Video summary

How the AI Bubble Could Burst (4 Warning Signs Are Here)

Main summary

Key takeaways

Finance

Summary

The video argues that an AI-related downturn could hurt the S&P 500 because a large share of the index is concentrated in AI-linked companies, while much of their spending depends on a small number of AI customers. It outlines two possible ways the boom could break: AI-lab customers reduce spending or fail to meet their commitments, and debt or other long-term financing obligations become harder to support.

The presenter says all four historical crash warning signs are present to some degree, though a decisive pullback by major cloud spenders has not yet occurred.

How the AI spending cycle works—and where it may be vulnerable

  • Nvidia supplies chips to major cloud and tech companies, including Microsoft, Amazon, Google, Meta, and Oracle. Together, these companies are on track to spend more than $690 billion on data centres and chips this year, according to the video.
  • Nvidia’s latest filing reportedly showed that 70% of its unpaid bills were owed by five customers; the video says their identities are not known.
  • Citing investor Steve Eisman, the presenter says about 70% of AI revenue at Microsoft, Amazon, Google, and Oracle comes from just OpenAI and Anthropic.
  • The video describes a cycle of investment and spending: Nvidia and Microsoft invest in AI companies; those companies pay cloud providers for computing; cloud providers then buy more Nvidia chips. Nvidia-backed CoreWeave is presented as another example of money circulating back toward Nvidia.
  • The presenter identifies customer concentration and uncertain demand as risks. Customers may reduce spending, build their own tools, or switch to cheaper models. The video gives the following examples:
    • Uber used its 2026 AI-coding-tools budget in four months, then capped spending at $1,500 per employee per month per tool.
    • Microsoft is moving most engineers from Anthropic’s Claude Code to its own tool.
    • AT&T found that using cheaper open-source models for some coding work could reduce costs by up to 56%, with quality falling about 2%.
    • Ramp data cited in the video indicates that 1% of companies account for about 80% of spending on OpenAI and Anthropic.

Two ways the boom could break

  1. AI-lab spending falters. OpenAI may need continuing fundraising to meet its plans. The video estimates about $36 billion in revenue this year, a target of $350 billion by 2030, and an expected cumulative cash burn of roughly $278 billion. It says OpenAI raised $122 billion in March and could exhaust that funding around 2028 without additional capital. Anthropic is described as profitable, but it has also made large commitments to cloud providers and depends on customers continuing to pay. If either lab cuts spending or cannot pay, the effects could spread to indebted providers such as Oracle and CoreWeave, then to cloud companies and chip suppliers.
  2. Debt and commitments strain companies. The video says the share of cloud-giant spending funded with debt rose from 9% in 2024 to 32%. It also cites approximately $3 trillion in data-centre commitments among nine large tech companies—about five times their data-centre and chip spending over the prior year. A Meta Louisiana data centre was financed with $27 billion in bonds through a separate company set up with Blue Owl; Meta reportedly owns 20% and Blue Owl 80%. The presenter also warns that longer depreciation assumptions can make current profits look higher: Meta’s server-life assumption rose from 5 to 5.5 years, reportedly reducing 2025 costs by $2.9 billion. Michael Burry is cited as estimating such accounting effects could add $176 billion to cloud giants’ reported profits over 2026–2028.

The video notes that large cloud companies are still funding much of their investment with cash and may be better able to withstand a slowdown than highly leveraged providers such as CoreWeave.

Is it a bubble?

  • The presenter says JPMorgan found that AI-related companies accounted for about two out of every three dollars of S&P 500 profits and spending since ChatGPT launched.
  • The video contrasts current valuations with the dot-com period: software stocks reportedly traded at about three times the rest of the market’s price-to-earnings level in 2000, versus about a 20% premium today.
  • It characterizes the concern as potentially an “earnings bubble” rather than a valuation bubble: current prices may be supported by real profits, but those profits depend on continued AI-related spending.
  • In a Bank of America October survey cited by the presenter, almost four in ten fund managers said cloud-giant spending was the most likely trigger for a wave of debt problems.

Four warning signs cited

  1. Rapid price increases: A Harvard study of US industries from 1926–2014 found higher crash odds when an industry’s stocks doubled in two years. Over the prior two years, the Nasdaq 100 was up about 45%, while AI chip stocks were up about 117%.
  2. Rising interest rates: The presenter says the Federal Reserve raised rates in September and signalled another hike before year-end, making new borrowing more costly for companies such as Oracle and CoreWeave.
  3. Nervous lenders: Risky-company borrowing spreads were said to be about 3.3 percentage points, below the cited 20-year average. But CoreWeave reportedly faced borrowing costs around 14.8%, and the cost of insuring Oracle debt against default reached its highest level since 2008 in July.
  4. Spending far ahead of revenue: Goldman Sachs estimates cloud giants need about $300 billion a year in AI revenue to break even on their spending, while AI had added about $70 billion a year to cloud revenue so far. For comparison, the video says telecom-company revenue grew about 19% from 1998–2000 while spending rose about 74%.

The presenter’s assessment is that all four signs are appearing in some form, but the major cloud spenders have not yet made the broad spending pullback that ended the telecom boom.

Portfolio guidance offered

The presenter gives three risk-management suggestions while emphasizing uncertainty about whether or when a downturn will occur:

  • Avoid investing with borrowed money. A sharp decline could trigger a forced sale and lock in losses.
  • Do not panic-sell based on alarming headlines. The presenter cautions that trying to time the market may leave investors out during a recovery.
  • Invest only money that will not be needed for at least five years. Money earmarked for near-term goals—such as a house or education—may not be suitable for market risk.

Assets, companies, and instruments mentioned

  • Indices and markets: S&P 500, Nasdaq 100, South Korean stock market.
  • Stocks and companies: Nvidia (NVDA, also mentioned in a promotional offer), Microsoft, Amazon, Google, Meta, Oracle, CoreWeave, OpenAI, Anthropic, Uber, AT&T.
  • Sectors and themes: AI chips, cloud computing, data centres, software, telecom.
  • Instruments and financial concepts: Stocks, S&P 500 funds, bonds, debt, borrowing costs, default insurance, depreciation, cloud contracts and commitments, US stocks and options (in the sponsor promotion).

Disclosures and promotions

The video includes a Moomoo-sponsored promotion for a paper-trading competition, trading incentives, and account-opening rewards. The promotion is subject to terms and conditions.

The accompanying disclosure says Kelvin Learns Investing is not licensed to provide financial advisory services in Singapore. It also states that the video is for informational purposes, does not constitute financial advice, and reflects the channel’s independent opinion.

Presenter and sources mentioned

Presenter: Kelvin, of Kelvin Learns Investing.

Sources or people cited in the video: Steve Eisman; Nvidia’s latest filing; Ramp; The Wall Street Journal; Michael Burry; JPMorgan; Goldman Sachs; Bank of America’s October fund-manager survey; Harvard economists’ study of US industries; Fitch.

Rate this summary

Your feedback will help improve summaries.

Improve this summary

Reprocess with a stronger model when the summary feels incomplete or inaccurate.

Pro

Translate summary in another language

Pro

Ask questions to this video

Chat for follow-up questions, clarifications, and source-backed answers.

Coming soon

Share this summary

Original video