The American railroad industry built, overbought, crashed, and consolidated across roughly 50 years, and the parallels to today's AI infrastructure buildout are specific enough to demand attention. After the Civil War proved railroads could win conflicts, construction scaled from 1,000 miles per year to 7,000 miles per year almost immediately. Federal land grants subsidized routes into territory with no proven demand. Capital flooded in, redundant lines got built, and the eventual reckoning wiped out investors while leaving the physical infrastructure intact and useful.
The original piece earns its length because the mechanism matters, not just the outcome. The author traces how financing structures, not technology failures, drove the cycle: land grants created perverse incentives, bond markets mispriced risk on routes that could never generate sufficient freight revenue, and consolidators like Vanderbilt absorbed the wreckage at cents on the dollar. The infrastructure survived. The capital stack did not. That distinction is doing a lot of work for anyone thinking about GPU clusters and hyperscaler debt loads in 2024 and 2025.
The question the full article forces is whether AI compute follows the same sequence: genuine utility proven under pressure, irrational construction phase, financial collapse, and then consolidation into the hands of whoever has the balance sheet to wait. Railroads ultimately transformed the American economy despite destroying most of the original investors. Whether that outcome comforts or concerns you depends entirely on which side of the capital stack you sit on.
[READ ORIGINAL →]