See "Spending on AI data centers is so massive that it’s taken a bigger chunk of GDP growth than shopping—and it could crash the American economy"[1]
The US is having a giant AI boom and a recession in the rest of the economy.
Historically, the classic example of this is railway mania of the mid 19th century.[2] That started in 1830, with the Liverpool and Manchester Railway.[3] This was when the industrial revolution got out of beta. There were earlier railroads, but with dual tracks, stations, signals, tickets, schedules, and reasonably good steam engines, the Liverpool and Manchester worked like a real service. It was profitable. Then lots of others started building and over-building railroads, with varying degrees of success. See Panic of 1847.
It was really too early for good railroads. Volume production of steel didn't exist. Early railroads were built with wood and iron, not very well. Around 1880-1900, everything was rebuilt in steel, and got bigger, better, and safer.
Consider the early Internet. We had TCP/IP working across the US in the early 1980s, but it wasn't a big deal commercially for another 10-15 years, it wasn't everywhere until the early 2000s, and it wasn't out of bubble mode until 2010 or so.
The really scary thing is all the value in the AI boom is predicated on the belief that the technology is "early" and that it will improve over time. We're seeing the opposite. Instead, all the competing models are basically converging on the same benchmark performance numbers, as we saw yesterday with the gpt-5 debacle. This suggests that performance is actually topping out, which makes intuitive sense if advancements in LLMs are proportional to their training data. They've already used up all the data. So it very well could be what we see right now is basically as good as it gets, or at least approximately so. The market is not ready for that.
> We're seeing the opposite. Instead, all the competing models are basically converging on the same benchmark performance numbers...
That question is really important. Comments on that? Maybe LLMs are asymptotically approaching some fundamental limit for that technology.
What goes on inside an LLM hasn't changed much in years. More data and more compute is thrown at that little kernel that makes it all go. Additional gimmicks are added around the core algorithm. But there's not much progress down at the bottom.
There's no reaching into the net and extracting a reliable "don't know".
Is there an upper limit to what that architecture can do?
AGI may not be reachable by this route.
Maybe someone will find a way around that limit. The field has way too much money, too much visibility, and too many people. The technology works, although it has limits.
Money can't (necessarily) buy a breakthrough. It could take 10 or 100 or 1000 years, we just can't know. This kind of technical risk isn't usually so prominent. It's usually more diffuse with fallbacks, offramps, less palatable but workable alternatives.. This time it seems like the technical risk is looming very, very large.
Historically, the classic example of this is railway mania of the mid 19th century.[2] That started in 1830, with the Liverpool and Manchester Railway.[3] This was when the industrial revolution got out of beta. There were earlier railroads, but with dual tracks, stations, signals, tickets, schedules, and reasonably good steam engines, the Liverpool and Manchester worked like a real service. It was profitable. Then lots of others started building and over-building railroads, with varying degrees of success. See Panic of 1847.
It was really too early for good railroads. Volume production of steel didn't exist. Early railroads were built with wood and iron, not very well. Around 1880-1900, everything was rebuilt in steel, and got bigger, better, and safer.
Consider the early Internet. We had TCP/IP working across the US in the early 1980s, but it wasn't a big deal commercially for another 10-15 years, it wasn't everywhere until the early 2000s, and it wasn't out of bubble mode until 2010 or so.
[1] https://fortune.com/2025/08/06/data-center-artificial-intell...
[2] https://en.wikipedia.org/wiki/Railway_Mania
[3] https://youtu.be/pDEnsraYx3k?t=505