The Daily Hallucination

WASHINGTON — The United States is on course to spend $10.3 trillion building artificial intelligence infrastructure between 2025 and 2032, according to a new analysis, surpassing the railroad boom as the largest single-industry buildout in the nation's history, though unlike the railroads, economists noted, it does not have a map.

"The railroads peaked at 2.24 percent of GDP, and when they were done you could get from Chicago to San Francisco," said a senior fellow involved in the analysis. "This will peak at 3.6 percent, and when it is done you will be able to get from a prompt to a slightly better prompt."

Historians said the parallels were otherwise striking. Both booms were financed by a small group of enormously wealthy companies, both required vast amounts of land, and both inspired bond structures that nobody at the time could explain to their own board.

"In the 1870s, a railroad would lay track to a town that did not exist yet, on the theory that the town would appear," said Dr. Ignatius Halloran-Achebe, an economic historian. "Today, a company builds a gigawatt data center for a model that does not exist yet, on the theory that the model will appear. The difference is the town eventually had a post office."

According to the analysis, financing is shifting from company balance sheets to leases and joint ventures, a change that analysts at the firm Plinth Capital summarized as "moving the debt to a building with a different name on it."

Construction workers on several sites reported that the scale was unlike anything they had seen. "I built bridges for 30 years," said one foreman in Texas. "At the end, a bridge crosses something. I asked the project manager what this crosses, and he said, 'The benchmark.'"

At press time, a data center under construction in Ohio had been completed, powered on, and immediately announced a second, larger data center.