The End of the AI Trade?
There’s a pattern that has repeated with every great technology for 150 years.
Every time this country builds something enormous, the same thing happens. Every single time. And it has never once — not one time in a hundred and fifty years — worked out the way the investors expected.
There is always a shift – from the companies building the technology to the ones using it – that destroys investors in the first group of stocks and delivers life-changing returns for those in the latter.
The same shift will happen with AI stocks – not if, only when. The question is, is it happening right now? Because there is over 150 years of precedent for how the final stage of AI is going to play out.
Start with the railroads. The most important thing America built in the 19th century. Thirty-five thousand miles of new track laid between 1866 and 1873. It worked. It connected a continent.
And by 1877, twenty percent of American railroad track mileage was in receivership. A fifth of the network — bankrupt. Forty percent of all railroad bonds were in default.
Railroad stocks lost sixty percent of their value. When Jay Cooke and Company went under in September of 1873, the New York Stock Exchange closed its doors for ten days for the first time in its history.
Now jump to the internet; fiber optics. In the five years after 1996, telecom companies poured more than 500 billion dollars into cable.
By the early 2000s, less than 2 percent of it was being used.
Global Crossing raised about 20 billion dollars, laid a hundred thousand miles of undersea fiber, and filed for bankruptcy in January of 2002. Its assets sold for pennies on the dollar of what it cost to build.
So who got rich off all that fiber? Google. Amazon. Netflix. Companies that never laid a single mile of it.
They built their empires on bandwidth that was practically free — because somebody else had already gone broke providing it.
The same is true of the airlines. The airplane might be the most transformative machine of the 20th century. Warren Buffett said that as of 1992, all the money made by every airline company in this country since the dawn of aviation added up to zero. Absolutely zero. Between 2000 and 2008 they lost another sixty billion. And this year the industry’s own forecast has it earning about 6.8% on capital that costs them 8.2% — which means it is still, right now, in 2026, losing money for the people who fund it.
Look, here’s the thing…
The technology always works. That’s not the question. The question is who gets paid for it. And the answer is almost never the people who built it. It’s the people who use it.
The only question is: When will the shift happen?
Last month’s blow up of the Situational Awareness hedge fund happened for 2 reasons:
- Aschenbrenner used 4:1 leverage on the long side
- AI infrastructure stocks cratered while software ripped higher
He was long infrastructure – Coreweave, Nebius, Micron, etc. All of those stocks fell 30-50% in a month.
At the same time, he was short software – Workday, Microsoft, Adobe, Salesforce. Those stocks all went up.
Throw in some leverage, and that’s how to turn $45 billion into $10 billion in a month.
Painful lesson.
But it showed a shift in the AI theme. The market didn’t crash; money just walked out of one end of the AI trade and into the other.
Many investors assume that those who build the new technology will see the greatest investment yield. But there is no relationship between those two things. There never has been.
Amazon spent about 131 billion dollars on capital expenditure in 2025 — roughly 95 cents of every dollar the whole business generated in operating cash, poured straight into the ground. This year they’re guiding to $220 billion.
Oracle’s free cash flow went negative, and its long-term debt nearly doubled.
And CoreWeave — one of Leopold’s biggest positions — is carrying term loans at 11-15% interest. All so the business can steadily burn cash.
Valuations have gotten, uhhh… optimistic. And they won’t be justified forever. Many of the big winners of the last three years are nearing long term peaks.
The AI infrastructure trade that has been powering this market for the last four years is likely to shift in the next 12 months.
It happened to railroads in a decade. Fiber took about five. We are currently 4 years into the AI buildout which, to me, feels like the 8th inning.
I’m not saying the stock market is going to crash. But I do expect to see rotation. Money will shift from stocks building a commodity (compute) that will only become cheaper every year, to the companies using the commodity to build an empire.
And there is a simple test to determine which you have. Ask yourself this:
When compute becomes meaningfully cheaper, will it help this company or hurt it?
