Will NASDAQ fall at least 65% from its historical peak at any time before the end of 2030?

TL;DR

My forecast is 9%. Only the dot-com collapse crossed 65% from a fresh Nasdaq peak in the index’s full history; the 1973–74 bear market and global financial crisis stopped at 59.9% and 55.6%, respectively (FRED, AJ Bell/Refinitiv). The main path to YES is an AI investment bust compounded by recession, persistent inflation, or credit stress; an AI disappointment by itself is more likely to cause a conventional 30%–50% bear market.

Context

The Nasdaq Composite closed at 25,373.85 on July 31, 2026, versus its record close of 27,093.90 on June 2, leaving it 6.3% below the running peak. Unless a new high occurs, the qualifying close is 9,482.865, requiring another 62.6% fall from the current level over the remaining 4.42 years (FRED).

The setup is stretched but not yet distressed. Vanguard placed broad US equity valuations at effectively the highest percentile in its historical model as of June 30, while the Nasdaq remained concentrated in technology and a few mega-cap firms (Vanguard, Nasdaq Composite fact sheet). At the same time, the July 31 VIX close was 15.99, the July 30 high-yield spread was 2.84 percentage points, and the Chicago Fed financial-conditions index was −0.554 for the week ending July 24, all inconsistent with an acute funding crisis (Cboe, FRED high-yield spread, FRED NFCI).

Evidence

The historical backbone is the full set of completed, local close-to-close Nasdaq bear markets of at least 20% since inception. Local peaks are used here to compare crash severity; the contract itself uses the running all-time high.

Only one of these 14 bear markets crossed 65%. Treating that single fresh-peak event as a constant hazard over the 55.48-year FRED history gives an 8% chance over the remaining 4.42 years. A Jeffreys-prior rare-event calculation gives 11%, showing how much the answer depends on prior assumptions when the effective sample is one (FRED).

Daily-return resampling and standard volatility models give lower estimates because they tend to break the multi-year dot-com decline into short, ordinary return blocks. Longer blocks that preserve regimes move closer to the historical event-rate estimate. This is consistent with research finding that block design materially affects simulated maximum drawdowns and that the largest market declines behave like outliers rather than scaled-up normal fluctuations (Journal of Economic Dynamics and Control, Johansen and Sornette). I combine the event count, regime-preserving models, and broader US-market history into an 8% statistical anchor.

Valuation and concentration move the estimate upward, but not dramatically. Nasdaq’s June 2026 fact sheet put technology at roughly 64% of the Composite and its ten largest securities at about 53% (Nasdaq). Yet long-run bubble research warns against assuming that a boom must reverse: a 2026 study covering US markets from 1792 through 2024 found that booms predict higher volatility, not reliably lower returns, while an earlier study of 42 global markets found a five-year crash rate of 9.8% after a three-year doubling versus 8.4% unconditionally (NBER 2026, NBER 2015).

AI capital spending supplies the clearest current crash mechanism. Nasdaq projects roughly $700 billion of 2026 capital expenditure by the largest technology firms, up 60%–70% from 2025, while also reporting stable-to-improving aggregate leverage because the leaders remain highly profitable (Nasdaq). A filing-based cross-check for Microsoft, Alphabet, Amazon, and Meta—using non-synchronized fiscal years—shows aggregate capex rising from about $237 billion to $409 billion while free cash flow fell from $231 billion to $194 billion (Microsoft, Alphabet, Amazon, Meta).

The financing channel is growing but is not yet systemic. The Bank of England reported that the five major hyperscalers represented more than 15% of year-to-date US investment-grade issuance by early May 2026, while AI issuers accounted for 41% of non-refinancing high-yield issuance; it also judged the outstanding AI-debt stock modest at the start of 2026 (Bank of England). The BIS highlighted circular financing, long leases, special-purpose vehicles, and poor disclosure as possible amplification channels if hyperscalers cut spending (BIS). I read this as a meaningful tail risk, not evidence that a credit cascade has begun.

Current macro data partly offset those risks. June unemployment was 4.2%, and broad credit and financial conditions remained loose (FRED unemployment, FRED high-yield spread, FRED NFCI). The offset is incomplete: the Federal Reserve reported May headline PCE inflation of 4.1% and core inflation of 3.4%, which could constrain aggressive easing during another supply or energy shock (Federal Reserve).

I therefore raise the 8% statistical anchor for extreme valuations, concentration, and the AI financing cycle, then subtract for strong current earnings, cash generation, resilient banks, and the absence of present credit stress. The net result is 9%.

What's non-obvious

The global financial crisis creates a counting trap. The March 9, 2009 close of 1,268.64 was 74.9% below the still-unrecovered March 2000 record of 5,048.62, so the literal running-peak criterion was met again during 2009 (FRED). But this was not a second independent fresh-peak event: the threshold had already been crossed during the dot-com collapse, and the old record remained in force. Today’s record is fresh, so a GFC-sized 55.6% local decline would not be enough.

AI capex cuts both ways. It concentrates index risk and creates a telecom-style overinvestment pathway, but it is also supporting real equipment demand, construction, cloud revenue, and current earnings. The likely failure mode is not that AI has no value. It is that capacity, depreciation, and financing costs rise faster than monetization, producing simultaneous earnings cuts and multiple compression across firms that are both buyers and sellers within the same ecosystem.

Uncertainties

• The directly relevant sample contains one event. Small changes in prior assumptions or event definitions move the estimate by several percentage points (FRED).
• A clean, long-run Nasdaq Composite valuation series was unavailable. Broad-market CAPE and Nasdaq-100 fundamentals are imperfect proxies for the Composite.
• Off-balance-sheet leases, private-credit structures, guarantees, and circular AI financing remain poorly disclosed. The location and size of eventual losses cannot be measured confidently (Bank of England, BIS).
• Stationary return models understate persistent bubble-collapse regimes, while extreme-value and long-block models can overfit the sole dot-com episode. This model uncertainty is larger than the apparent precision of any single numerical estimate.