On the one hand, the United States government is trying to reverse three decades of deindustrialization with tariffs on China. On the other hand, investment in AI is recreating exactly the phenomenon that destroyed part of the American industry in the 1990s.
History repeats itself, but this time knowing what is going to happen.
Why is it important. Derek Thompson, business reporter for The Atlantic, has identified a pattern that rewrites what we thought we knew about American industrial decline. China not only stole jobs but American capital abandoned them early.
In an interview with the investor Paul Kedrosky for his podcast Plain EnglishThompson presents his thesis:
- In the nineties, the massive deployment of the Internet and telecommunications absorbed brutal amounts of money.
- That money had to come from somewhere. He left the factories. Small manufacturers saw financing becoming increasingly more expensive. Just at that time, China was entering the World Trade Organization and trade barriers were falling.
It wasn’t bad luck. It was cause and effect.
The context. Technology companies are going to spend about $400 billion this year building infrastructure for AI. To put it in perspective: the Apollo program that took the United States to the Moon cost about 300 billion adjusting for inflation. That was ten years. This is a year.
Data centers have accounted for half of US economic growth in the first six months of 2025. The forecast is that investment exceeds 500,000 million annually in 2026 and 2027.
Meanwhile, American consumers are spending $12 billion a year on AI services. The difference between what is invested and what is earned is abysmal.
The panoramic. The problem is structural. If you manage an investment fund with 500,000 million, you have two options:
- You can distribute that money among a hundred small factories that need five million each.
- Or you can write ten $50 billion checks to AI projects.
The first option means managing a hundred different companies. Sit on dozens of tips. Do constant monitoring. The second means ten meetings a year. The choice is obvious.
- A manufacturer that wants to take advantage of the moment to bring production back to the US finds that borrowing money is very expensive.
- Banks compare their project with the returns that AI promises.
There is no color.
The irony. Trump has built his economic policy on tariffs that force companies to manufacture in the US. But investment in AI is making it more expensive exactly what the tariffs are trying to make cheaper: producing locally.
- Tariffs raise the price of importing from China.
- AI raises the cost of financing local production.
The net effect may be zero for the industry, but with higher prices for everyone.
The figures. Building a modern data center involves…
- That 60% of the budget goes to NVIDIA chips.
- The rest is divided between refrigeration, electricity and construction.
- The physical building is the cheapest part.
Geography also counts. Northern Virginia concentrates a good part of the investment. Areas that were rural ten years ago are now surrounded by industrial facilities that operate 24 hours a day.
Yes, but. There is a way out that did not exist in the nineties: set up data centers outside the United States. India and the Middle East are receiving huge investments because electricity is cheaper and your neighbors, ahem, complain less.
But that makes the original problem worse. If the money goes to data centers in other countries, there is even less left for American factories.
Between the lines. Kedrosky uses a simile that sums it all up: a death star that absorbs capital. In the nineties that star was the Internet. Now it’s AI. The factories, in both cases, are collateral damage.
The difference is that in the nineties no one saw it coming. Now yes.
In Xataka | Spain has a railway giant in the shadows. And he just got the “contract of the century”
Featured image | Cemrecan Yurtman
GIPHY App Key not set. Please check settings