September gets blamed for being the market’s worst month, but the calendar isn’t what has me worried this year. What concerns me is what happens when a market already heavily dependent on artificial intelligence spending, leverage, and a handful of giant companies finally has to prove those expectations are justified.
September is the only month with a negative long-run average return. Since 1928, the S&P 500 (^GSPC -0.71%) has declined an average of 1.2% in September, worse than every other month. Since 1950, it has finished positive just 44% of the time, the only month below a coin flip. Nine of the 40 worst monthly losses in market history landed in September, more than any other month.
The recent record is uglier — in 4 of the past 5 Septembers, the market fell by an average of 4.2%, more than triple the long-term average decline. When September finishes in the red, the average loss is 3.8%. Here is the honest caveat: Remove a handful of outlier years, and September performs positively slightly more than half the time. Seasonality is not destiny, and one useful filter matters: When the index sits above its 200-day moving average entering September, the average return flips to positive 1.3% with 60% of occurrences higher.
Image source: Getty Images.
Why 2026 is probably going to be different
The seasonal statistic is background noise. The foreground is how AI is being financed.
Hyperscaler capital expenditure has gone from 33% of cash flow from operations in 2023 to an estimated 93% in 2026. Read that again. The cash piles are gone. Sell-side estimates for 2026 capex among five United States companies now sit at $697 billion, up $173 billion since January alone. Aggregate hyperscaler capex is forecast near $625 billion this year, with $5 trillion projected through 2030.
When spending consumes nearly all of operating cash flow, the marginal dollar must come from debt. Some analysts flag exactly that: investments increasingly reliant on debt, counterparty and opacity risks in contracts, and complex circular financing structures that create a notable risk of overstretch.
This is what actually concerns me. In my opinion, the AI economy has become a web in which the same dollars circulate as revenue among multiple companies. Nvidia invests in cloud and model providers; those providers commit back to hyperscaler infrastructure, and hyperscalers buy chips from Nvidia. For example, Amazon agreed to invest $15 billion in OpenAI, with an additional $35 billion contingent on certain conditions.
Bloomberg framed the situation similarly and in blunt terms. If a company has its supplier as a major shareholder, it becomes more likely to keep buying, whether the purchase makes commercial sense or not. And the risk concentrates precisely when a handful of buyers represent most of the market, which describes AI exactly. If demand disappoints, the sponsor loses twice: the customer stops buying, and the equity stake falls.
Today’s Change
(-0.71%) -54.67
Index Level
7,631.47
Key Data Points
Day’s Range
7,611.20 – 7,663.63
52wk Range
6,316.91 – 7,816.70
The valuation gap
Analysts at Goldman Sachs put a number on the expectations embedded here. AI-related companies have added roughly $27 trillion in market value since late 2022, against a baseline estimate of $9 trillion for the present value of AI-related capital revenues. Seven months earlier, that gain was $19 trillion, so the gap widened by $8 trillion in half a year.
Closing that requires AI companies to capture an unusually high share of economic gains and sustain above-average profits longer than normal cycles allow. Goldman’s analysts explicitly warn investors may be overestimating profit durability, particularly for infrastructure suppliers.
What worries me most
The “Magnificent Seven” stocks now represent about a third of the S&P 500 market cap, the highest index concentration in modern market history. Five are direct AI beneficiaries. A standard 60% United States equity allocation already carries heavy AI exposure, whether you choose it or not.
So September’s pattern is not a threat to AI investors. The threat is that a stretched, debt-financed, circularly funded trade meets the month most prone to profit-taking, with almost no valuation cushion left to absorb the impact. If a drawdown would force you to sell, your position is already too large.
