Is AI a bubble? The numbers both camps cite
Nobody has settled it. Here are the figures each side points to, with who reported them and when, as of 22 November 2026. This page is not investment advice, and it makes no forecast.

| Figure | Number | Reported by | Date | Who cites it |
|---|---|---|---|---|
| Yearly AI infrastructure spending | About $800 billion | Gary Gensler, via CNN | 16 Nov 2026 | Bubble camp |
| Big Tech spending versus three years earlier | 5 times higher | Gary Gensler, via CNN | 16 Nov 2026 | Bubble camp |
| Projected AI infrastructure spending to 2030 | $7 trillion, against Gartner revenue of no more than $3.6 trillion | Axios | 16 Nov 2026 | Bubble camp |
| Anthropic revenue, nine months to 30 Sep | $33.24 billion, up about 1,200% | CNBC | 18 Nov 2026 | Other camp |
| Anthropic run rate revenue, end of October | About $81 billion, up $10 billion in a month | CNBC | 18 Nov 2026 | Other camp |
| Anthropic IPO | $83 a share, about $1.5 trillion, $81.3 billion raised | CNBC | 18 Nov 2026 | Both |
| Anthropic net loss, nine months to 30 Sep | $104 billion, mostly from changes in the value of convertible notes | CNBC | 18 Nov 2026 | Both |
| OpenAI third quarter | About $10 billion revenue, up about 50%; $10.9 billion adjusted operating loss | Axios, citing the Wall Street Journal | 16 Nov 2026 | Both |
| Morningstar fair value for Anthropic | $92 a share, against the $83 IPO price | Fortune | 18 Nov 2026 | Other camp |
| Nasdaq Composite after the dot com peak | Peak 5,048.62 on 10 Mar 2000; down 78% by Oct 2002 | Wikipedia, Dot-com bubble | Historical | Both |
Read that chart with care. Axios reports the two figures side by side, but they come from different sources and the revenue number is a ceiling, not a result. The sections below say which camp reads what into each row.
What does the camp that calls it a bubble say?
Its case is that spending has run ahead of what anyone has earned. Gary Gensler, a former SEC chair who is now a professor at MIT, said it plainly in the CNN interview: "It's already a bubble." He expects a repricing and a softer economy before November 2028, and said "I think history will show that they loaned too much money, too fast."
Other named voices CNN lists on this side are Ray Dalio, Harvard economist Kenneth Rogoff and Mike Novogratz, whom CNN names as having issued similar warnings without quoting them. Elon Musk, in an October 2026 post on X, said the 200 plus neocloud companies face "guppy level death and consolidation" and that "Some will make it through, but not many."
PCWorld's 8 signs the AI bubble may pop, published 20 January 2026, lists the arguments in plain order. Gains are concentrated in a few large companies. Money circles between AI firms and chipmakers. Spending outruns revenue. Small models that run on a laptop improve. The rally has lasted longer than most. Power and grid delays hold up data centers. Consumers are tired of AI features. Trade and Taiwan risk hang over the chips. Several of its numbers, such as OpenAI spending $150 billion for about $15 billion of revenue, carry no source on the page, so treat them as the author's claim.
The money trail is the part Axios stresses. It says Nvidia has helped finance the buildout by investing in customers and backing large projects, that the companies increasingly lean on debt and on each other, and that rising borrowing costs and a flood of new bonds are making lenders wary. That is the circular financing PCWorld warns about, now described by a news outlet with dates attached. Axios adds that political opposition to data centers could delay construction and raise costs.
Since that piece ran, the picture has moved. Axios reported that Nvidia chief Jensen Huang resigned on Thursday 12 November 2026, and CNN also describes his sudden exit and that CFO Colette Kress is interim CEO. Joe Brusuelas, RSM's chief economist, said his departure could trigger a "wholesale bursting." Vanda Research found retail traders became net sellers of US stocks in the 5 November sell off.
What does the camp that says it is not a bubble say?
Its case is that the technology is real and the money, so far, is following results. Ed Yardeni, of Yardeni Research, told CNN: "It's a self-correcting bubble." In the same interview he said "We're not looking at 1999-2000 all over again." Gensler himself says "The major companies will survive," and expects the breakage among neocloud companies.
The two camps overlap more than the headlines suggest. MIT economist Ricardo Caballero told Axios: "I have no doubt that AI will be transformative. That does not preclude a financial bubble." Technology investor Paul Kedrosky made the same point from the other side: "No sane person questions whether AI works." Columbia professor Stijn Van Nieuwerburgh said "AI can deliver enormous value to users" while warning that Huang's exit matters for the buildout.
On revenue, the strongest data point for this side is Anthropic. CNBC reports $33.24 billion of revenue in the nine months to 30 September, up about 1,200% on a year earlier, and an annualized run rate near $81 billion at the end of October. The bubble camp reads the same company's $104 billion nine month net loss as a warning, though CNBC says most of it came from changes in the fair value of convertible notes rather than from running costs.
Is AI a bubble like the dot com bubble?
Both camps use the dot com crash as their yardstick, and they read it in opposite ways. According to Wikipedia's account, the Nasdaq Composite peaked at 5,048.62 on 10 March 2000 and fell 78% from its peak by October 2002. Gensler says the coming downturn could be as significant as that one, but not as harsh as 2008. Yardeni says this is not 1999 and 2000 again.
Calling something a bubble also does not decide whether the technology survives. The World Economic Forum noted in October 2025 that every boom since Dutch tulip mania in 1637 has drawn a comparison to a past bubble, and that the dot com and tulip references are convenient shorthand rather than a measurement. PCWorld makes a different point: past rallies it counts lasted a few years, and this one has already run longer. It does not source those durations, so check them before you repeat them.
A fair summary of the comparison is that the dot com crash shows how far prices can fall, and does not show whether today's prices are too high. That depends on future revenue, which no one has. The AI predictions scorecard shows how forecasts in this field have fared.
What did the Anthropic IPO show?
It showed demand, and it showed a price below some earlier hopes. Anthropic priced its IPO on 18 November 2026 at $83 a share, above the guided range of $70 to $78, and was set to begin trading on the Nasdaq as ANT on 19 November. CNBC reports about $81.3 billion raised; Fortune puts it near $79.4 billion before the over allotment option. CNBC calls it the largest IPO on record, ahead of SpaceX's June offering.
Axios reports that the IPO terms imply a valuation well below earlier hopes of as much as $2 trillion, and CNBC puts the fully diluted market cap at about $1.5 trillion. Morningstar analyst Malik Ahmed Khan put fair value near $92 a share and said "Our valuation reflects the significant growth we expect from AI adoption." We could not confirm how the stock traded on its first day, so this page does not report it. Sam Altman and Dario Amodei disagree on much about the industry, including how fast to spend.
Questions to ask an AI about the AI bubble
Use a model with web search on, and ask it to separate fact from opinion. These work in Claude, ChatGPT, Gemini or Whizi, and none of them asks for a prediction.
- Sourced table: "Make a table of the five most cited figures in the AI bubble debate, with the number, the person or outlet that reported it, the date and a link. Mark any figure you can only find on one site."
- Steelman both sides: "Give the strongest case that AI is a bubble and the strongest case that it is not, in four sentences each, each with a named person and a dated source."
- Check a headline: "Here is a headline: [paste]. What does the number measure, what period does it cover, and who reported it first?"
- Dot com check: "Compare the dot com crash and today's AI spending on three measurable points, and say where each comparison breaks down."
- Define the terms: "Explain the difference between revenue, run rate revenue and net loss, with one example from a recent IPO filing."
- Spot the stake: "For each person I paste below, say whether they sell, fund, regulate or study AI, and how that might shape what they say: [names]."
What mistakes make AI bubble numbers misleading?
Four errors cause most of the confusion, and each has a fix.
- Comparing spending to revenue over different years. The fix: put both on the same period, and say when a source did not.
- Treating run rate as earned revenue. The fix: Anthropic's roughly $81 billion run rate and its $33.24 billion nine month revenue are different measures, so quote both with labels.
- Reading a net loss as operating cash burn. The fix: check the filing's explanation. CNBC says most of Anthropic's $104 billion loss came from the fair value of convertible notes.
- Repeating a number one site reports. The fix: label it as that site's report, as this page does for Axios on Gartner, or leave it out.
One app for comparing sources on the bubble debate
Whizi is one app where you can ask Claude, GPT and Gemini the same question with web search on, which is how you spot where models disagree on a figure. Compare puts two answers side by side on Pro, and its first week is $1.99. The free plan gives 10 messages a day on Whizi AI, Auto and Claude Haiku, with no card. See Whizi pricing.
What can AI not tell you about the AI bubble?
It cannot tell you whether prices are right, and it cannot give you a safe call on any investment. A model can summarize the arguments above, and it can also repeat a stale or invented number. Ask for a link and a date for every figure, and open the page. How to fact check AI answers turns that into a two minute routine, and why AI gets things wrong explains the mechanism.
For money decisions, talk to a licensed adviser. Related reading: AGI vs AI and will AI take over the world.
- Write down the date and source of every AI bubble figure you use
- Check whether a number is spending, revenue, run rate or net loss
- Put spending and revenue on the same years before comparing
- Note whether the speaker sells, funds, regulates or studies AI
- Label any figure only one site reports as that site's report
Frequently asked questions
Is AI a bubble right now?
Experts disagree, as of November 2026. Gary Gensler told CNN on 16 November it is "already a bubble," while Ed Yardeni calls it "a self-correcting bubble." This page makes no forecast. It lists the dated figures each side cites so you can read them yourself.
Will the AI bubble burst?
No one can say, and this page does not predict it. Gensler expects a repricing and a softer economy before November 2028, according to CNN. Caballero says AI can be transformative and still a financial bubble. Compare dated sources before you rely on any one view.
Is AI a bubble in the stock market?
Some analysts say yes and others say prices follow results. Axios reported a sell off on 5 November 2026, when retail traders became net sellers of US stocks, and Anthropic then priced its IPO at $1.5 trillion on 18 November. Neither fact settles the question, and this page is not investment advice.
Is the AI bubble like the dot com bubble?
The comparison is the most used and the most disputed. The Nasdaq Composite peaked on 10 March 2000 and fell 78% by October 2002, per Wikipedia. Gensler expects a downturn as significant as that one, and Yardeni says it is not 1999 and 2000 again.