AI bubble: will it burst, and what it means for AI risk

Is AI a bubble or the future?

Several people on the record give the same two-part answer: some of the money is overexcited, and the technology still matters.

• -Sam Altman (OpenAI), August 2025: "Are we in a phase where investors as a whole are overexcited about AI? My opinion is yes." And straight after: "Is AI the most important thing to happen in a very long time? My opinion is also yes." He compared it with the dot-com era: "When bubbles happen, smart people get overexcited about a kernel of truth."3
• -Jamie Dimon (JPMorgan), October 2025, to the BBC: "AI is real, AI in total will pay off. Just like cars in total paid off, and TVs in total paid off. But most people involved in them didn't do well."4
• -Sundar Pichai (Google), November 2025, to the BBC: Reuters reports he called the moment "extraordinary" but saw "elements of irrationality" in the market, adding: "I think no company is going to be immune, including us."5
• -Jerome Powell (US Federal Reserve), 29 October 2025: asked about the 1990s, he said the most highly valued companies "actually have earnings", unlike the dot-com firms, which "were ideas rather than companies". His summary: "So it's really a different thing."6

So "bubble or the future?" is a false choice for most of the people quoted: they see both. Whether the prices are too high is a question for markets. This page doesn't answer it.

Will the AI bubble crash the economy?

Central banks and the IMF have said a sharp fall is a risk worth watching, without forecasting one.

• -Bank of England, 8 October 2025: "The risk of a sharp market correction has increased." On one backward-looking measure, US share prices were "comparable to the peak of the dot com bubble", though on a forward-looking measure they "remained below the levels reached during the dot com bubble". Among the things that could trigger a fall, the committee listed "disappointing AI capability/adoption progress or increased competition".1
• -IMF, 8 October 2025: Managing Director Kristalina Georgieva said "Today's valuations are heading toward levels we saw during the bullishness about the internet 25 years ago" and "If a sharp correction were to occur, tighter financial conditions could drag down world growth".17
• -IMF, 13 October 2026: Georgieva said: "We now have an economy that is standing on one leg, the AI leg. What if it wobbles?" In a downside scenario, the IMF said a steep drop in AI investment could lead to a sharp fall in asset prices, "ultimately cutting global GDP by 1.1% by 2028, including a 1.8% drop in US output" (Reuters' summary). The IMF's Jason Wu: "If there are questions about the monetisation of AI, the equity investments will deplete in value, the debt may be difficult to service, and the revenue commitments in the future may not materialise."2 Two days later, at the Annual Meetings in Bangkok, Georgieva used the same image again: "We have a world economy that is like on one leg" (Nation Thailand, 15 October 2026).18
• -Bank of England, 30 September 2026: "Equity valuations for AI companies fell sharply in July." The fall was "amplified by an unwinding of stretched positions", and "Despite significant losses for some leveraged investors with concentrated positions, there was no spillover to core markets." Its view: "The risk of a sharper correction persists, notably if there is a more significant shock to earnings expectations reflecting concerns around the pace of AI development or adoption."19

These are scenarios and warnings, not predictions. What happens if the AI bubble bursts depends on how big the fall is, who holds the debt, and what else is going on in the economy at the time; none of the sources above puts a date or a probability on it.

When will the AI bubble burst?

Nobody knows, and the people quoted on this page say so.

• -Benjamin Todd (writer on AI and careers), April 2024: "The end of stock bubbles often have no obvious trigger. At some point, the stock of buyers gets depleted, prices start to move down, and that causes others to sell, and so on."8
• -Yarrow Bouchard (EA Forum), December 2025, who thinks a bubble is likely: "Nobody can predict when a bubble will pop, even if they can correctly call that there is a bubble."12
• -Jamie Dimon, October 2025: he saw a higher chance of "a meaningful drop in stocks in the next six months to two years" than markets were pricing, and said "The level of uncertainty should be higher in most people's minds" (CNN's report of his BBC interview).4

As of 16 October 2026, AI shares have already had one sharp fall: the Bank of England says equity valuations for AI companies "fell sharply in July" 2026, without spilling over into core markets.19 Both the Bank and the IMF still describe a bigger correction as a risk, not as something that has happened.192 We don't add a forecast of our own.

AI bubble vs dot com: what happened last time?

The dot-com crash is the comparison almost everyone reaches for.

• -The fall: the Nasdaq Composite peaked at 5,048.62 on 10 March 2000; by October 2002 it had fallen to 1,114.11, "a total loss of 78% against its peak" (Wired, 2010; The Guardian, 2005).15
• -What survived: MIT Technology Review, December 2025: the dot-com bubble "wiped out a lot of companies, which sent ripples across the world, but it left behind the infant internet—an international network of cables and a handful of startups, like Google and Amazon, that became the tech giants of today." And: "The companies that went bust back then didn't have the money to last the distance. Those that survived the crash thrived."7
• -Where the comparison breaks: Powell (October 2025) said today's leaders have earnings; the Bank of England (October 2025) found valuations above dot-com levels on one measure and below on another.61

The lesson that matters for this page: the dot-com crash ruined many companies and investors, but it did not stop the internet. A commenter on the EA Forum made the same point about AI in December 2025: "the internet carried on growing after a bunch of overpromisers who misspent their capital fell by the wayside."13

Will an AI bubble burst slow down AI advancement?

This is the question that links money to risk. There are arguments both ways.

Why a crash could slow AI down

• -Frontier AI runs on money. Epoch AI, which tracks the field, reports that the compute used to train frontier language models has grown about 5× a year since 2020, and that training costs have grown about 3.5× a year; frontier AI labs "have raised more than $370 billion" (page updated 5 February 2026).9 A trend that steep needs investors to keep paying.
• -Less capital, fewer bets. Benjamin Todd wrote in April 2024 that "A crash in AI stocks could cause a modest lengthening of AI timelines, by reducing investment capital. For example, startups that aren't yet generating revenue could find it hard to raise from VCs and fail." His own estimate: a 50% fall in AI stock prices "could maybe lengthen my estimates for when transformative AI will arrive by a couple of years".8
• -A crash might be a signal. Todd: "A crash in AI stocks (depending on its cause) might also tell us that market expectations for the near-term deployment of AI have declined." The Bank of England also listed "disappointing AI capability/adoption progress" as one possible trigger.81

Why a crash might not change much

• -Crashes and capability can come apart. Todd thought a sell-off "could happen even if capabilities keep advancing (e.g. maybe because real world deployment is slow)".8 Something like that showed up in 2026: in the same record that notes July's sharp fall in AI share valuations, the Bank of England says "AI capabilities had continued to advance across leading providers", including in "models' ability to complete complex tasks without human direction".19
• -Survivors keep building. The dot-com record above: the companies with money to "last the distance" thrived.7
• -Released models don't disappear. The International AI Safety Report 2026, chaired by Yoshua Bengio, says: "Open-weight model releases are irreversible. Once released, model weights cannot be recalled." It adds that open-weight releases "have narrowed the capability gap with leading closed models".11 Whatever happens to share prices, models already published stay in the world.

What a crash could do to the safety side

Todd also warned that a crash could hit AI safety work itself: "the wealth of many donors to AI safety is pretty correlated with AI stocks", so "an AI crash could easily cause a tightening in the funding landscape". He expected a shift in mood too: people warning about AI "could get branded as alarmists", and "it would likely become harder to push through policy change for some years as some of the urgency would drop out of the issue."8 Bouchard expected the opposite on money ("Not because it will make funding dry up — it won't") but hoped a pop would make the safety community rethink "the imminent AGI narrative".12

Is another AI winter coming?

An "AI winter" is a period when money and interest in AI research dry up. There have been two, by the usual account: roughly 1974–1980 and 1987–2000 (Wikipedia; sources disagree on the exact dates).14 The first is usually tied to the UK's 1973 Lighthill report and US defence funding cuts; the second to the collapse of the market for specialised LISP machines in 1987.14

Even that history is argued over. Historian Thomas Haigh wrote in Communications of the ACM in December 2023 that his data sources do not support "the idea of a broadly based AI winter during the 1970s": the main AI interest group had 1,241 members when the Lighthill report came out in 1973, about twice its 1969 level.16

Is another one coming? Our keyword research found that the top search results for "is ai winter coming" are prediction markets, not analysis, and none of the sources we opened makes a dated forecast. Russell noted in January 2026 that "many observers describe a bubble that is about to burst", citing underwhelming revenue and signs that large language models have "plateaued"; he did not say a winter is coming.10 A winter in the old sense would mean research funding and progress stalling, which is a different thing from share prices falling. More on the term: glossary.

Why risk researchers say capability, not stock prices, drives the risk

The case for worrying about advanced AI, as set out on our p(doom) and superintelligence guides, is about what AI systems can do and whether people can control them. Share prices are one input into how fast that happens; they are not the risk itself. Three people on the record:

• -Stuart Russell, January 2026: a burst means "severe" economic damage; no burst means breakthroughs toward AGI, and "AI developers have no cogent proposal for how to control such systems, leading to risks far greater than economic damage."10
• -Benjamin Todd, April 2024: even a 50% stock fall "wouldn't have a huge impact on my all considered view about how many resources should go into AI safety." He also said it made him "want to be more cautious about hyping short term capabilities, since that makes it sound like the case for AI Safety depends on them."8
• -EA Forum commenter "David T", December 2025, a sceptic of near-term AGI: "The AI bubble popping … is somewhat orthogonal to the imminent AGI hypothesis".13

People who think AGI is far away and people who think it is close both say the same thing here: a crash tells you about money, and only indirectly about machines. If capabilities stall, the risk picture changes for that reason; if they keep improving, it doesn't matter much that the stock market had a bad year. Forecasts like AI 2027 are about capability milestones, and our AI 2027 tracker checks them against what actually happens, not against share prices.