Investing Classics: The Capital CycleWhere the Money Went, and What It Does to Returns

Edward Chancellor, ed., Capital Returns (2015), Introduction. The book collects Marathon Asset Management's investment commentary from 2002–2015.

The framework's power comes from an asymmetry: demand is hardest to forecast, supply is comparatively observable. More importantly, capital formation lags—two to five years typically separate "these returns look attractive" from "the capacity is running." New supply therefore tends to arrive exactly as the thesis behind the hottest demand starts to wobble. That lag is the source of the excess return.

Dry bulk shipping. The Baltic Dry Index hit its all-time peak of 11,793 on May 20, 2008, and fell to 663 by that December—-94% in seven months. The lesson is not the crash but what followed: ships ordered in 2006–2007 only launched through 2010–2012, so supply arrived long after demand had collapsed, holding the industry in low returns for close to a decade. Owners placed those orders while looking at record freight rates.

Three failure modes. When demand suffers a structural break, this is not a cycle but an ending—capacity in film and print media shrank drastically without returns recovering, because the end of clearing was zero. In licence-protected industries capital cannot enter or exit freely, and the cycle is severed by administrative force. And the time scale: the framework works over three to seven years.

• Am I forecasting demand, or observing supply? Is there public data for the latter?
• How much capacity has this industry added over three years? How much is still under construction or in transit?
• Holding demand flat, what happens to returns once announced capacity lands?
• At the end of this industry's clearing, is there a better structure—or zero demand?

Howard Marks, The Most Important Thing (2011), Chapter 8, "Being Attentive to Cycles."

Both are theories of cycles; they measure different things. Sentiment is a fast variable—it can flip from greed to panic in weeks, which makes it useful for timing but silent on long-run earning power. The capital cycle is a slow variable—once capacity is poured it depreciates for a decade, which makes it nearly useless for timing but decisive for where an industry's ROIC settles over five years. The two can diverge: in 2001 telecom equities had already halved while fiber was still being laid. The ideal entry is when both point the same way—the market is disgusted and the capital has already left.

DRAM. Elpida went bankrupt in 2012 after an oversupply-driven price collapse; Micron completed its acquisition in 2013, and the industry consolidated from a crowded field to three players. The consensus then was "forever cyclical, never profitable"—sentiment was terrible, but the supply side had already cleared. By fiscal 2018 Micron posted record results, with non-GAAP net income of roughly $14.7 billion. The symmetry then delivered: record returns called capex back, and fiscal 2019 fourth-quarter revenue fell 42% year over year with operating income down roughly 85%.

In the March 2020 pandemic crash the capital cycle gave no signal at all—supply cannot change in a few weeks—yet the opportunity was a once-in-a-decade one. When the fast variable dominates, the slow one is mute. Conversely, through 2021–2022 it genuinely was sounding an alarm, and that alarm rang for eighteen months before anything happened. It gives direction, not timing; using it as a timing tool is the most expensive misuse.

• Am I drawn by a cheap price (sentiment), or by an improving structure (supply)?
• Do the two signals agree or diverge? If they diverge, which do I back, and why?
• Is my holding period long enough for a supply-side change to play out—at least three years?

Buffett's 1985 letter to shareholders, reviewing the closure of Berkshire's textile operation.

① Capex / depreciation. Sustained above 2 for years means capacity is compounding and returns will almost certainly be diluted; persistently below 1 means the industry is shrinking itself and survivors are accumulating pricing power. Note that the depreciation life itself can be adjusted—and the direction of that adjustment is a signal in its own right.

② The financing window. The hallmark of a cycle top is not a high share price but how easy financing has become—money chasing projects rather than projects hunting money. This forecasts the ending earlier than any valuation metric.

③ Management's incentives. Once the scorecard shifts from ROIC toward market share, installed base, and size rankings, capital discipline has already broken. The parade metaphor describes the mechanism precisely: every individual expansion clears its own hurdle rate, yet collectively they cancel each other's returns. This is not stupidity—it is the equilibrium of a prisoner's dilemma.

Berkshire's textile business. Closing it in 1985, Buffett admitted that each round of equipment spending had passed its own arithmetic—but after every round all players had spent the same money, the lower costs immediately became the industry's new price baseline, returns returned to where they started, and everyone simply had more chips on the table. Not investing meant exiting sooner; investing meant a collective futility.

Heavy capex is not itself a sin. When expansion happens in an industry with real entry barriers, the leader's heavy spending actually widens the moat: leading-edge fabs run capex far above depreciation for years, yet precisely because the threshold is so high, the number of followers keeps falling. The distinction is not how much was spent but whether that spending makes entry harder or easier for newcomers. Watch the accounting boundary too: in asset-light models real capacity expansion may not show up in owned capex at all.

• What has this industry's capex/depreciation been over three years? Rising or falling?
• Over the past 24 months, has financing in this industry become easier or harder?
• Does my company's expansion raise or lower the barrier for new entrants?
• In management's scorecard, how much weight sits on ROIC versus size metrics?

Buffett's 2007 letter to shareholders, on the great, the good, and the gruesome.

The telecom investors of the late 1990s were right about demand—internet traffic did go on to grow by orders of magnitude. They were wrong about supply: everyone believed the same correct thing at the same time, and so laid fiber at the same time. Being right does not constitute a return; the return comes from being right and others not having done it yet. Once a correct call becomes consensus, it converts automatically into capital inflows—and therefore automatically into returns being destroyed.

Fiber. Between 1996 and 2001 US telecom companies issued more than $500 billion in new bonds; on OECD figures, telecom infrastructure investment approached $230 billion in 2000, while OECD telecom revenue grew only about 7.2% a year over the period, slowing to 1.6% in 2001. Only a single-digit percentage of the fiber laid was ever lit for years afterward, and roughly 85% was still dark at the end of 2005. Global telecom equities lost more than $2 trillion in market value between 2000 and 2002; Global Crossing filed for bankruptcy on January 28, 2002. And the internet itself succeeded completely—the returns went to those who later picked up that fiber at liquidation prices.

An observation on the present (not a forecast): combined 2025 capex at Microsoft, Alphabet, Amazon, and Meta was roughly $410 billion, with 2026 guidance rising toward the $700 billion range. Alongside it came a reversal in depreciation policy: Amazon extended server useful life from five years to six effective January 2024, then moved it back to five a year later (effective January 2025, citing the faster pace of technology development in AI), disclosing that the change would reduce that year's operating income by about $700 million. The economic life of the assets is being cut by the owners of the assets themselves.

The fiber analogy is imprecise in two ways. First, the main builders back then were highly leveraged pure infrastructure players, while today's principal spenders hold enormous operating cash flow; destroyed returns need not mean insolvency, and the pain more likely takes the form of years of low ROIC. Second, fiber was a one-time capital good while compute depreciates fast, so supply will not sit in a glut for a decade—but that also means the spending must be repeated every year. Only the method transfers: count the supply, compute the depreciation, watch how easy the financing is.

• Is my view of this technology also held by everyone else? Then where does my excess return come from?
• Is this wave of spending funded by operating cash flow, debt, or equity issuance?
• Does the accounting life of these assets match the economic life I believe in? Who is adjusting it?
• If demand arrives but returns do not—the most common outcome historically—what happens to my position?