The Stock Market Peaks Long Before Capex Slows: Five Historical Episodes

The idea in one line

Equity markets peak before capital expenditure (capex) stops growing, and not by a little. That is the pattern across the great technological revolutions, and the argument examined here is that the AI investment cycle should be expected to follow it. Read this as a framework for thinking about timing, not as a call on today's market.

The trap: waiting for capex to slow

Start with how many investors are thinking. AI is a revolution that is changing the world and will keep changing it. It is also going to cost a great deal of money for more than one or two companies, and more people are becoming aware that the scale of what is being built may be too much.

Yet nobody wants to leave early. The revolution is good for the economy, it transforms things, and there may still be a stretch of the ride left. So investors set themselves a marker: when capex slows down, I will get out.

This is the problem. The plan sounds sensible because it ties the exit to something observable. But it does not work that way, and the reason is timing. The market tends to peak first and capex keeps rising afterwards, so a signal that arrives only when capex slows arrives after the top.

What the history shows

The evidence comes from a chart that compares the peak of the stock market with the peak of capex in several big technological episodes. In each case, capex went on rising after the market had topped out:

• Railways: after the market peaked, capex kept rising for roughly two years.
• Utilities, around 1929 to 1930: capex kept rising for about a year after the market top.
• Japan's bubble, 1989 to 1991: capex peaked 18 months after the market did.
• The dotcom bubble of 2000: capex kept increasing for about eight months after the market peak.
• A fifth episode dated 2014: capex kept going for about six months.

The lags run from six months to about two years. The lag is never zero and never negative in any of these cases. At no point did capex turn first and give investors the warning they were hoping for.

How close is the dotcom parallel?

Of the five, the dotcom bubble is singled out as perhaps the most similar to the present, with a caveat: it is not equal by any means, but it has a certain resemblance. That is a modest claim and it should stay modest. The similarity is a reason to take the dotcom lag of around eight months seriously, not proof that the same timing will repeat. The evidence does not explain why the lag differs from one episode to the next, so it offers no way to predict which one the AI cycle would resemble.

Why capex gives no warning

The central point is short: capex will not give notice. It is not an alarm bell. The sequence investors should picture goes like this.

• The market reaches its peak.
• Nobody sees it at the time, because a peak is only recognisable afterwards.
• Capex carries on growing for quite some time: many months, perhaps a year, perhaps a year and a half or even two.
• Meanwhile investors sit near the top, reassured by the numbers.

That last step is the dangerous one. While capex is still rising, it feels like confirmation. The reasoning runs: capex is still growing, so this will keep going up. That is the mental error. Rising capex after the peak is not evidence that the market is still healthy. In every episode in the chart, capex kept growing for a while after the market had already turned.

So the marker does two kinds of damage. It delays the exit, because the signal comes late. And it gives false comfort in the meantime, because the very variable you are watching is still pointing up. The investor holds on, convinced the rule is protecting them, while the rule is in fact keeping them in.

Why investors fall for it anyway

The psychology is plain. Everyone is aware that things may be getting too far. But nobody wants to give up the remaining stretch of a revolution that is changing the world and is good for the economy. Choosing capex as a marker lets an investor feel prudent and stay fully invested at the same time. It postpones the decision to a future moment, and that future moment is defined by a variable that moves later than the market. The same tension is examined in avoiding FOMO investing in the 2026 tech rally.

The argument does not claim that the AI revolution is fake or that it will not matter. It accepts that it is changing the world and will go on doing so. The claim is narrower and about timing: a real revolution and a market that has run ahead of it can coexist, and in past episodes the market peak came before the investment peak.

What the evidence does not say

It is worth being clear about the limits. The evidence does not give the reasons why markets topped before capex in each of these cases. It does not give a date or level for a possible peak in the current cycle, and it does not offer an alternative exit signal. The honest conclusion is the one the history supports: the capex marker is unreliable, because in all the cases shown it lagged the market by between six months and about two years.

The takeaway

If you plan your exit around the day capex slows, history says you will probably be leaving after the top, possibly long after. The practical lesson is to distrust any exit rule that depends on a variable known to lag. The lag in the five episodes ran from about six months to about two years, and during all that time the data looked reassuring. Where the AI trade's leadership has already shifted inside the market is covered in the AI trade has split.

ImGeld publishes research, not investment advice.