In December 1996, the most powerful financial official on Earth publicly suggested the stock market had lost touch with reality. Alan Greenspan's "irrational exuberance" speech is now shorthand for calling the dot-com bubble — and he was, eventually, proven right. Here is what the shorthand leaves out: the Nasdaq roughly quadrupled after his warning before it peaked in March 2000. An investor who exited on the Fed chairman's signal spent 39 months watching the market run away from them — and even at the very bottom of the 2002 wreckage, the S&P 500 never fell back to the level of the speech.
Being right about the bubble was not enough. It has never been enough.
That is the uncomfortable baseline for today's dominant market question. A handful of AI-linked megacaps now drive an outsized share of index returns, valuations sit well above historical averages, and every week produces a credible essay arguing that this is 1999 again. Some of those essays may age brilliantly. The historical record says that acting on them is a different problem entirely — one with a much worse payoff profile than the essays imply.
What the Dot-Com Tape Actually Shows
The 2000 crash is the strongest argument the bubble-callers own, so examine it honestly — from both directions.
The damage was real and specific:
- Nasdaq, March 2000 to October 2002: a 78% peak-to-trough decline; the index needed 15 years to reclaim its high.
- Cisco: the era's dominant infrastructure company, profitable throughout, spent more than two decades below its 2000 price.
- S&P 500, 2000–2009: a negative total return across ten full years — the "lost decade."
But the timing record from the same era is just as specific, and it points the other way:
- Warning-to-peak gap: Greenspan's 1996 call preceded the top by over three years and a ~4x Nasdaq run. Prominent bear calls in 1997, 1998, and 1999 fared the same.
- The exit that "worked" still required a re-entry: an investor who sold in early 1999 and — paralyzed by the crash they correctly predicted — re-entered in 2004 underperformed the investor who simply held a diversified portfolio through the entire event.
- The diversified experience: a global 60/40 portfolio, rebalanced annually, endured the 2000–2002 decline at survivable depth and finished the "lost decade" comfortably positive, because international stocks, value stocks, and bonds all worked while US large-cap growth did not.
The dot-com crash did not punish investors who owned technology inside a diversified plan. It punished concentration on the way down — and it punished market timing on the way up, both before the peak and after the bottom.
One more dot-com fact deserves permanent residence in your memory. Amazon fell 94% in the crash. The thesis "the internet will transform commerce" was completely correct, the company survived, and an investor still needed to sit through a near-total drawdown to collect. Whether AI is "real" and whether current prices are safe are separate questions. They were separate in 1999 too.
The Trap: Why Bubble Warnings Feel Actionable
The pull toward "doing something" about an expensive market is not an information problem. It is a wiring problem, and it has two components.
First, the regret asymmetry behind action bias: losing money in a crash you stayed invested for feels like negligence, while missing gains from cash feels like prudence. The feelings are backwards relative to the math. Since 1996, US markets have delivered several multi-year stretches of elevated valuations and near-continuous bubble commentary — an investor who went to cash at the first credible warning of each cycle missed the majority of two decades of compounding. The crash they avoided was smaller than the advance they forfeited, in every cycle on record.
Second, the re-entry problem. Selling requires only conviction; getting back in requires a signal that never arrives. If prices fall after your exit, further falls feel imminent. If prices rise, the market looks even more overvalued than when you left. Every price path strengthens the case for waiting. This is how investors who correctly sold in early 2000 ended up watching the 2003 recovery — and 2009's, and 2020's — from the sidelines. The Schwab timing study quantified the stakes across 20 years of data: the gap between perfect annual timing and simply investing immediately was about 12% cumulative, while the gap between investing immediately and waiting in cash was roughly 200%.
Bubble calls are free. Both transactions they demand — the exit and the re-entry — are among the most expensive trades in market history.
The Strategic Filter: Replace an Unanswerable Question With an Answerable One
"Is AI a bubble?" is unanswerable in advance — in 2000, the answer was yes; in 1996, acting on yes cost a fortune; in 2010, 2015, and 2020, credible yeses were wrong. Discard the question. Replace it with three that have objective answers.
What is my actual concentration? A cap-weighted US index fund currently assigns roughly a third of its weight to a handful of AI-adjacent megacaps. That is not a flaw — cap-weighting is how index investors captured those companies' entire ascent — but it is an exposure you should hold knowingly. Investors who added tech or AI-themed funds on top of an index core are frequently running 45–55% effective exposure to one narrative without having decided to.
Does my allocation survive the bear case? Not "will the crash happen" — "if 2000–2002 happens to the AI cohort, does my portfolio's drawdown stay inside the tolerance I set in my policy?" International equities, bonds, and disciplined weights answered this in 2000. They are the same answer now.
Is anything trimming the winners automatically? A rebalancing rule harvests the run-up mechanically — selling a slice of what soared, buying what lagged — without requiring a bubble verdict from anyone. It is the only "sell high" mechanism in finance with no forecast inside it. If the AI cohort keeps compounding for a decade, you participate the whole way. If it unwinds, you spent years quietly banking gains into the rest of the portfolio.
An investor with honest answers to all three questions has extracted everything useful from the bubble debate — without placing a timing bet in either direction.
Testing the Whole Argument on Real Data
Every claim above is checkable against market history, which is the appropriate standard for an article asking you to distrust dramatic narratives.
The Backtester covers the relevant experiments directly. Run a US-concentrated portfolio against a globally diversified one with the date range set across 2000–2009, then again across the full 30-year window — the reversal between the two views is the entire concentration argument in two charts. Then price the timing trade you may be tempted by: model an exit at a past peak with re-entry 12 or 24 months later, and compare it against holding. The re-entry drag tends to surprise people more than any crash statistic.
The Asset Allocation sandbox answers the concentration question for your own holdings: the look-through view shows your true combined weight in the megacap cohort across every fund you own — the number most investors have never actually computed — and the historical stress overlays show what your current mix would have experienced in the 2008 and 2020 drawdowns. If the output sits inside your written tolerance, the AI debate becomes something you can read for entertainment. If it doesn't, you've found a portfolio decision that is worth making — and it has nothing to do with predicting a top.
You cannot know whether this is 1996, 1999, or neither. You can know your concentration, your worst-case drawdown, and your rebalancing rule — and those three numbers are the only part of the bubble question that was ever yours to control.