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

TL;DR

I put the probability at 7.3%. The historical backbone is one qualifying collapse in 55.5 years, while every other running-peak bear market stopped short of 65% (Nasdaq-sourced FRED history). AI capex, concentration, and rich valuations raise the tail, but profitable market leaders and a resilient financial system keep NO at 92.7% (Nasdaq, Federal Reserve). (federalreserve.gov)

Context

The Nasdaq Composite closed at 26,690.62 on August 7, 2026, against a record close of 27,093.90 on June 2, 2026. It was 1.49% below the record. The current qualifying level is 9,482.87, or 35% of the peak (FRED/Nasdaq, updated August 7, 2026 at 10:38 p.m. CDT).

The peak is dynamic. New records would raise the numerical threshold but leave the required drawdown at 65%. A further rally therefore does not make YES safer; it could instead create a larger bubble that later collapses.

Evidence

The clean dataset contains 13,994 non-missing official closes from February 5, 1971 through August 7, 2026. Units are index points with February 5, 1971 equal to 100; frequency is daily close; the series is not seasonally adjusted. The table is the full set of running-all-time-high cycles whose maximum closing drawdown reached 30% (full FRED/Nasdaq series).

Only the dot-com cycle qualified. The 1973–1974 collapse stopped five percentage points short. The global financial crisis produced a 55.63% fall from its October 2007 local high, but that was not a new all-time-high cycle: the 2000 record still stood (FRED history).

Treating the one initiating event in 55.5 years as a Poisson hazard gives:

$$1-\exp\left[-\left(\frac{1}{55.5}\right)\times4.40\right]=7.6\%.$$

Here, 55.5 is the observed index history in years and 4.40 is the remaining forecast horizon. Weak-prior Bayesian versions, with priors centered on one event every 50–80 years, put the historical anchor around 6%–8%. The sample is one event, so the apparent precision is illusory.

Path models produce a lower anchor. Stationary block resampling of the same daily series gives roughly 1%–3%, depending on block length, while defensible heavy-tail or regime models center near 4%–5%. Short blocks break up the multi-year dot-com sequence and understate coherent bubble collapses. Very persistent GARCH specifications do the opposite: volatility can explode unrealistically over four years. I use 4.5% as the path-model estimate rather than taking either extreme literally.

Current conditions push upward. The June 30 Nasdaq fact sheet showed 3,374 securities, but the ten largest accounted for 53.29% and technology for 63.51%. Nvidia and Apple alone represented 19.88%. A common shock to AI demand, chips, cloud spending, or regulation can therefore move the whole index.

Valuation is elevated, but this is not March 2000. The Nasdaq-100, used as a large-cap proxy because it dominates the Composite’s market value, traded at 24.96 times next-12-month earnings on June 2, 2026, versus a 22.94 ten-year average (Nasdaq dashboard). Nasdaq’s historical comparison estimates that the Nasdaq-100 exceeded 100 times trailing earnings around the dot-com peak, with nearly three-quarters of its value either above 60 times earnings or unprofitable (Nasdaq comparison).

AI investment creates the main new tail channel. A May 15, 2026 estimate put the five largest hyperscalers’ 2026 capital spending near $690 billion, equal to 94% of projected operating cash flow; it also identified $822 billion of undiscounted future lease commitments (PIMCO). The BIS documented more than $100 billion of hyperscaler bond issuance in 2025 and growing use of off-balance-sheet data-center vehicles backed by leases, guarantees, private credit, and bank funding lines. The New York Fed described a plausible cascade from weaker hyperscaler demand to data-center developers and their lenders.

The counterweight is the wider financial system. The Federal Reserve’s July 2026 report called it sound and resilient, with bank capital high and business and household debt relative to GDP at its lowest level since the early 2000s. The same report said asset valuations remained above historical norms and hedge-fund leverage remained elevated (Federal Reserve). This is a vulnerable market, not an active systemic crisis. (federalreserve.gov)

I combine three views: a 7.2% Bayesian historical-hazard estimate with 45% weight, a 4.5% path-model estimate with 30% weight, and an 11% current-regime scenario estimate with 25% weight. The result is 7.34%, which I report as 7.3%.

What's non-obvious

The 2009 trough looks like a second qualifying crash if measured against the stale March 2000 record: it was 74.9% below that old peak. But the index had already crossed the 65% threshold during the dot-com collapse. Counting 2009 as another independent event, or counting hundreds of overlapping start dates around 1999–2000, sharply overstates the base rate (FRED history).

The simple “AI equals dot-com” story is also wrong. Today’s leaders are far more profitable and much less expensive than the firms dominating the 2000 index. The risk has moved. It now sits in the scale and duration of capex, leases, data-center vehicles, private credit, and the concentration of index weight in firms exposed to the same investment cycle (Nasdaq, BIS). A 65% fall probably needs both an AI earnings and valuation reset and a second force such as recession, persistent high rates, forced deleveraging, war, or a credit accident.

Uncertainties

My subjective 90% range is 2%–16%. The main gaps are:

• The relevant Nasdaq sample has one qualifying initiating event. Statistical confidence intervals are too wide to settle the question.
• There is no clean, timely Composite-wide forward valuation series. The Nasdaq-100 is an imperfect proxy (Nasdaq dashboard).
• Hyperscaler capex estimates include non-AI spending, and lease and private-credit exposures remain opaque. Public data do not reveal the full loss-bearing chain (New York Fed).
• Return models cannot assign reliable probabilities to novel wars, cyber failures, fiscal crises, market-structure failures, or a technology boom that first becomes much larger before breaking.

Better long-dated option data tied directly to the Composite, consistent daily histories for comparable global indexes, and issuer-level disclosure of data-center leases and guarantees would narrow the range.