J.P. Morgan is warning of an emerging AI bubble in the stock market, pointing to extreme stock concentration and chip-trading patterns that echo the dotcom era. The bank says "signs of investor exuberance" are spreading across AI-related assets just as a tiny group of companies carries an outsized share of the entire US market's gains, raising the stakes if sentiment turns.
An AI-driven market built on a handful of stocks
Since ChatGPT launched in 2022, just 42 AI companies in the S&P 500 have driven roughly 65 to 80 percent of the index's profits, revenues and investments, according to J.P. Morgan. That is a remarkable share of an index built from 500 names, and it means the broad market's health now rests on a very narrow base of mega-cap technology and semiconductor firms.
The concentration shows up across several metrics the bank tracks:
- The ten largest US stocks now make up about 40 percent of the S&P 500's market cap, up from 17 percent in 2015.
- Leveraged chip ETFs have quintupled their influence on global markets since early 2024.
- Hedge funds are heavily exposed to chip and hardware stocks, while retail traders pile into semiconductor options.
The bank does add global context: despite the jump, the US still ranks among developed markets with relatively low concentration, with only India and Japan less concentrated. That nuance matters, but it does little to offset how quickly a few names have come to define US equity returns. For how these companies are valued, see our business coverage.
Nvidia's slipping share and China's open-source squeeze
Nvidia still holds the largest slice of the AI accelerator market, but J.P. Morgan estimates its share is falling from 85 percent in 2023 to around 75 percent by 2026. The erosion is not from a rival GPU maker but from the cloud giants designing their own silicon. Custom chips such as Google's TPUs and Amazon's Trainium are said to cut operating costs by 30 to 40 percent versus Nvidia GPUs, a gap large enough to reshape spending. Anthropic, for one, has committed to running its Claude models on Trainium for the next decade.
Margin pressure is mounting on the model makers too. The bank flags that fast-growing revenue at labs like OpenAI and Anthropic comes with massive compute costs and unclear future profitability. Rising token prices could push customers toward cheaper open-source alternatives, and there are early signs of exactly that: average token prices are falling, and Chinese open-source models are closing in on top-tier performance at a fraction of the cost. If frontier intelligence becomes a commodity, the premium pricing underpinning today's valuations gets harder to defend.
Why the AI bubble risk matters across markets
Taken together, J.P. Morgan argues AI is stacking up concentration risk across three layers at once: financial markets, physical infrastructure and the broader economy. Tech investment's share of economic growth is rising even as cloud providers' free cash flow margins shrink and their debt financing grows. That combination, heavy capital spending funded increasingly by borrowing, is precisely what makes a downturn self-reinforcing rather than contained.
The warning is not the bank's alone. NYU finance professor Aswath Damodaran has voiced a similar concern, cautioning that an AI crash could hit harder than the dotcom bust because so much of the real economy is now wired into the same handful of suppliers and customers.
None of this is a prediction of an imminent crash. Markets can stay concentrated and expensive for long stretches, and AI revenue is real and growing. But the bank's message is that the margin for error has narrowed: when 42 companies underwrite most of an index's earnings, any stumble in chip demand, model economics or capital discipline reverberates far beyond the firms involved. Track the latest developments on our AI news page.
Frequently asked questions
Is J.P. Morgan saying the AI market will crash?
No. The bank is flagging concentration risk and signs of exuberance, not forecasting a specific crash. Its point is that the market has become unusually dependent on a small group of AI and chip stocks, which raises the consequences if those names falter.
Why is Nvidia's market share falling?
Cloud giants are designing custom chips, such as Google's TPUs and Amazon's Trainium, that J.P. Morgan estimates cut operating costs by 30 to 40 percent versus Nvidia GPUs, pulling some demand away from off-the-shelf accelerators.
What does stock concentration mean for ordinary investors?
It means broad index funds are more exposed to a few mega-cap stocks than in the past, so the fortunes of a handful of AI companies have an outsized effect on portfolios that look diversified on paper.
