Capital Economics believes the artificial-intelligence boom that has driven U.S. stocks to record highs is entering its late stages, warning that several measures of market speculation are approaching levels associated with previous bubbles.
The research firm is not calling for an immediate market top. Senior markets economist James Reilly still expects the S&P 500 to climb from around 7,650 to 8,250 by the end of 2026, which would represent roughly 8% additional upside. Capital Economics then expects the index to fall to 6,500 by the end of 2027, a decline of about 21% from its year-end target.
Reilly’s concern is less about one expensive stock than the number of late-cycle signals appearing at the same time. Capital Economics tracks eight indicators associated with equity bubbles, including valuations, earnings expectations, market concentration, share issuance, leverage and foreign demand.
Several are now near historical extremes.
The strongest warning comes from earnings expectations. Long-term S&P 500 earnings-growth forecasts have reached a record, according to Capital Economics, while near-term expectations are approaching levels last seen around the dot-com boom. A large portion of that optimism is concentrated in technology and AI-related companies.
That leaves less room for disappointment if AI spending remains enormous but eventually produces slower-than-expected profit growth.
Concentration and New Stock Supply Are Rising
The S&P 500 has also become unusually dependent on a relatively small group of companies.
As of early September, the index’s ten largest positions accounted for about 38% of its weight. Nvidia alone represented more than 8%, followed by Apple, Microsoft, Amazon, Alphabet, Broadcom, Meta and other companies heavily exposed to AI investment or technology spending.
That concentration has worked extremely well while the largest companies have delivered strong earnings.
It also means disappointing results from a handful of companies can increasingly affect the entire index.
Capital Economics is watching equity issuance for a similar reason. Net issuance has turned positive, while the U.S. market has already experienced a major surge in IPOs and secondary offerings.
Companies raised roughly $307.7 billion through U.S. equity issuance during the first half of 2026, more than double the $137.4 billion raised during the same period last year. It was the second-largest first-half total since at least 2001.
Historically, strong markets often encourage companies and early investors to sell more shares while valuations are favorable. Capital Economics argues that another significant wave of IPOs or secondary offerings would strengthen the case that the AI rally is entering a more speculative phase.
Foreign ownership of U.S. equities has also reached a record, according to Reilly, another sign of how much global capital has crowded into American stocks.
Not Every Bubble Signal Is Flashing Red
The bearish case is not complete.
Capital Economics says leverage has moved in a concerning direction but has not yet reached the extremes typically associated with the final stage of a bubble. Volatility also remains relatively subdued rather than showing the behavior seen around some previous market peaks.
More importantly, corporate earnings are still providing fundamental support for the rally.
S&P 500 earnings grew roughly 33.5% from a year earlier during the second quarter, according to LSEG data cited by Reuters. About 86% of companies reporting results beat analyst expectations, well above the long-term average. Barclays recently raised its 2026 S&P 500 earnings estimate to $365 per share and lifted its year-end index target to 7,950.
That strength separates the current market from a purely speculative rally with little underlying profit growth.
AI spending also remains extremely strong.
Oracle’s latest quarter offered a fresh example. Its cloud infrastructure revenue jumped 121% from a year earlier, total revenue increased 30% to $19.3 billion, and its contracted revenue backlog reached $664 billion as demand for AI computing continued to expand.
Those numbers make it difficult to argue that AI demand itself is already collapsing.
The risk is that expectations may have risen faster than even strong operating results can consistently satisfy.
The Market May Need Better Results Just to Keep Rising
Capital Economics’ warning therefore depends heavily on what happens next to earnings rather than simply whether companies continue spending on AI.
High expectations are manageable when companies repeatedly exceed them. They become more dangerous when earnings merely meet forecasts or when massive capital expenditures fail to produce the margins and cash flows investors have priced into stocks.
Another factor is valuation. The S&P 500’s forward price-to-earnings ratio recently fell to about 19.2 as earnings estimates increased, its lowest level in roughly 17 months. That makes the market less expensive on a conventional forward-earnings basis than it was earlier this year, even while other measures used by Capital Economics remain stretched.
That mixed picture helps explain why the firm still expects stocks to rise before its projected downturn.
Reilly’s thesis is not that investors should expect an immediate collapse. It is that the combination of aggressive earnings assumptions, extreme concentration, rising issuance and record foreign exposure increasingly resembles the later stages of previous market booms.
The next several earnings seasons may provide the clearer test.
If Nvidia, Microsoft, Alphabet, Amazon, Meta, Oracle and other major AI beneficiaries continue turning record capital spending into faster profit and cash-flow growth, elevated expectations could remain supported longer than skeptics expect.
If earnings revisions begin falling while share issuance and market concentration continue rising, Capital Economics’ late-stage warning would become much harder for the market to ignore.