Even if the AI bubble pops, we could keep the power

The warnings were about money. Ben Thompson, the technology analyst who writes about aggregation theory, was describing how the AI buildout could stall if investors run out of cash before revenues arrive to pay for it. Then he offered his test for a good bubble.

"It has to be power," he said.

What Thompson means is that the most useful bubbles leave something physical behind after shareholders lose their money. He compared AI to railroads, which he called the last buildout on this scale. Building a railroad took a decade or more to pay off, but it was funded with short term money, and in the 1870s, he said, the world literally ran out of money.

The dot-com bust, he argued, did the same kind of good in the end. He said Google picked up unused fiber for next to nothing after the bust, and that the core of the internet still runs on fiber laid by WorldCom, a company that failed. The railroads left lines whose value still flows today, including BNSF, which he said is still throwing off money.

For AI, Thompson said, he doubts graphics chips and data centers will be that kind of legacy. He said the graphics chips do not last that long, and after allowing that data centers count for something, he asked what the lasting thing would be. Electricity does, because energy undergirds everything, in his telling.

He said the United States has already brought more power online than he expected, through generators installed directly at industrial sites, natural gas development in West Texas, and restarted nuclear plants. A power constraint is still coming, he believes, but he thinks the squeeze has arrived later than people had anticipated.

"If we're in a world where this all blows up and we have way too much power, that is an amazing world to be," he said. Humans have always lived with too little energy, he added, so abundance is hard to even imagine.