Two months ago, JPMorgan believed that deleveraging in US equities had already cleared out the prior excess. Now it is taking back half of that statement.

In its September 30 "Flows and Liquidity" weekly report, the bank's global markets strategy team said that "excess" in some leverage and equity positioning metrics has returned to the market and could become a headwind for fourth-quarter equities—the report's main title is "Equity Market Fragility." Strategist Nikolaos Panigirtzoglou wrote: "Elevated equity positioning and leverage have re-emerged, and although not to the same extent as in June and July this year, they still pose some challenges for fourth-quarter equities."

Seven indicators: where leverage has come back from

The report breaks the return of this "excess" into seven chains of evidence. First, leverage built through US equity index futures has rebounded and is close to year-to-date highs. The bank's positioning gauge is based on US Commodity Futures Trading Commission (CFTC) data and tracks asset managers' and leveraged funds' positions in S&P 500, Dow Jones, Nasdaq and their mini contracts, calculated as a share of open interest. Second, a broader composite gauge of equity positioning peaked in September, after earlier highs in January 2026 and August 2025. Third, short interest in the world's largest equity ETF, the SPY tracking the S&P 500, began bottoming in early September after hitting an all-time low. Fourth, previously elevated short interest in semiconductor ETFs (SMH and memory-stock ETF DRAM) has normalized, indicating that "the prior short-covering is largely over." Fifth, momentum signals show that trend-following traders (CTAs) have begun rebuilding longs in Nasdaq, South Korea's KOSPI, Taiwan and Nikkei, "but still far below prior extremes"; the report also noted that the relative trade popular from April to June—"sell Chinese internet companies (HSCEI), buy Korean and Taiwanese equities"—reappeared in August and September. Sixth, the ratio of leveraged equity ETF assets to underlying stock market capitalization has picked up in recent weeks and is now at about 70% of the range between the April low and the June high; the rebound in memory-related leveraged products is less obvious and remains at about 50% of that range. The report warned that if the overall size of leveraged ETFs continues to expand, it could again become a risk point for equities as a trigger for deleveraging and a source of excessive volatility. Seventh, and what the report views as the most concerning: margin-account leverage.

The hardest piece of evidence: $1.45 trillion and 4.5% of GDP

JPMorgan uses the New York Stock Exchange margin account net debit balance as a proxy for US retail investor leverage. The report said the metric was at a "very high level" in August, and the deleveraging in June and July barely changed it, so it "remains a major vulnerability for equities." The report explained the mechanism: hedge funds can add leverage more easily and at lower cost through options and futures, while retail investors are more constrained by the Federal Reserve's Regulation T, which allows borrowing up to 50% of the purchase price when buying securities on margin; net debit balances equal margin debit balances minus the sum of cash account and margin account credit balances. Market-side data corroborates this judgment. Monthly statistics from the Financial Industry Regulatory Authority (FINRA) show that US margin debt rose by about $36.6 billion, or 2.6%, in August to $1.4538 trillion, the second-highest on record after the June peak of $1.502 trillion; so far this year it has increased by about $228 billion, or 19%, and is up about 37% year over year. Over a longer horizon, investor borrowing has increased by $847 billion, or 140%, since the end of 2022, while the S&P 500 rose 98% over the same period—leverage has expanded clearly faster than market capitalization itself. Relative to the size of the economy, margin debt now equals about 4.5% of US GDP, above the roughly 3.6% cycle peak in 2021 and 2.8% during the 2000 dot-com bubble. JPMorgan Chairman and CEO Jamie Dimon has also publicly said that "market leverage is quite high." One distinction in methodology is needed here: JPMorgan uses NYSE net debit balances, while FINRA publishes customer margin account financing balances. The two have different statistical scopes and should not be directly equated or added together.

Why JPMorgan remains bullish on tech: three fundamental pillars

Despite flagging positioning and leverage headwinds, the weekly report did not turn bearish. It stressed that the bank's equity research team still believes the tech/AI sector has fundamental support, and that "even another equity VaR shock similar to June and July would not derail the tech bull market." Support comes from three areas: First, memory prices remain in an uptrend, providing fundamental support for memory makers—long a high-beta segment within tech/AI. This is consistent with this publication's earlier storage tracking: a Goldman Sachs report on October 1 expected DRAM and NAND contract prices to continue rising in the fourth quarter and identified enterprise SSDs as the only accelerating category. Second, hyperscalers' capital expenditure expectations have been significantly revised up. Based on aggregated bottom-up analyst consensus, combined capex for the five largest companies (Google, Amazon, Meta, Microsoft and Oracle) is expected to be $804.9 billion in 2026 and $1.0918 trillion in 2027, above the $758 billion and $925 billion expected on July 1. This consensus level is broadly in line with market expectations cited in Goldman Sachs' September 24 report (about $800 billion in 2026 and $1.1 trillion in 2027); Goldman's own forecasts are more aggressive, at $1.2 trillion in 2027 and $1.4 trillion in 2028, and it estimates that for these investments to break even on average annual capex in 2026 and 2027, about $300 billion in annual AI revenue would need to be generated in the coming years. Third, AI compute prices remain firm. JPMorgan uses Compute Desk's Hopper US Index as a proxy—the index aggregates on-demand and reserved prices for Nvidia H100/H200 rentals from US neocloud providers, in dollars per GPU per hour—and noted that after continued pressure at the end of 2025, compute prices have shown signs of improvement in recent months; the report argued that the higher compute prices are, the stronger hyperscalers' ability to maintain or improve margins, and that the index's trend also challenges some investors' assumptions that "overestimate equipment obsolescence and depreciation." Independent industry data points in a similar direction: compute pricing agency Ornn's H200 settlement price index was reported at $5.96 per GPU per hour on October 1, up 32.2% over 30 days, with a three-month range of $4.01 to $5.96.

The other side: deteriorating breadth, "if it's not AI, it's failing"

Beyond the warnings on positioning and leverage, September also left several less healthy marks on US equities themselves. For the full month, the S&P 500 edged down 0.4%, 1.9% below its August record high; the Nasdaq Composite rose 1.8% and set a record high on September 22; but the Dow Jones Industrial Average fell 4.3%, and the small-cap Russell 2000 dropped 5.4%. Interactive Brokers chief strategist Steve Sosnick summed this up as a market where "if it's not AI, it's failing." A more specific breadth signal appeared on September 21: the S&P 500 rose 1.49% that day to 7,764.70, just about 0.4% below its record close, yet the number of constituents hitting 52-week lows (30) exceeded the number hitting 52-week highs (7), the first such combination of "index at highs, net new highs negative" since December 1999; according to Dow Jones Market Data, 59.2% of S&P 500 constituents were already at least 20% below their own record highs. The same set of facts has also produced two voices within JPMorgan. Just one day before the weekly report warned that "positioning and leverage excess has reappeared," the bank's market intelligence team led by Andrew Tyler turned tactically bullish in a September 29 client note, citing stronger-than-expected macro data, a stabilizing bond market and lower oil prices, and expecting tech and semiconductors to continue outperforming. The other side of the disagreement comes from the scale of AI investment itself. Morgan Stanley strategist Michael Wilson said that another rise in oil prices and refined product prices is the main near-term risk preventing the market from reaching his year-end target; UBS chief economist Arend Kapteyn estimated that AI-related investment and the wealth effect from rising AI stocks have accounted for more than 80% of US economic growth, while capex in non-AI sectors is "almost zero"—both a support for growth and a source of concentration risk.

A two-way judgment

For the fourth quarter, the report actually offers a two-way judgment: on one hand, the "excess" in positioning and leverage has returned through multiple channels, with margin-account leverage the most stubborn and the one closest to retail investors; on the other hand, JPMorgan believes the three fundamental pillars of tech/AI—memory prices, trillion-dollar capex and firm compute prices—remain intact, so the more likely scenario is amplified volatility rather than the end of the bull market. The report did not provide index-level forecasts, but it narrowed the observation window sharply: what really needs watching is whether any of these three pillars begins to loosen.