J.P. Morgan warns of AI market red flags and chips

J.P. Morgan warned on June 27, 2026 that "there are signs of investor exuberance" in AI-related financial markets, calling out concentrated gains, speculative activity in semiconductors, and mounting infrastructure risks. The bank lays out technical and market indicators it says echo late-1990s excess even as compute-driven businesses scale rapidly.

What red flags did J.P. Morgan identify?

J.P. Morgan pointed to four specific warning signs: semiconductor stocks deviating sharply from their 200-day moving average similar to the dotcom bubble, heavy hedge fund exposure to chip and hardware names, margin lending in Korea that has tripled since 2020, and options trading in semiconductor stocks at five times the 2020 level. The bank also highlights that leveraged chip ETFs, which amplify price swings, have quintupled their influence on global stock markets since early 2024, and that retail traders are piling into semiconductor options.

Those market-structure signals sit alongside operational concerns: leading AI labs have fast-growing sales but massive compute costs, and J.P. Morgan warns future profitability for those businesses is unclear. The bank frames these items as a stack of risks that interact — technical trading patterns, concentrated ownership, rising leverage, and high operating costs.

How concentrated are the gains and which firms are carrying them?

J.P. Morgan says concentration is extreme: since ChatGPT launched in 2022, just 42 AI companies in the S&P 500 have driven roughly 65 to 80 percent of the entire index's profits, revenues, and investments, and the ten largest US stocks now account for about 40 percent of the S&P 500's market cap, up from 17 percent in 2015. The bank notes that despite this increase the US still ranks among markets with relatively low concentration globally, with only India and Japan less concentrated.

Nvidia remains dominant in AI accelerators, but J.P. Morgan projects its share slipping from 85 percent in 2023 to an estimated 75 percent by 2026, as custom chips from major cloud providers gain ground. The bank points to cloud-provider silicon like Google TPUs and Amazon Trainium, saying those custom chips cut operating costs by 30 to 40 percent compared to Nvidia GPUs. Anthropic's commitment to run Claude on Amazon's Trainium for the next decade is cited as a concrete example of that shift.

Why it matters

Concentration and leverage change how shocks propagate. When profit and investment flows are carried by a narrow set of firms, an abrupt repricing of those names can ripple through broader indexes and investor portfolios. J.P. Morgan ties the trading and leverage signals to operational risks at AI labs, where high compute costs and uncertain paths to profitability raise the odds that rising token or compute prices could force customers toward cheaper models or cloud silicon.

The bank is not alone in flagging systemic exposure: NYU finance professor Aswath Damodaran warned an AI crash could hit harder than the dotcom bust, a comparison J.P. Morgan echoes with its blend of valuation, leverage, and infrastructure concerns.

What to watch

Watch the four market signals J.P. Morgan lists: deviations from 200-day moving averages in semiconductors, hedge fund positioning in chip stocks, Korean margin lending (which has tripled since 2020), and options volume in semiconductor names (about five times the 2020 level). Also monitor Nvidia's market share trajectory from 85 percent in 2023 toward J.P. Morgan's 75 percent estimate for 2026, and adoption of cloud-provider silicon and shifting token prices, both cited as cost pressure points that could alter vendor and customer behavior.

Taken together, the bank frames these items as multiple layers of concentration risk across markets, infrastructure, and the broader economy, rather than a single isolated bubble. The next concrete confirmatory signals will be whether the trading and leverage metrics roll back or accelerate, and whether firms materially shift compute footprints to cheaper custom chips.