Infrastructure Booms, Financial Busts

A recurring shape in American economic history: a country needs a network before anyone can prove it will pay for itself, private capital rushes in to build it, the network gets built and outlives every company involved, and the financiers who took the risk are very often ruined along the way. Railroads, the electric grid, and the fiber-optic backbone of the internet all followed this arc; highways and airports mostly didn't, because government took the financing risk instead of private capital. Editorial synthesis of a well-documented pattern, not a forecast.

The pattern

Infrastructure has a strange economic double life. The country gets a durable, high-value network out of the process almost every time. But because nobody can prove the demand exists until the thing is already built, the money that funds construction is raised on a projection, not a fact — and projections used to raise billions of dollars in a hurry are, reliably, too optimistic. Overbuilding, price collapse and a wave of bankruptcies follow the boom as surely as the boom follows the initial need. The asset keeps working after the crash; the capital structure sitting on top of it usually does not survive to see that.

Case 1 — US railroads (1830s–1900)

America's rail network was financed almost entirely by private capital — state charters, stock sold to British and American investors, and, from the 1850s, direct federal land grants — competing to connect the same cities and territories several times over because no single company could be sure it would win the route. The Panic of 1873 began with the collapse of Jay Cooke & Co., overextended on Northern Pacific Railway bonds, and dragged the country through roughly five years of depression as dozens of railroads defaulted. Two decades later the Panic of 1893 put an estimated quarter of US rail mileage into receivership at once, including major trunk lines like the Philadelphia & Reading and the Erie. Both panics wiped out bondholders and shareholders on a huge scale. The track itself mostly stayed in the ground, was reorganized under new owners, and kept running.

Case 0 — toll turnpikes came first (1792–1830s)

Roads ran the same experiment a generation earlier, on a smaller scale. Starting with the Philadelphia and Lancaster Turnpike in 1792, American states chartered thousands of private turnpike corporations to build and toll roads that individual towns couldn't afford alone. Economic historians who have gone back through the surviving company records (notably Daniel Klein and John Majewski) find that most turnpike companies never paid their shareholders a meaningful dividend — the tolls rarely covered the cost of construction and upkeep — even though the roads themselves clearly paid off for the towns and farmers who used them for free-riding commerce. It is the same private-loss, social-gain split the later, larger booms would repeat at far greater scale.

Case 2 — the electric grid's holding-company pyramids (1900s–1930s)

Electrification needed enormous upfront capital for generating plants and distribution wires serving customers who, in many towns, didn't yet own an appliance. The financing solution that won was the utility holding company — a corporate structure pyramiding one holding company on top of another, each layered with its own debt and preferred stock, controlling an operating utility several levels down with a sliver of actual equity. Samuel Insull built the largest of these empires, ultimately controlling a network of utilities serving millions of customers across the Midwest through a tower of holding companies. When the 1929 crash hit the stock financing his structure depended on, the whole pyramid unwound; Insull's empire collapsed in 1932 in one of the largest corporate failures in American history to that point, wiping out shareholders up and down the chain. The lights, generating stations, and wires kept working throughout — Congress responded not by tearing down the grid but by passing the Public Utility Holding Company Act of 1935 to outlaw the pyramid structure that had financed it.

Case 3 — the dot-com fiber glut (1996–2002)

Deregulation and internet-traffic projections in the late 1990s set off a race to lay long-haul fiber-optic cable, financed largely by newly created carriers selling high-yield bonds against forecasts of exploding bandwidth demand. WorldCom and Global Crossing were the two most prominent builders; WorldCom's 2002 bankruptcy, driven by accounting fraud on top of the overbuilt network, was at the time the largest in US history, and Global Crossing's bankruptcy the same year erased more than ten billion dollars of value built on roughly 100,000 miles of cable. A widely cited estimate from that period put the share of newly laid fiber actually in use (“lit”) at only a small fraction of installed capacity — the rest sat dark for years. That dark fiber is a large part of why video streaming, cloud computing, and — later — AI training traffic had cheap, abundant bandwidth to grow into once demand caught up with the capacity a bust-era buildout had already paid for.

Case 4 — roads and airports built the other way

By the mid-20th century, the US ran the same problem — a needed network too big and too uncertain for any single private balance sheet — through a different financing structure, and got a visibly different outcome. The Interstate Highway System, authorized by the Federal-Aid Highway Act of 1956, was financed roughly 90% by a federal Highway Trust Fund built from fuel taxes, with states covering the rest; there was no private bond wave and no comparable investor wipeout, because the risk of the projection being wrong sat with the federal government from the outset. Commercial airports followed a similar public path — the large majority are owned by cities, counties, or public authorities and financed through municipal revenue bonds backed by landing fees and concessions, with federal grants (formalized in the Airport and Airway Development Act of 1970 and its successors) covering much of the rest. Neither network produced a Panic-of-1893 or an Insull-scale collapse, because neither asked private financiers to underwrite the initial bet on demand.

The comparison

The hypothesis, stated plainly

• Networks need to exist before demand can prove itself. No amount of careful modelling substitutes for actually building the railroad, the grid, or the cable, so the capital that funds construction is always betting on a projection rather than a fact.
• Private capital racing to fill that gap reliably overbuilds. Competing companies chasing the same uncertain demand, each individually rational, collectively lay far more track, wire, or cable than the resulting market can support at a price that repays construction cost.
• The bust falls on the financiers, not the asset. Rail track, generating plant, and fiber conduit are durable and keep functioning through reorganization; what doesn't survive is the specific capital structure — bonds, holding-company equity, junk debt — that financed the bet.
• Government-led financing skips the private-ruin phase — it doesn't skip the risk. Highways and airports show that putting the state's balance sheet behind the projection from day one avoids the Insull- or WorldCom-style private wipeout. It does not mean the projection was safer; it means taxpayers, not bondholders, absorb a wrong one.

Critique

• Survivorship bias in which busts get remembered. The railroads, utilities, and telecoms that are cited here are the spectacular failures; plenty of 19th- and 20th-century infrastructure ventures financed privately and returned their investors a perfectly ordinary profit, and don't make it into the popular history of the boom.
• Public financing has its own failure mode. Government-backed infrastructure isn't immune to bad projections — it just converts a bankruptcy story into a cost-overrun or under-utilization story paid for through taxes and tolls instead of a stock-price collapse, which is a different kind of loss, not the absence of one.
• The four cases aren't on identical footing. Railroads and utilities financed genuinely novel, uncertain technology; by the 1950s the case for highways and by the 1970s the case for airport capacity was far better established, which on its own would predict smaller forecasting errors regardless of who financed them.

What it means for the AI data-centre race

The current buildout of AI data centres follows the private-capital template far more closely than the highway or airport one. As this site's capex supercycles piece lays out, the hyperscalers are financing an investment wave in the historical band of the railway manias, but doing it faster and with far more concentrated capital — increasingly not just out of their own cash flow, but through debt issuance, GPU-collateralized loans, and off-balance-sheet joint ventures with private-credit funds designed to keep the exposure a step removed from any single company's balance sheet. That layering of debt on top of debt to finance a projection nobody can yet prove is precisely the structure that undid Insull's utility pyramid in 1932 and the telecom bond issuers in 2002 — and, like both of those, it is happening with essentially no government backstop of the kind that spared highways and airports their own version of this story. The sobering read of the historical pattern is not that the compute being built now will turn out to be wasted — the railroads, the grid, and the fiber all say the opposite, that the capacity gets used eventually and becomes the backbone of the next several decades of growth. It is that the specific firms, funds, and lenders financing this particular boom should not assume that being on the right side of history is the same thing as being solvent at the end of it.