Revenue disappoints
If AI revenue stops growing fast enough to cover the capex, investors stop paying for future growth. This is the gap Bain sizes at up to $800 billion by 2030.
Is AI a bubble? On the evidence as of November 13, 2026, our answer is bubble territory: the AI Bubble Meter scores today 67 out of 100, against 87 for the dot-com bubble at its March 2000 peak. Score the 10 warning signs yourself, from AI capex vs revenue to circular financing, and see where you land.
Each slider starts at our sourced baseline (November 13, 2026). Read the evidence, move any slider you disagree with, and compare your total with the dot-com peak of March 2000. Nothing leaves your browser.
Your AI Bubble Score
67Bubble territory
Dot-com peak: 87 · Our baseline: 67
Spending, prices and financing have run well ahead of proven profits. A sharp correction would not be a surprise.
Is spending on AI data centers running far ahead of what AI earns?
Now: Microsoft, Alphabet, Amazon and Meta guide to roughly $720-740 billion of 2026 capital spending, and Meta's free cash flow turned negative in Q3 2026. On the revenue side, OpenAI's annualized revenue was near $50 billion in October 2026, and Bain estimates AI needs $2 trillion a year of revenue by 2030. Unlike 2000, cloud capacity is sold out: Azure grew 48%.CNBC, Oct 29, 2026 (AI capex vs AI revenue, opens in a new tab)Meta Q3 2026 (AI capex vs AI revenue, opens in a new tab)Bain via Bloomberg (AI capex vs AI revenue, opens in a new tab)
2000: Telecom companies invested more than $500 billion in the five years after 1996, mostly in fiber. By 2002 only about 2-5% of US fiber capacity was lit.Brookings (AI capex vs AI revenue, dot-com, opens in a new tab)FDIC, 2002 (AI capex vs AI revenue, dot-com, opens in a new tab)
Baseline 7/10 · Dot-com 10/10
Are stock prices far ahead of earnings?
Now: The S&P 500 forward P/E was 17.7 on November 5, 2026, its lowest in three years, because earnings estimates rose much faster than prices. The long-run Shiller CAPE is still near 41, close to its 1999 record.FactSet, Nov 2026 (Stock valuations, opens in a new tab)Shiller CAPE (Stock valuations, opens in a new tab)
2000: The Shiller CAPE peaked at 44.2 in December 1999. Leaders traded at extreme multiples: Cisco at about 200 times earnings.Shiller CAPE (Stock valuations, dot-com, opens in a new tab)Motley Fool (Stock valuations, dot-com, opens in a new tab)
Baseline 5/10 · Dot-com 10/10
Does the market depend on a handful of AI stocks?
Now: Three stocks made up about 21% of the S&P 500 in October 2026, the most ever, and the ten largest are close to 40% of the index. Nvidia alone was worth about $4 trillion on November 12, 2026.24/7 Wall St, Oct 7, 2026 (Market concentration, opens in a new tab)Nvidia market cap (Market concentration, opens in a new tab)
2000: At the March 2000 peak the ten largest stocks were roughly 25-29% of the S&P 500, less concentrated than today.Informa Connect (Market concentration, dot-com, opens in a new tab)
Baseline 9/10 · Dot-com 6/10
Are suppliers funding their own customers?
Now: Nvidia invests in OpenAI (a $30 billion stake, after a planned $100 billion deal was dropped in March 2026) and backstops CoreWeave capacity; AMD gave OpenAI warrants tied to GPU purchases; Oracle holds a $300 billion OpenAI contract. The IMF's October 2026 stability report names circular financing as a risk amplifier.IMF GFSR, Oct 2026 (Circular and vendor financing, opens in a new tab)Capacity, CoreWeave (Circular and vendor financing, opens in a new tab)
2000: Lucent, Nortel and Cisco lent billions to telecom start-ups that bought their gear. Lucent alone had about $6.7 billion of customer-financing commitments in September 2000; many borrowers later failed.Lucent 10-K/A, 2000 (Circular and vendor financing, dot-com, opens in a new tab)
Baseline 8/10 · Dot-com 8/10
Is the build-out increasingly paid for with borrowed money?
Now: Hyperscalers and related firms sold about $225 billion of bonds in the first half of 2026, and Moody's counts about $1.2 trillion of off-balance-sheet obligations. Oracle's credit default swaps reached high-yield levels in October 2026, and Meta's Hyperion data center sits in a $27 billion off-balance-sheet joint venture.Fortune, Jul 31, 2026 (Data center debt, opens in a new tab)Oracle CDS (Data center debt, opens in a new tab)
2000: Telecom carriers borrowed heavily to lay fiber, and the bust ended in record defaults: Global Crossing filed with $12.4 billion of debt in January 2002 and WorldCom with about $41 billion months later.CBS News (Data center debt, dot-com, opens in a new tab)
Baseline 7/10 · Dot-com 9/10
Are private AI companies priced on hope rather than revenue?
Now: OpenAI raised money at a $1.25 trillion valuation in November 2026, and Anthropic was valued at $965 billion in May 2026 on a run-rate above $47 billion. That is roughly 15-25 times revenue. AI took about two thirds of global venture funding in Q3 2026.The Star/Bloomberg, Nov 7, 2026 (AI startup valuations, opens in a new tab)AP via WTOP, May 2026 (AI startup valuations, opens in a new tab)
2000: US venture funding hit about $100 billion in 2000 and companies with little or no revenue went public; the average 1999 IPO jumped 69% on its first day.NYU study (AI startup valuations, dot-com, opens in a new tab)
Baseline 7/10 · Dot-com 10/10
Are ordinary investors piling in with borrowed money?
Now: FINRA margin debt hit a record $1.5 trillion in June 2026 and stayed near it in September. Google searches for "AI bubble" reached a five-year high in late October 2026, and more than 65% of S&P 500 companies mentioned AI on Q2 2026 earnings calls.FINRA margin data (Retail mania and leverage, opens in a new tab)Google Trends (Retail mania and leverage, opens in a new tab)
2000: Margin debt peaked in March 2000, day trading was a national craze, and companies added ".com" to their names to lift their share prices.Advisor Perspectives (Retail mania and leverage, dot-com, opens in a new tab)
Baseline 8/10 · Dot-com 10/10
Are customers getting real value, or just experimenting?
Now: About 20% of US businesses used AI in May 2026 (37% of large firms), a share that has barely moved since late 2025, and an MIT study found 95% of generative AI pilots showed no measurable return. But ChatGPT had 900 million weekly users in February 2026 and cloud AI revenue is growing fast.US Census, May 2026 (Adoption and payoff, opens in a new tab)MIT report (Adoption and payoff, opens in a new tab)
2000: Internet use was exploding, but most dot-coms had no path to profit; Pets.com went from IPO to liquidation in under a year.
Baseline 4/10 · Dot-com 6/10
Will today's chips lose value before they pay for themselves?
Now: Big tech depreciates servers over 5-6 years, while Nvidia ships a new generation every year. Michael Burry estimates this understates depreciation by about $176 billion over 2026-28. H100 rental prices are far below their 2023-24 peak but rose again during 2026 as demand stayed high.Outlook Business on Burry (GPU depreciation, opens in a new tab)
2000: A fiber glut left about 97% of installed fiber unused in 2001, and bandwidth prices collapsed, wiping out the value of the network that had just been built.Deseret News, 2001 (GPU depreciation, dot-com, opens in a new tab)
Baseline 5/10 · Dot-com 10/10
Are the people closest to AI saying it is overheated?
Now: Sam Altman said investors are "overexcited" (Aug 2025), Jeff Bezos called it an "industrial bubble" (Oct 2025) and Sundar Pichai saw "elements of irrationality" (Nov 2025). The IMF's October 2026 report lists an AI correction among three salient risks, and Michael Burry bet against Nvidia and Palantir.Axios on Bezos (Insider and smart-money warnings, opens in a new tab)Pichai via Computing (Insider and smart-money warnings, opens in a new tab)IMF GFSR (Insider and smart-money warnings, opens in a new tab)
2000: Alan Greenspan warned of "irrational exuberance" in December 1996, more than three years before the peak, and tech insiders sold heavily through 2000.Federal Reserve, 1996 (Insider and smart-money warnings, dot-com, opens in a new tab)
Baseline 7/10 · Dot-com 8/10
The score is the average of the 10 signals, times ten. Baseline and dot-com scores are AI Wins' judgment of the cited numbers, checked November 13, 2026. Not financial advice.
Bubbles rarely end on valuation alone. These are the triggers analysts watch most.
If AI revenue stops growing fast enough to cover the capex, investors stop paying for future growth. This is the gap Bain sizes at up to $800 billion by 2030.
If one hyperscaler trims its data center budget, Nvidia, chip suppliers, utilities and neoclouds all lose their biggest customer at once.
Rising default insurance on debt-funded builders such as Oracle and CoreWeave, or a failed bond deal, could cut off the money that pays for the build-out.
A model that matches the best at a fraction of the compute would cut demand for GPUs and data centers, as DeepSeek briefly did in January 2025.
Higher interest rates make long-dated AI bets worth less today, and new rules on power, chips or AI use could slow spending.
If a lab cannot pay a supplier that also invested in it, losses spread through the web of cross-investments.
2025-10-03
Jeff Bezos calls AI an "industrial bubble"
He added that industrial bubbles leave behind useful infrastructure even when investors lose money.
2026-03-31
OpenAI raises $122 billion at an $852 billion valuation
The largest private funding round on record at the time.
2026-05-28
Anthropic is valued at $965 billion
Demand for Claude pushed its run-rate above $47 billion; it filed confidentially for an IPO days later.
2026-07-08
Nvidia sheds about $1 trillion in under two months
Bloomberg reported the slide from its May 14 high sent its valuation to pre-AI-boom levels.
2026-10-13
IMF flags an AI correction as a salient risk
Its Global Financial Stability Report also names circular financing deals as a risk amplifier.
2026-10-29
Big tech lifts 2026 capex toward $740 billion
Earnings week: Meta reports negative free cash flow and "AI bubble" searches hit a five-year high.
2026-11-05
The Nasdaq Composite falls into a correction
More than 10% below its October 6 record, as investors question returns on AI spending.
Each warning sign shows today's numbers and the 1999-2000 dot-com numbers, with dated source links.
Every slider starts at our baseline. Move any you disagree with, from 0 (healthy) to 10 (extreme).
Your 0-100 AI Bubble Score sits on the same meter as the dot-com peak, and each sign shows you vs 2000.
Copy the link or a text summary. Your scores are saved in the link.
Parts of it look like one. On the evidence as of November 13, 2026, the AI Bubble Meter scores 67 out of 100 (bubble territory). Spending, market concentration, circular financing and margin debt look stretched, but unlike in 2000 the leading AI companies earn real and fast-growing revenue, cloud capacity is sold out, and the S&P 500 forward P/E (17.7 in November 2026) is far below dot-com levels.
Today's market is more concentrated than in 2000 and its debt and circular deals rhyme with the telecom bust, but valuations are much lower: the S&P 500 traded at about 18 times forward earnings in November 2026, while Cisco traded near 200 times in 2000. Scored on the same 10 signs, the dot-com peak gets 87 out of 100 against 67 today. After its March 10, 2000 peak the Nasdaq fell 78% by October 2002.
Circular financing is when a supplier invests in or lends to a customer that then spends the money on the supplier's products. In AI, Nvidia has invested in OpenAI and backstops CoreWeave, and AMD gave OpenAI share warrants tied to GPU purchases. It inflates reported demand if the customer cannot pay its own way. Lucent and Nortel did the same with telecom start-ups before 2001.
The usual triggers are AI revenue growing too slowly to cover data center spending, a hyperscaler cutting capex, credit stress at debt-funded builders such as Oracle or CoreWeave, a much cheaper model that cuts demand for chips, or higher interest rates. A bubble can also deflate slowly rather than pop, as earnings catch up with prices.
Each of the 10 warning signs gets a score from 0 (healthy) to 10 (extreme). The AI Bubble Score is their average times ten, from 0 to 100: under 25 is no bubble, 25-44 frothy but grounded, 45-64 a bubble warning, 65-79 bubble territory and 80 or more dot-com mania. Our baseline today is 67; the same signs scored for March 2000 give 87. It is a way to organize the evidence, not a forecast or financial advice.
Evidence was checked against its sources on November 13, 2026. Scores are AI Wins' judgment of public numbers, not a prediction. This tool is for education and is not financial or investment advice; AI Wins is not affiliated with any company or person named.
See when experts expect AGI, the bet behind the spending.
Open toolCheck whether an AI project pays for itself.
Open toolSee how exposed your job is to AI automation.
Open toolPlain-English definitions of the terms in the news.
Open toolThe bubble debate moves with every earnings call. Get the AI news that matters in the free daily AI Wins digest.
Get the free digestOnly positive AI stories, delivered daily or weekly. Free, no spam, unsubscribe anytime.