The market has become dangerously dependent on one story
AI has spent the past several years turning almost every concern into background noise.
Inflation remained stubborn? AI earnings were strong. Bond yields climbed? AI demand was stronger. Valuations stretched? Data-center spending kept accelerating. Geopolitical risk increased? Nvidia, hyperscalers and other technology leaders continued producing the numbers investors wanted to see.
That confidence has pushed global equities to record levels, but it has also created a market increasingly dependent on one assumption: the AI investment cycle can keep getting bigger without breaking the economics behind it.
Panmure Liberum strategist Joachim Klement thinks that assumption could begin failing as soon as 2027.
His year-end target for the S&P 500 is 5,000, implying roughly 36% downside from current levels. Such a decline would be the market’s worst since the global financial crisis and stands dramatically apart from the rest of Wall Street. Strategists tracked by Bloomberg are expecting about 14% upside on average.
Klement is an outlier. That does not make the warning irrelevant.
Markets become most vulnerable when one narrative grows powerful enough to explain away almost everything that could go wrong.
$713 billion has to earn a return eventually
The scale of the AI infrastructure buildout is becoming difficult to comprehend.
Hyperscalers could spend approximately $713 billion on data centers in 2026, more than twice last year’s total. Another increase is expected in 2027.
Hundreds of billions are flowing into GPUs, networking, cooling systems, power generation, land and enormous computing facilities. Suppliers are reporting record demand, semiconductor companies are expanding capacity, and analysts continue raising earnings forecasts across the AI ecosystem.
Yet the money funding that expansion is not unlimited.
Klement argues that free cash flow at the major hyperscalers has already been heavily depleted by capital spending. At the same time, borrowing costs have moved sharply higher. Financing another round of infrastructure becomes considerably less attractive when companies must issue expensive debt to build assets whose eventual returns remain uncertain.
A company can generate excellent earnings today while simultaneously making increasingly aggressive bets on tomorrow.
AI does not need to stop growing for the bubble thesis to work. Spending only needs to become too expensive relative to the returns it produces.
That distinction should make investors uncomfortable.
The most dangerous part may be the confidence
Bubble warnings are easy to dismiss after a market has spent years proving bears wrong.
Every pullback has recovered. Every concern about AI spending has been followed by another enormous capex announcement. Every debate about valuation has been answered with stronger earnings.
That history conditions investors to believe the next warning will also be wrong.
Klement described a market focused almost entirely on technology earnings, with investors willing to excuse macroeconomic, credit and financial headwinds because the AI story remains strong.
The psychology matters.
When investors expect bad news to be temporary and good news to validate the entire market, risk can accumulate quietly. Positions become crowded. Valuations assume continued execution. Capital spending forecasts extend years into the future. Even companies outside the immediate AI trade begin benefiting from the wealth and economic activity surrounding it.
Eventually, expectations can become harder to beat than the actual business.
A semiconductor company growing 50% can disappoint if the market expected 70%. A hyperscaler increasing AI revenue can still be punished if capital spending grows faster. Data-center demand can remain enormous while investors decide the returns no longer justify the price being paid for those earnings.
Bubbles rarely burst because the underlying technology suddenly becomes useless. They burst when reality stops improving fast enough to support the expectations built around it.
A correction in AI would not stay inside technology
The concentration of the rally makes that risk much larger.
AI spending now touches semiconductors, utilities, nuclear power, natural gas, networking, construction, copper, data centers and industrial equipment. Trillions of dollars in market value are tied directly or indirectly to expectations that the buildout continues.
A slowdown would therefore travel far beyond a handful of technology stocks.
Chip demand could soften. Data-center projects could be delayed. Utilities might reassess power forecasts. Equipment orders could slow. Highly valued software companies could face multiple compression. Investors suddenly looking for liquidity would likely sell far more than just AI names.
That is how a concentrated theme can become a market-wide event.
Klement is not forecasting an immediate collapse. In fact, he still believes markets could rise further before the danger becomes acute. His concern is centered roughly six to nine months ahead, and his recommended response is preparation rather than panic.
His key technical signal is the S&P 500’s 200-day moving average. A sustained break below that level would prompt him to move heavily toward defensive sectors such as food, pharmaceuticals and tobacco.
Whether that exact signal works is less important than the principle behind it: know what would cause you to change your view before the market forces you to make the decision under pressure.
The bubble does not have to pop tomorrow to matter today
Bullish investors have a credible argument. Corporate earnings remain strong, AI demand is real, and another year of profit growth could push equities higher. Citigroup is among the firms that still sees solid 2027 earnings supporting further market gains despite higher rates and geopolitical risks.
Both ideas can be true for a while.
AI may continue transforming the economy. Nvidia may keep selling enormous quantities of chips. Hyperscalers may continue building data centers at an extraordinary pace. Stocks could even reach new highs.
None of that eliminates the possibility that investors eventually paid too much for the growth.
Klement’s 36% downside target is an extreme forecast, and there is no reason to treat it as inevitable. What makes it worth paying attention to is the combination underneath it: enormous capital spending, shrinking free cash flow, expensive borrowing and a market increasingly convinced that AI earnings can overcome nearly every other risk.
History tends to punish that kind of certainty.
The scariest part of an AI bubble would not be discovering that artificial intelligence was overhyped. It would be discovering that AI was real, transformative and enormously valuable, but investors still managed to pay far too much for it.
Explore More Stories in Markets
Disclaimer
Wall Street Access is an independent financial publisher. Wall Street Access is not a financial advisor. No company mentioned has compensated Wall Street Access for the preparation, publication or distribution of this article. This content is provided for informational and educational purposes only and is not investment advice or a recommendation to buy, sell or hold any security. Investing involves risk, including the possible loss of principal.
Prices, market capitalizations, financial figures and other market data are based on information available as of the date and time of publication and may change without notice. Information is drawn from public sources believed to be reliable but is not guaranteed to be complete or accurate. Opinions and expectations expressed are our own and may change as new information becomes available.
Readers should conduct their own due diligence and contact a licensed financial professional before making any investment decision. Past performance does not guarantee future results. Any material ownership interest or conflict of interest, where applicable, will be disclosed separately.
