AI optimism is colliding with classic market warning signs
J.P. Morgan is warning that the AI trade may be showing symptoms of excess, with market gains increasingly concentrated in a small group of companies and speculative activity building around semiconductors. The bank’s concerns, as reported by The Decoder, do not amount to a forecast of immediate collapse. But they do suggest that the market’s most celebrated growth story is becoming more fragile as more capital crowds into the same names, instruments, and assumptions.
The warning comes after several years in which generative AI has reshaped capital spending, product roadmaps, and equity narratives across the technology sector. Since ChatGPT’s launch in 2022, enthusiasm around AI infrastructure has helped drive one of the strongest runs in recent market history, especially for chipmakers and hardware suppliers. J.P. Morgan’s argument is that the scale and structure of that rally now deserve closer scrutiny.
A small group of companies is carrying much of the market
One of the sharpest red flags in the report is concentration. According to the source text, just 42 AI companies in the S&P 500 have driven roughly 65% to 80% of the index’s profits, revenues, and investments since 2022. That means the market’s AI story is not broadly distributed. It is highly dependent on a relatively small cluster of companies.
The concentration is even more pronounced at the top. The ten largest U.S. stocks now account for about 40% of the S&P 500’s market capitalization, compared with 17% in 2015, according to J.P. Morgan. That kind of concentration does not automatically imply overvaluation, but it does raise the stakes. When gains are driven by a narrow leadership group, indexes can look healthy even while underlying breadth weakens. It also increases the risk that a setback in a few dominant firms could reverberate across broad portfolios.
The bank reportedly notes that, in global context, the U.S. still does not rank as the most concentrated market. Even so, concentration has risen sharply enough to become a material part of the AI investment debate. Investors are not just betting on technology adoption. They are making increasingly large, increasingly correlated bets on who will capture the profits.
Semiconductor trading is drawing the closest comparison to past bubbles
The most pointed comparison in the report concerns semiconductors. J.P. Morgan says technical patterns in the chip rally resemble those seen during the dotcom era. In practical terms, that means prices may be running far ahead of long-term trend lines, while investor positioning grows more aggressive.
The source text lists four warning signs highlighted by the bank. Semiconductor stocks are diverging from their 200-day moving average as sharply as they did during the dotcom bubble. Hedge funds are more heavily invested in chip stocks than ever before. Margin lending on the Korean stock exchange has tripled since 2020. And options trading in semiconductor stocks is now five times the 2020 level.
Those are not the signals of a calm market. They point to leverage, crowding, and short-term speculation, especially in a sector seen as the essential infrastructure layer for AI. The report also says leveraged chip exchange-traded funds have quintupled their influence on global stock markets since early 2024. Products that magnify price swings can intensify both rallies and reversals, adding another layer of instability if sentiment shifts.
Nvidia remains central, but the competitive picture is changing
J.P. Morgan’s caution is not only about valuations. It is also about the possibility that today’s dominant economics may prove less durable than the market assumes. Nvidia still holds the largest share of the AI accelerator market, but the bank estimates that its share could decline from 85% in 2023 to 75% by 2026.
That would still leave Nvidia in a commanding position, but the direction matters. The source text says custom chips from major cloud providers such as Google’s TPUs and Amazon’s Trainium can reduce operating costs by 30% to 40% compared with Nvidia GPUs. If hyperscalers and large model providers can lower training and inference costs with in-house or alternative hardware, the competitive moat around incumbent suppliers may narrow over time.
The report cites Anthropic’s commitment to run Claude on Amazon’s Trainium for the next decade as one sign that buyers are willing to pursue alternatives. This does not imply a near-term displacement of Nvidia, whose ecosystem and performance lead have shaped the entire AI buildout. It does indicate, however, that the market is beginning to test whether pricing power at the infrastructure layer can hold indefinitely.
High growth does not settle the profitability question
J.P. Morgan also flags a tension at the model-provider layer. Leading AI labs such as OpenAI and Anthropic are generating fast revenue growth, but compute costs remain enormous, and long-run profitability is still uncertain. That gap matters because the current AI investment cycle assumes not just adoption, but durable monetization across software and services.
The source text notes another pressure point: if token prices rise too much, customers may move workloads to cheaper open-source models. It also says there are already signs that companies are shifting tasks to lower-cost systems and that average token prices are falling. Those trends suggest price competition could intensify even as demand grows, putting pressure on margins for model providers.
In other words, the AI market may be confronting a familiar pattern from past platform shifts. Infrastructure leaders enjoy explosive demand, application players chase scale, and investors reward growth aggressively. But once competition expands and buyers become cost-sensitive, the economics can look less straightforward than the early narrative implied.
A warning, not a verdict
The significance of J.P. Morgan’s assessment is that it reframes the AI boom as both a technology story and a market-structure story. The bank is not arguing that AI lacks substance. It is arguing that genuine technological importance can coexist with investor exuberance, concentrated exposure, and unstable pricing dynamics.
That distinction matters. Some bubbles form around empty ideas. Others form around real transformations whose long-term impact is genuine, even if short-term valuations overshoot. The current AI cycle may ultimately produce lasting shifts in computing, enterprise software, and industrial productivity. But J.P. Morgan’s warning is that investors should not confuse strategic importance with immunity to speculative excess.
For now, the AI market remains powered by strong demand, large capital commitments, and a belief that the winners will define the next era of computing. The bank’s report suggests a harder question is emerging underneath that confidence: not whether AI matters, but whether the market has become too dependent on a narrow and increasingly leveraged version of that bet.
This article is based on reporting by The Decoder. Read the original article.
Originally published on the-decoder.com
