The S&P 500 Has Fallen in 56% of Septembers Since 1928. Here's What That Means for Artificial Intelligence (AI) Stocks.

The real mechanism here is not the calendar; it is positioning in a crowded, high-duration factor trade. NVDA is less exposed to a generic September wobble because the market is still pricing a multi-quarter capex cycle, but it can de-rate quickly if investors use seasonality as a pretext to trim winners. MU is more levered: memory names tend to move on incremental changes in pricing and utilization, so any broad risk-off in September can create a sharper drawdown than the fundamentals alone justify.

The next 2-4 weeks matter more than the month itself. The Fed meeting and MU’s late-September earnings are the obvious catalysts, but the second-order issue is whether hyperscaler capex commentary stays intact into Q4. If AI spend is merely paused rather than cut, any September selloff should be bought; if one of the large buyers signals sequencing delays, semis can underperform for months even without an immediate demand collapse. A softer tape would also likely rotate capital toward lower-beta AI enablers and equipment names rather than away from the theme entirely.

Consensus is missing that the biggest risk is not seasonal weakness, but crowding and valuation sensitivity in the same names everyone uses as the AI proxy. That argues against chasing strength, but it also means a 5-10% pullback in NVDA/MU on macro noise is more likely a liquidity event than a thesis break. What would falsify that view: MU missing its guide or sounding cautious on HBM/DRAM pricing, or NVDA trimming its growth outlook by more than a low-single-digit percentage point; absent that, dips should be treated as entry windows rather than signals to exit.