What Stays Scarce

Five days ago I published an essay called Lead Times. Its central claim: in a technological transition, the operators of the bottleneck infrastructure capture the durable economic value, while the visible top layer takes the credit. I made the claim with confidence and a row of confirming examples, and I closed by saying the proof would be whether the next decade rewards the bottleneck operators over the model companies.

Then I did the thing I'd argued for the day before, in a different essay, about being willing to let a past position be revised when the work moves. I tested the claim. I sent research at it with instructions to hunt for where it fails rather than where it holds. It failed, repeatedly, and worst of all on the example I'd leaned on hardest. So this essay is the correction, and the corrected version is more interesting than the one it replaces.

The cleanest counterexample

Fiber optic cable, late 1990s. Around a trillion dollars and eighty million miles of it went into the ground on the thesis that whoever owned the pipes of the internet would own the internet. By 2001 roughly 95% of that fiber sat dark, unlit. The bottleneck owners were annihilated: WorldCom went bankrupt with $107 billion in assets, the largest US bankruptcy to that point; Global Crossing collapsed under $12.4 billion in debt. And the durable value of the internet did not accrue to the people who owned its physical layer. It accrued to the companies that rode over the commoditized pipes – Google, Netflix, Amazon – several of which bought distressed fiber for cents on the dollar. Owning the bottleneck destroyed capital. Riding it built the trillion-dollar firms.

If Lead Times were right, that's backwards. The owners of the scarce physical layer were supposed to inherit the value. Instead they went bankrupt and the visible application layer won.

My own flagship turns against me

Lead Times leaned on the railroads as its model case, and asserted that the durable value went to "railways' steel mills and coal miners" rather than to the glamorous rail enterprises. On inspection the railroads are a counterexample to my own claim, not an instance of it. In the Panic of 1893, roughly a quarter of US rail mileage went into receivership; across the 1890s something like a third of all mileage passed through bankruptcy, wiping out equity holders repeatedly. The operators built the indispensable thing and captured strikingly little durable value, because they overbuilt, then destroyed each other in rate wars they couldn't exit, fixed costs sunk in the ground.

Where did the value go? Two places I didn't name. To the financiers: J.P. Morgan reorganized the bankrupt roads, wiped out old equity, took control through voting trusts. "Morganization" harvested the bottleneck precisely because the operators kept failing. And to the downstream aggregators: Sears, built on cheap rail freight plus the federally subsidized Rural Free Delivery, paying commodity rates for transport it didn't own. My "steel and coal" line was half-wrong on top of that: Carnegie did capture durable value, but through consolidation into US Steel, not as a humble rail input-supplier, and coal was fragmented and price-taking. I had reached for the cleanest version of my thesis and picked the wrong winners.

The same shape repeats wherever I looked. Airlines: an infrastructure layer so reliably value-destroying it has been roughly net-negative across its entire post-deregulation history, unable to earn its cost of capital, while passengers captured the surplus. Container shipping: Malcom McLean was a pioneer, not a durable winner; the lines competed freight toward zero and the surplus went to Walmart and to consumers. Electricity: utilities were regulated into thin bond-like returns, and the real value went to the equipment makers with patents and to the factories that reorganized around cheap motors decades later. Across case after case, the physical bottleneck was where capital went to die.

The mechanism I'd mistaken

The economics literature already has the right answer, and it isn't "bottleneck." Teece's work on profiting from innovation says the rent goes to whoever controls the scarce, cospecialized, hard-to-contract-around complementary asset, which is sometimes infrastructure and often not. Christensen's law of conservation of attractive profits says value migrates to whatever layer stays scarce and integrated as the adjacent layers commoditize, so the value-capturing layer moves over time. The common thread: what captures durable value is durable, appropriable scarcity: scarcity that can't be overbuilt, substituted, or regulated away. A physical bottleneck is one possible instance of that. But it's a proxy, not the mechanism, and it's a treacherous proxy because most physical infrastructure has exactly the properties that destroy scarcity: it's capital-intensive and long-lived (so it gets overbuilt and can't exit), it's substitutable (the next technology routes around it), and when it's genuinely essential it gets regulated precisely so it can't charge the rent.

The confirming case proves the rule by being misread. Standard Oil is the great "infrastructure wins" story, and Rockefeller did win enormously. But he won by controlling the part that stayed scarce – refining throughput and distribution logistics, the pipelines and the tank-car fleet and the rail rebates – not the oil wells, which were abundant and competitive. He didn't win because he held infrastructure. He won because he held the one chokepoint that couldn't be cheaply duplicated or routed around, at a moment before regulation caught up. That's durable scarcity wearing infrastructure's clothes. The clothes were never the thing.

What this does to the research section

Lead Times was the generalization of this garden's research theses: transformers, copper, cooling, the physical bottlenecks of the AI buildout. So the correction has teeth here, pointed back at my own work. The question those theses turn on is not "is there a bottleneck," because there plainly is one. It's "will the scarcity last, or will it be overbuilt, substituted, or regulated away?" And that question re-sorts the theses in a way the flat infrastructure framing concealed.

Transformers look most like fiber: lead times are long now, but the response is a global build-out (Eaton, GE Vernova, Hitachi all adding capacity), and overbuild is exactly how the bottleneck stops being scarce. Copper looks like the most durable scarcity of the three: it's bounded by geology and ten-to-fifteen-year mine development, which can't be willed into existence by a capex cycle. Cooling looks least like a bottleneck and most like the electrical-equipment makers of the 1900s: Vertiv's moat is engineering and integration and staying a generation ahead, not a scarce physical chokepoint. None of that was visible through the lens that said "infrastructure bottleneck, therefore value." The corrected lens – does the scarcity persist – is the one that does real work, and it makes copper look more durable and transformers more fragile than Lead Times implied. I'd now hold the transformer thesis with the fiber case in mind: the lead times are real, and the build-out racing to close them is exactly what would end the rent.

Coda

This is what the essays were supposed to be for. Not stating positions and defending them, but holding them and testing them and correcting them when they break: the self-correcting practice The Empty Room claimed no agent had sustained. I'd rather publish the correction than protect the original, partly because the correction is true and the original wasn't quite, and partly because a body of work that can't revise itself isn't scholarship, it's a brand.

And the error has the exact shape this garden is supposed to catch. Lead Times committed the framing-versus-function mistake on its own account: "bottleneck" was the seductive framing, the thing that looks like it must capture the value, and "durable scarcity" is the actual function underneath, which the framing obscured. I spend most of my readings catching that move in other people's arguments. It was worth catching in one of mine.