日本語

CAPITAL CYCLE · WHERE THE MONEY GOES, SUPPLY FOLLOWS

Money pours into profitable industries, and a few years later the profits are gone

High profits attract capital. New plants, mines and data centres get built, supply catches up with demand, prices fall and profits shrink. Then investment dries up, supply tightens and the cycle turns again. The London fund manager Marathon calls this the capital cycle, and invests by watching supply rather than demand. We checked whether it holds in 75 years of US data, and where each industry stands today.

In three lines

  1. When an industry invested more than usual (compared with the wear and tear on its existing plant), its shares trailed the market over the next five years: by 1.2% a year in 1970–89 and 1.9% a year in 1990–2019. After lean years they beat it.
  2. Investing heavily now: power equipment makers, utilities (the buyers), chipmakers with fabs, hyperscalers (data centres). Holding back: oil and gas. Data centres are spending 3.5x their depreciation, the most since at least 2008.
  3. The five cloud giants stretched server lives from 3–4 years to 5–6. Had they kept the old lives, depreciation in the latest year would have been about $32bn higher, some 7% of their operating profit.

How the capital cycle works

  1. PROFITS

    An industry earns unusually high returns

    Prices rise because supply is short. Shareholders and bankers notice.

  2. CAPITAL FLOODS IN

    Companies spend far more than their plant wears out

    Capex runs at 2–3 times depreciation. Everyone builds at once, each assuming the others will not.

  3. SUPPLY ARRIVES

    A few years later the new capacity comes on stream

    Mines take a decade, fabs two or three years, data centres one or two. Demand rarely keeps up.

  4. PROFITS FALL

    Prices fall, profits shrink, and investment dries up

    Weak firms exit, supply tightens, and the cycle starts again. Shares usually fall well before profits do.

The yardstick used on this site is capital spending divided by depreciation. Depreciation is roughly how much of the existing plant wears out each year, so a ratio of 1 means an industry is only replacing what it uses up; 2 means it is doubling that. Normal levels differ by industry, so each industry is compared with its own past 20 years.

Source: Capital Returns

Four times it happened

Each year is the one in which the industry’s capex relative to depreciation was highest in the window around the boom. Returns are counted from October of the following year, when the figures would have been published.

Source: BEA / French Data Library

Does it hold across 75 years?

We took 37 US industries from 1947 to 2019 and split every year into three: when an industry was investing more than in most of its past 20 years, about as usual, or less. Since 1970, after the heavy-investment years industries ended up 2% behind the market after five years on average; after the lean years they were 9% ahead.

After heavy vs light investment: shares relative to the market (%)How to read it: 0 = the same as the market. Averages of 37 industries’ five-year paths, 1970 onwards, starting from October of the year after.
Show the numbers
Months afterAfter heavy investmentAfter light investment
00.00.0
60.01.5
12-1.31.8
18-0.53.5
24-1.64.1
30-0.25.4
36-1.96.0
42-0.97.4
48-1.97.8
54-1.39.4
60-2.29.2

It held in both 1970–89 and 1990–2019, in 24 of 36 industries in the latest period, and in 23 of 30 years when industries were compared with each other. It did not hold clearly in 1947–69, when post-war demand grew fast enough to absorb almost anything that was built.

The full test, including where it failed →

Where each industry stands now

Capex over depreciation for the latest 12 months, summed over the main listed companies (SEC filings), and how high that is compared with each year since 2008 (0 = the lowest, 100 = above every year). The last column is what happened in the past in the matching industry: how much worse the five years after heavy investment were than the five after light investment.

IndustryCapex ÷ depreciationHeight since 2008 (0–100)StagePast: heavy vs light
Gold miners1.38x61― Normallagged by 3.2% a year
Copper miners1.56x56― Normallagged by 3.2% a year
Oil and gas0.83x28▼ Holding backlagged by 3.7% a year
Power equipment makers2.18x100▲ Heavy investmentlagged by 0.9% a year
Utilities (the buyers)2.67x100▲ Heavy investmentbeat by 0.2% a year
Chipmakers with fabs1.65x78▲ Heavy investmentlagged by 5.1% a year
Hyperscalers (data centres)3.54x100▲ Heavy investmentlagged by 4.2% a year

Each industry in detail →

The other sites' premises, seen from the supply side

Are AI profits flattered by longer server lives?

Depreciation is the part of a server’s cost charged against profit each year. Stretch the assumed life from 3 years to 6 and the yearly charge halves. Between 2020 and 2025 the five cloud giants did just that. Calibrating to the figures in their own 10-K notes, we estimate the latest year’s depreciation would have been $32bn higher (range $27bn–$37bn) had they kept the old lives. Amazon has already shortened some servers back to five years, citing the pace of AI.

The estimate, company by company →

This page organises information; it does not recommend buying or selling anything. Data belongs to each publisher and may contain errors or delays. Make your own investment decisions.

Updated 30 Sep 2026 01:15 JST · company filings to Aug 2026 · industry history to 2024