The AI bubble is one of the most debated risks in global markets. After a multi-year run in everything tied to artificial intelligence, one camp sees valuations that have outrun reality and the other sees a boom backed by real profits. Officials have stopped treating the question as a footnote: in the survey of market contacts published with the Federal Reserve's May 2026 Financial Stability Report, AI was the second most-cited potential shock, behind only geopolitical and oil risks.
This article looks at what the term actually means, the evidence on both sides, what a burst would do to the stock market, and what it could mean for currencies.
What "AI Bubble" Actually Means
The phrase describes the hypothesis that a part of the stocks linked to artificial intelligence trades at prices too high for the revenues and profits those companies will actually be able to generate. It would be a financial bubble in the classic sense, with expectations growing faster than fundamentals.
The companies involved go well beyond a few chipmakers. They include cloud providers, software houses, data center operators, the suppliers of power and equipment behind them, and any business that promises to build artificial intelligence into its model.
One misunderstanding is worth clearing up at the start. The question is not whether the technology is useful. A technology can transform the economy and still produce valuations that reality fails to match. The Bank for International Settlements made the point in its 2026 annual report: the canal mania of the 1830s, the British railway mania of the 1840s, the electrification boom of the 1920s and the dot-com boom of the late 1990s all shared "a genuine technological breakthrough that attracted capital in excess of what commercial returns could ultimately justify."
The Case That It Is a Bubble
The arguments are real. The first is concentration. The ten largest companies accounted for about 40% of the S&P 500 at the end of 2025, against a peak of about 27% during 2000. The International Monetary Fund noted in October 2025 that the group of mega-cap stocks known as the Magnificent 7 made up 33% of the index on its own, and that its measure of concentration was substantially higher than during the dot-com bubble. By July 2026 the Bank of England estimated that AI-related companies represented around half of the S&P 500, up from around a quarter in 2022. What that concentration does to the relationship between the index and the average stock, and the indicators that measure it, is explained in market breadth: why the index rises while your stocks don't.
The second is the scale of the spending and the way it is paid for. According to the BIS, the five largest hyperscalers, the giants that build the computing infrastructure behind AI, are set to spend more than a trillion dollars on AI-related capital expenditure from 2025 through 2026. The BIS adds the detail that matters: these commitments are outpacing the earnings and free cash flow of the companies making them, which has led some of them to issue debt. By a Reuters count, the five sold about $121 billion of US corporate bonds in 2025, against an average of $28 billion a year between 2020 and 2024. Borrowing has also moved off the balance sheet: a joint venture with the private-credit manager Blue Owl, announced in October 2025, carries the roughly $27 billion cost of a single Meta data center campus in Louisiana, and the Bank of England has flagged such structures as a possible additional source of fragility in stress. A build-out paid for out of profits mostly puts shareholders at risk. One financed with borrowed money extends that risk to lenders and the credit market. The IMF has also flagged the spread of circular financing, in which AI developers, chipmakers and large technology firms act at the same time as each other's customers, investors and financiers.
The third is valuation. The cyclically adjusted price-to-earnings ratio of the S&P 500, which compares prices with ten years of inflation-adjusted earnings, stood at about 41 on 18 September 2026. Its record is 44.19, set in December 1999. In the words of an August 2026 analysis by European Central Bank economists, US valuations are "close to their historical peak."
The Case That It Is Not
The opposing arguments are just as concrete. Many of the companies leading the AI boom are already highly profitable, with large revenues, deep cash reserves and established businesses. That sets them apart from many internet companies of the late 1990s, which were valued almost entirely on expectations.
The IMF drew the same distinction in its October 2025 Global Financial Stability Report, which judged US equity valuations to be stretched, but less so than in the dot-com period. Earnings have so far held up their side of the bargain. More than 80% of S&P 500 companies beat analysts' estimates in the first quarter of 2026, by the IMF's count. In the July 2026 earnings round Amazon, Alphabet and Meta all raised their capital expenditure guidance for the year, and the one apparent reduction, at Microsoft, was a change in lease accounting and not a cut.
The honest conclusion is less clear-cut than the word bubble suggests. Some companies, some segments or some expectations may be overpriced without the whole AI sector being destined for a general collapse. It is also worth remembering that the IMF passage offering the reassurance on valuations is the same one that delivers the warning on concentration. The two facts arrive together.
What Happens to the Stock Market If It Bursts
The first effect would be the most obvious one: a sharp repricing of the stocks most exposed to the theme. Companies whose prices embed very aggressive growth would take the deepest cuts, especially if revenues or profits fell short. A single session has already shown how fast that can happen. On 27 January 2025, after the Chinese start-up DeepSeek released a model that raised doubts about how much computing power AI really needs, Nvidia fell about 17% and lost close to $600 billion of market value, the largest one-day loss for any company on record.
The damage would not stop at the AI names. Because the indices are so concentrated, a fall in a few very large companies drags the whole market with it. The arithmetic is simple: if the ten largest stocks make up 40% of an index and fall by 30%, the index loses 12% before any other stock moves. The Bank of England's Financial Policy Committee said as much in October 2025: any AI-led price adjustment "would have a high level of pass-through into the returns for investors exposed to the aggregate index."
How deep could it go? History offers one reference point. The Nasdaq Composite closed at 5,048.62 on 10 March 2000 and at 1,114.11 on 9 October 2002, a fall of about 78%. It took until 23 April 2015, fifteen years later, for the index to set a new closing record. The S&P 500 lost 49% over broadly the same period. Cisco, briefly the most valuable company in the world in March 2000 at about $555 billion, fell about 90% and did not close above its March 2000 share price again until 10 December 2025, more than twenty-five years on. Cisco remained a large and successful business throughout. Its shareholders had simply paid too much.
Official scenario work points to the same order of magnitude. In its July 2026 Financial Stability Report, the Bank of England modeled an AI-related correction starting in the United States: a 45% fall in the US equity market over six quarters, corporate credit spreads wider by 350 basis points and a persistent depreciation of the dollar, enough to take 2.2 percentage points off UK output. It is a hypothetical scenario, not a forecast, and the Bank notes that its macroeconomic effects are no larger than those of the stress tests UK banks have already passed.
Then come the indirect effects. A downgrade of expectations would slow investment in data centers, semiconductors, energy infrastructure and cloud services. By some estimates cited by the BIS, AI-related investment added about one percentage point to US real GDP growth in 2025, and the BIS warns that disappointing returns "could trigger a sudden pullback in financing and turn the capex boom into a protracted investment bust." The most indebted companies, and those that depend on outside financing, would be the most exposed.
Households are the other channel. The IMF's July 2026 update warns that a sharp repricing "would likely weigh on private consumption through wealth effects," and the exposure has never been larger: US households held a record 48% of their financial assets in equities in the second quarter of 2026, according to Federal Reserve data, compared with 39% at the peak of the dot-com boom. The effect would not stay in the United States. ECB economists estimate that euro-area households own about €440 billion of US technology stocks, mostly through funds.
The summer of 2026 offered a partial rehearsal. The Philadelphia Semiconductor Index set a record on 22 June and then fell more than 28% by late July, its worst month since 2008, while chip stocks worldwide shed more than $1 trillion over a few sessions and the Korean market dropped more than 30% from its peak. The S&P 500 nevertheless went on to a new record on 13 August, and a severe repricing in one part of the AI trade was absorbed by the broad index, which is either evidence of resilience or a sign that the largest names have not yet been tested. The backdrop has since become less forgiving: on 16 September 2026 the Federal Reserve raised its policy rate by a quarter point, to a range of 3.75% to 4%, its first increase since 2023.
What It Would Mean for Currencies
For forex traders the natural question is where the money would go. The textbook answer is the safe haven currencies: the dollar, the yen and the Swiss franc. The record is less tidy than the textbook.
During the dot-com bust the dollar kept rising for almost two years after stocks peaked and turned down only in early 2002. In the 2008 crisis it surged as investors around the world scrambled for dollar funding, while the yen was the strongest major currency of all: EUR/JPY fell by more than 30% between July and October 2008. In March 2020 the dollar jumped again, and the yen and the franc rallied first and were then sold as the rush for dollars took over.
April 2025 broke the pattern. When US tariffs sent the S&P 500 down about 12% in four sessions, the dollar fell with it. Between 2 and 21 April it lost 8.5% against the franc and 6.1% against the yen, while the euro gained 5.9% against the dollar. A San Francisco Fed study of the episode noted that safe-haven currencies such as the franc, the yen and the euro generally appreciated against the dollar. In the two Nasdaq corrections of 2026, however, the older pattern returned: the dollar rose and both the yen and the franc lost ground against it, in a year dominated by war in the Middle East and an oil shock. The Bank of England's AI scenario, for its part, assumes that the dollar weakens and stays weak, the opposite of the textbook reflex. As Andrew Hauser, deputy governor of the Reserve Bank of Australia, put it in March 2026, "the dollar has never been a perfect hedge for all risk-off events."
The practical reading is that the currency response depends on the nature of the shock. Funding stress tends to favor the dollar. A loss of confidence centered on US assets may not. In a risk-off phase the relationships between pairs also tighten and shift, which shows up in the correlation table, where the 30-day correlations among 29 markets are recalculated several times a day.
What Traders Can Watch
The bad news comes first: in a correction of the kind that would follow a burst, no automatic protection exists. What traders can do is monitor the conditions that raise the odds of an abrupt repricing, along with their own exposure to it.
- Portfolio concentration. Heavy exposure to a few companies, one sector or one theme amplifies volatility when sentiment turns. With the top ten stocks at about 40% of the S&P 500 at the end of 2025, holding the index is less diversified than it sounds.
- Valuations versus results. Late in a boom, attention shifts from stories about future growth to the concrete ability to produce revenues, profits and cash flow. In February 2026 Nvidia beat expectations for its quarter and still lost about $260 billion of market value in the following session.
- Leverage. Positions financed with borrowed money amplify moves in both directions, and forced selling feeds on itself. The Bank of England reported in July 2026 that hedge funds' equity prime brokerage balances were at record levels, up around 40% in a year, with positions increasingly concentrated in sectors such as semiconductors.
- Correlations between assets. Under stress, assets that normally behave differently start moving in the same direction, which reduces the benefit of diversification exactly when it is needed. The IMF observed in April 2026 that equities and long-term government bonds had been selling off together in the major economies.
For markets, the real question is not only whether an AI bubble exists but how much of the future is already in the price. The more prices anticipate results that still have to be demonstrated, the more sensitive investors become to quarterly earnings, investment plans, margins and any sign of a slowdown. Herd behavior does the rest, on the way up and on the way down.
