Is The AI Bubble (finally) Popping?

Summary

The video argues that the AI investment boom may be becoming more fragile—not just because valuations could fall, but because AI infrastructure is increasingly financed with debt. A downturn in demand could therefore affect lenders and company cash flows as well as stock prices. The presenter compares these risks with the dot-com era and with recent AI-related market declines in China and South Korea, while stressing that a sudden crash is not inevitable.

Main financial themes

• AI investment is shifting toward debt financing. The presenter says AI companies borrowed $489 billion in the first half of the year, mostly through bonds carrying interest rates above 10%. The video compares the inflation-adjusted amount with subprime lending originated in 2006.
• Infrastructure can look like an asset without being a safe investment. Buying data centers and GPUs shifts company resources from cash into property, plant, and equipment. Borrowing to do so adds liabilities alongside the assets. However, computer hardware can lose value quickly, and loans secured by it may become riskier if resale prices fall.
• Accounting treatment affects reported earnings. The video says some tech companies depreciate chips over six years, compared with a cited typical replacement horizon of one to three years for cutting-edge components. Longer depreciation schedules spread expenses over more years, potentially making near-term earnings look stronger. The presenter also notes that unusually tight hardware supply has so far supported resale and rental values.
• The financing chain may be circular. AI companies rent compute from cloud providers; some cloud providers borrow to build capacity; and GPUs may serve as loan collateral. Tech companies also rent infrastructure from one another. This can seem secure while demand and hardware values hold up, but the different links may depend on the same underlying AI demand.
• Large tech companies face pressure on both spending and core businesses. The presenter says Google/Alphabet has reported a second negative free-cash-flow quarter, something not seen in its 22 years as a public company. Search-ad growth reportedly slowed from 19% to 12% over two quarters, amid competition from AI-generated answers and higher infrastructure costs. The video also cites the cost of hosting a growing volume of AI-generated content on YouTube.
• Equity markets have started to wobble, but the declines cited are uneven. The video refers to some cloud-provider shares falling around 50% and roughly $5 trillion being erased from major AI-related companies collectively, while describing the Nasdaq 100 correction as about 10%. It notes that after strong gains over the prior three years, some stock prices had only returned to levels seen earlier in the year.
• Rising borrowing costs could amplify a downturn. The presenter says CoreWeave’s first major loan in 2023 carried a rate just under 15%, about 10 percentage points above the risk-free rate. By March, its borrowing spread reportedly narrowed to 2.25 percentage points over the risk-free rate, or about 5.9% all-in. A subsequent loan in August reportedly required a 5.5-point spread. Unsecured borrowing for some firms was said to exceed 15%.
• Credit protection is not a cure-all. The video describes credit default swaps (CDS) as insurance against bond default, and synthetic CDS as contracts that can reproduce bond-like exposure without directly lending to the borrower. It reports that premiums on Microsoft credit protection more than doubled, Alphabet’s more than tripled, and Oracle’s reached almost 300 basis points. For example, insuring $10 million of Oracle exposure would cost about $300,000 a year in premiums, on top of the loan’s other interest costs; the presenter estimates borrowing costs near 9%.
• Stock buybacks may give way to fundraising. The video says S&P 500 companies bought back a record $1 trillion of stock in the 12 months through September of the previous year. It contrasts this with companies increasingly selling equity and borrowing to support spending. Google is said to have announced an $85 billion equity raise and sold 19-year euro bonds at 5.25%.
• China and South Korea have already seen sharp AI-linked market falls. China’s STAR 50 AI-stock index reportedly fell 40% in four months, wiping out about $480 billion in shareholder value. South Korea’s market fell 40% from its June peak to its October low. The video says Samsung and SK Hynix together represented more than half of the Korean stock market, making it particularly exposed to chip demand and pricing.
• The presenter sees important differences between those markets and the U.S. Korean and Chinese listed firms are characterized as more concentrated in chip and component suppliers, whose sales are more directly exposed to pricing competition and demand limits. The video cites a Chinese memory-chip design about one-third smaller than the current generation, after which SK Hynix fell 10% in one day. It also says Korean chip exports more than tripled year over year in October, while unemployment remained broadly stable and retail sales improved in September.
• Hardware conditions have helped sustain the boom—for now. The video says rental rates for Nvidia H100 GPUs fell from about $7 per hour to $1.50, then rose again, with median hourly rates up around 13% year over year. It also cites an 80% profit margin in a record quarter for SK Hynix, attributing current strength partly to constrained memory-chip supply.

Risk mechanism described

The presenter outlines a potential chain of events:

1. AI usage or customer growth weakens.
2. Companies reduce or fail to renew compute leases.
3. Cloud providers and other owners sell excess GPUs and data-center equipment.
4. More hardware on the market pushes resale and collateral values down.
5. Loans secured by that equipment become less secure, while higher interest costs further strain borrowers.
6. Tighter credit forces additional spending cuts or asset sales.

The presenter’s central caution is that a modest decline in share prices might be manageable, but debt makes a smooth “deflation” of the AI boom harder: weak growth could trigger a more damaging feedback loop. The video suggests watching developments in China and South Korea and says policymakers should not automatically reach for a bailout merely because markets become unsettled. It does not predict that a crash is certain.

Assets, companies, sectors, and instruments mentioned

• Companies: Alphabet/Google, Meta, Amazon, Microsoft, Oracle, Nvidia, Intel, Anthropic, OpenAI, SpaceX, CoreWeave, Samsung, and SK Hynix.
• Indices: China’s STAR 50, Nasdaq 100, and S&P 500.
• Assets and sectors: AI and cloud-computing companies, data centers, GPUs and other chips, memory, search advertising, and technology stocks.
• Instruments: Corporate bonds, euro bonds, equity issuance, secured and unsecured loans, credit default swaps, and synthetic CDS.
• Hardware examples: Nvidia H100 GPUs and the RTX 5090.

Disclosures

The video describes itself as educational and not investment or financial advice. It also includes a paid sponsorship for Incogni.

Presenter and sources

• Presenter/source: How Money Works (unnamed narrator).
• Sources listed with the video: The Register, The Next Web, Forbes (Fiona Riley), 24/7 Wall St., and Yahoo Finance.