because the money was needed. AI changed that: unlike software (written once, sold a million times), a data centre must be built, filled with chips, wired into the grid, cooled, powered, and refitted every couple of years.

• Amazon spent $173 billion on capital expenditure over 12 months, while free cash flow was –$11.6 billion — spending more on the cloud than the whole business brings in.

2. The borrowing surge

• October 2025: Mark Zuckerberg told investors Meta's 2026 AI spending would be "notably larger". Meta then sought $30 billion in the bond market — the largest high-grade corporate bond sale on record not funding an acquisition. Orders hit $125 billion, with the longest tranche not repaid for 40 years.
• The five most active hyperscalers issued $121 billion of bonds in 2025 — more than four times their $28 billion/year average across 2020–2024.
• First half of 2026: Amazon, Alphabet, Meta and Oracle issued roughly $195 billion, up 80% on the whole of 2025. Amazon alone borrowed $54 billion in the first few months.
• Alphabet sold a £1 billion, 100-year bond maturing in 2126.
• Alliance Research counts $356 billion raised by AI-related borrowers in US credit markets through end-August, versus $174 billion in all of 2025.
• Almost all of it is fresh money: hyperscalers issued about $150 billion in dollar bonds this year, while only $23 billion of existing debt matures before end-2027.

3. The market's price for all this

• Goldman Sachs: AI-linked issuers pay 115 basis points over benchmarks for new debt, versus 78 bps for the broader investment-grade market.
• Vanguard: sluggish demand for a $25 billion Amazon deal is a sign investors are, in its words, "full".

4. Off-balance-sheet borrowing (SPVs)

When the front door gets expensive, issuers use the "side door": special purpose vehicles (SPVs) — separate companies that borrow the money, own the asset, and keep the debt off the parent's balance sheet.

• Prime example — Meta's Hyperion campus (Louisiana): financed via an SPV called Beignet Investor LLC (named after a Louisiana pastry). Blue Owl-managed funds own 80%; Meta owns 20% and leases the finished data centre back, receiving a $3 billion cash payment when the deal closes.
• The SPV raised $27 billion of A+ rated debt — including $18 billion from PIMCO and $3 billion from BlackRock — the largest private credit deal ever executed.

5. "Neoclouds" and GPU-backed debt

Smaller firms renting out GPU capacity borrow against the chips themselves. CoreWeave illustrates rapidly worsening terms:

That is one borrower's spread more than doubling in five months — a sign of "investor fatigue".

6. Crowding out everyone else

The pool of bond buyers is huge but finite. Pension funds, insurers, bond funds and foreign central banks each have a limited capacity for long-dated debt — so every dollar lent to Meta for 40 years is a dollar not lent to someone else.

• The five major hyperscalers plus Nvidia have issued roughly $320 billion of debt this year — equal to 68% of new long-term Treasury borrowing.
• A quote attributed in the transcript to "Columbia's [name garbled]": "Any other entity that needs capital, from the US government financing its $2 trillion per year deficit to an individual seeking a home mortgage or a car loan, is in competition for funds with these hyperscalers."
• 30-year Treasury yield: 4.7% → 5.5% in a year.
• 23 September: the US had to offer just over 5% to sell $70 billion of 5-year debt — the highest auction yield in 20 years — and demand still came in weaker than expected.
• A week earlier, 20-year Treasuries sold at 5.42% — the highest auction yield since 1986.
• From September, the Treasury doubled long-end bond buybacks from $2 billion to $4 billion per operation to steady that part of the curve.
• Treasury Secretary Scott Bessent called AI corporate issuance "almost yield agnostic" — i.e. these borrowers will pay whatever it takes, and everyone else must match them.

Who gets squeezed:

• Apollo: investment-grade corporate yields near 6%; leveraged loans near 10%.
• Morgan Stanley: $235 billion of software loans — half rated B-minus or lower — exposed as lenders rotate towards AI infrastructure.
• Emerging-market issuers face wider spreads; municipalities funding schools and water systems pay more; mortgage rates follow the 10-year Treasury.

7. The hidden balance sheet

• "Nic Asia" (transcript name; likely garbled) estimates the five major US tech giants carry around $1.65 trillion of off-balance-sheet obligations, against roughly $1.35 trillion officially reported — more debt off the books than on them.
• Moody's: over $820 billion of those obligations relate to data centres still being built.
• The risk remains because these structures rely on long-term leases and residual value guarantees: if the tenant walks away or the asset sells for less than promised, the tech giant covers the shortfall.

8. Lenders are hedging — credit default swaps

Credit default swaps (CDS) are insurance on a bond: you pay a yearly fee, and if the borrower defaults, you get paid.

• September: Oracle's 5-year CDS spread jumped 16% in a single week to a record high — about four times the investment-grade average.
• Apollo's Torsten Slok tracks a basket of Amazon, Google, Microsoft and Oracle CDS that has climbed above 100 bps — the highest in an 8-year series — while the big US banks sit at around 40 bps. It now costs more to insure against America's tech champions than against its largest banks.
• Oracle's 2056 bonds crossed an 8% yield for the first time. A year ago, Oracle's own $18 billion deal drew orders of up to $90 billion.
• Analysts warn a downgrade to junk would force index funds to dump roughly $120 billion of bonds.
• After Oracle issued a force majeure notice on the New Mexico project, $18 billion of loans tied to the development traded at around 90 cents on the dollar.

9. Circular financing and stretched collateral

• The AI capex boom features circular loops: chip makers take stakes in AI labs, and the labs spend that money on chips and cloud infrastructure (Nvidia–OpenAI, Nvidia–CoreWeave, AMD–OpenAI). Debt enters the loop via private credit and bank syndicates funding the infrastructure those startups use to pay the tech giants.
• Nvidia sold $25 billion of bonds in June — its first bond sale in five years.
• Depreciation risk: hyperscalers have stretched GPU depreciation schedules to 5–6 years, while critics like Michael Burry argue the real useful life is closer to 2–3 years. One estimate puts secondary GPU rental rates down 70% from their peak.
• Mismatch: CoreWeave's investment-grade loan matures in 2032, but the customer contracts supporting such loans typically last only about 3 years. Hyperion's debt runs 24 years, on a campus whose most valuable contents will likely be obsolete several times over — "loans that outlive their collateral".

10. The historical parallel: the dot-com fibre bust

• After the Telecommunications Act of 1996, US telecoms firms poured over $500 billion into fibre optic cable, justified by the claim that internet traffic was doubling every 100 days — it wasn't.
• Equipment makers like Lucent and Nortel even lent money to startup carriers so they could buy Lucent and Nortel's own gear. The Wall Street Journal later called the industry "a gigantic poker game".
• Global Crossing filed for bankruptcy in January 2002 carrying $12.4 billion of debt; WorldCom followed in July with $41 billion; more than 60 telecom bankruptcies came in all.
• Corning, the fibre maker, went from nearly $100 a share to around a dollar. Only 5% of the laid fibre was switched on in the years after the bust — yet that cable eventually became the backbone of the modern internet. The fibre survived; the bondholders didn't.
• Swap fibre for GPUs and Lucent for Nvidia and the setup looks similar — though estimates suggest today's buildout is growing five to six times faster as a share of the economy, so the stakes are much larger.

11. Why it matters to ordinary people

• Mortgage rates follow the 10-year Treasury.
• Pensions almost certainly own corporate bonds.
• Hold a bond index fund and you likely own some AI debt already.
• If Oracle is cut to junk, your fund could be one of the forced sellers.
• Either directly or indirectly, "you're already paying for the AI boom".

Closing argument

The giants themselves will very likely be fine — that was never the issue. The problem begins when companies this huge borrow at this scale, all at the same time, for years on end. The AI boom may be built by a handful of tech giants, but the financial stress it is creating is spreading through the entire economy.

Nic closes with an open question: is big tech borrowing sensibly to build the infrastructure of the future, or is it passing the cost of the bet on to every other borrower in the market?