Video summary

The AI Bubble Is About to POP… Here's the Warning Sign Nobody Sees

Main summary

Key takeaways

Finance

Summary

Ross Givens argues that AI-related stocks and infrastructure may be in a technology-bubble cycle. He says three of the four warning stages are already visible, but the key trigger—falling prices for renting AI computing capacity—has not yet appeared. He cautions that a real technological breakthrough does not prevent investors from suffering severe losses if they buy at inflated prices. He also argues that a broad selloff could create buying opportunities in companies and infrastructure that survive.

How AI economics work

  • AI models require costly training and ongoing computation on data-center chips, mostly Nvidia chips, unlike conventional software, which can be reproduced cheaply once written.
  • Citing Epoch.ai, Givens says the cost of obtaining the same level of AI performance has fallen roughly 13-fold per year since 2023. He argues that this benefits customers but could pressure AI providers’ economics if they must sell much more to offset falling prices.
  • He compares a free downloadable Xiaomi model scoring 46 on a popular benchmark with Anthropic’s Claude Opus at 58—about 80% of its score. He says their inference costs are $0.87 versus $20 per million words, respectively.

Four-stage technology-bubble framework

Givens’s historical framework is:

  1. Debt-fueled overbuilding: Companies build infrastructure faster than demand warrants, often using borrowed money.
  2. The price of the product or capacity falls: Revenue can decline while debt payments remain fixed.
  3. Credit tightens: Lenders charge more or stop lending as collateral values and repayment prospects weaken.
  4. Highly leveraged companies fail first: Their suppliers and even stronger companies’ stocks can also be hit in a broader selloff.

He points to railroad, radio, Nifty 50, dot-com, fiber-optic, and power-plant booms as historical parallels. As examples, he says British railroad investment rose from £13 billion in 1845 to £44 billion in 1847, with rail stocks later falling 64%. In the fiber-optic boom, he says the annual cost of renting a New York–Los Angeles express lane fell from $1.8 million in 2000 to $597,000, then $248,000—an 86% drop over two years. He also notes that the Nasdaq fell 78% after the 2000 tech-bubble peak, and investors who bought at the top waited about 15 years to break even.

AI spending, debt, and credit risks

  • Amazon, Microsoft, Alphabet, Meta, and Oracle reportedly spent $181.5 billion on buildings and equipment in the last quarter, up 87% year over year. Givens annualizes that pace to about $726 billion, nearly five times the level at the beginning of 2022.
  • Their cloud businesses generated $114 billion in revenue, up 44%, but spending growth had outpaced cloud-revenue growth for 10 consecutive quarters. He estimates they invest about $1.59 in new capacity for each dollar of cloud revenue.
  • He says Microsoft and Meta spent 96% of the cash their core businesses earned on AI development, while Amazon and Alphabet spent more than those core businesses earned.
  • Morgan Stanley estimates cited in the video put global data-center spending at about $2.9 trillion by 2028, with a $1.5 trillion gap between planned construction and what major tech companies can fund themselves. Givens says Morgan Stanley expects roughly $800 billion of that gap to come from private credit.
  • He describes weakening private-credit fundraising: retail private-credit funds reportedly raised $2 billion in Q2, versus about $11 billion a year earlier, while investors withdrew $5.9 billion. Some funds capped withdrawals at 5%. He says outflows exceeded inflows in both the first and second quarters, though redemption queues had shortened at 13 of 15 tracked funds.
  • A $38 billion loan for two data-center campuses being built by Vantage for Oracle and OpenAI’s Stargate project reportedly took months to sell down; some lenders offered their portions at a discount. A separate $5 billion Volta data-center loan was reportedly seeking higher interest and looser terms to attract buyers.
  • Other credit indicators he cites: Oracle’s cost of debt-default insurance reached a record; CoreWeave debt that initially required a 9.625% rate was said to demand 14.6% in the market; and $27.3 billion of bonds tied to Meta’s Hyperion data center were trading below par.
  • Givens says Amazon, Alphabet, Meta, and Oracle sold about $194 billion of bonds through July and notes that Alphabet issued 100-year bonds. The subtitle’s comparison with about $108 billion for all of 2025 is unclear, so the figures may not be directly comparable.
  • He also says the 10-year U.S. Treasury yield reached 5%, its highest level since 2007, and that the Fed raised rates in September.

Where he says the market stands

Givens sees substantial evidence of debt-funded investment and tightening credit, alongside early weakness in some AI-related stocks. From October 19 highs, he cites declines of 90% for NuScale Power, 86% for Oak Street, 74% for Supermicro, 68% for CoreWeave, and 62% for Oracle, while the S&P 500 was down less than 2% from its highs. He characterizes this as an early, concentrated selloff rather than a broad market crash.

The second stage, in his view, has not been confirmed: AI-computing rental prices were rising, not falling. He reports a median price of $3.81 per hour to rent an Nvidia H100, versus about $3 a year earlier—up 28%—and says rental prices across tracked chips had risen about 21% that year.

What he says to monitor

  • GPU rental rates: His main warning signal is a sustained decline over a couple of months. He argues that lower rental prices could reduce the value of chips used as loan collateral, prompt forced chip sales, and push rental prices down further.
  • Big Tech spending plans: He says Microsoft and Alphabet were due to report on October 28, with Amazon on October 29. He recommends watching whether those companies cut planned spending; he says such a cut would signal a crack in the investment boom.
  • Personal preparation: He advises investors to decide in advance what they might buy after a selloff and how much they are willing to pay. He offers no specific stock recommendations in this video, but says a later video would discuss five stocks he may consider.

Assets, companies, and instruments mentioned

  • Companies and AI providers: Nvidia, Oracle, OpenAI, Anthropic/Claude, Xiaomi, Amazon, Microsoft, Alphabet/Google, Meta, Supermicro, CoreWeave, NuScale Power, Oak Street, Vantage, JP Morgan, and MUFG.
  • Markets and infrastructure: AI stocks, data centers, cloud computing, GPUs and Nvidia H100 chips, railroad stocks, fiber-optic networks, power plants, software companies, and technology stocks.
  • Instruments and credit: Corporate bonds, data-center loans, private-credit funds, debt-default insurance/CDS, U.S. 10-year Treasuries, the Nasdaq, and the S&P 500.
  • Ticker symbols: No ticker symbols are spoken in the subtitles.

Disclosures and promotion

No explicit “not financial advice” disclaimer appears in the subtitles. Givens promotes his Black Ops trading service for $5 for a year, advertising live mentoring, a newsletter, and TradingView tools. This is a paid-service promotion, not a specific investment recommendation.

Presenter and sources

Presenter: Ross Givens (Ross Givens channel).

Sources named or cited in the subtitles: Bloomberg, Epoch.ai, Morgan Stanley, and “Stenger” (the attribution is unclear in the subtitles). An unnamed tracker is cited for GPU rental rates and private-credit fund data.

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