AI as the fourth industrial revolution: history says being right about the technology isn't enough

I'm building a disruptive innovation position, slowly, with AI exposure layered on top, and I'll keep writing about it as it grows. So the other side belongs on the record too. The last two times the world was this sure a technology would change everything, it was right. A lot of the people who invested in it lost money anyway.

Schwab said it first, and he didn't mean AI

"Fourth Industrial Revolution" isn't an AI term. Klaus Schwab coined it in a 2015 Foreign Affairs essay and made it the theme of Davos the following January. He meant a broad fusion of physical, digital and biological technology: sensors, robotics, 3D printing, gene editing, with AI as one strand among several. When Huang and others use industrial-revolution language for the AI buildout now, they're taking a bigger idea and narrowing it to one technology. That doesn't make them wrong. It does mean the phrase arrives with more authority than it's earned in this particular use.

The railways were real

Britain in the 1840s is the cleanest case. Railways changed how people and goods moved, and most of the lines laid then are still carrying trains. In 1846 alone Parliament passed 272 Acts setting up new railway companies, covering around 9,500 miles of proposed track. About a third of that mileage was never built. Some companies collapsed under bad planning, some were swallowed by rivals, and some were frauds from the start.

The share prices tell the rest. The railway share index hit 1,984 in August 1845. By April 1850 it was 673, down about two-thirds. Many investors had bought partly-paid shares, putting down a small deposit with the company free to call the balance whenever it liked. Through the crash, the companies called it. People who thought they'd risked 10% of a share's price found out they owed the other 90% on something now worth a fraction of what they'd paid.

The technology thesis was right. The railways got built, and the country ran on them for a century. The shareholders were a different story.

So was the internet

The dot-com case is closer to home, and it's the platform you're reading this on. Nobody argues the internet was overhyped as a technology. That didn't stop the Nasdaq closing at 5,048 in March 2000 and 1,114 in October 2002, a 78% fall. It didn't close above its 2000 high again until April 2015. Fifteen years.

That's the American version. From here it was worse. On the day the Nasdaq peaked in March 2000, a New Zealand dollar bought about 49.5 US cents. By the day it finally closed above that peak in April 2015, the same dollar bought about 76 cents. So a Kiwi who bought the Nasdaq unhedged at the top was still down roughly a third in their own currency on the day American investors got back to even. Currency cuts both ways, and Issue 10 was about a weak dollar quietly lifting overseas portfolios. But every "the market always comes back" chart you've seen was drawn in US dollars, and you don't spend US dollars. It's the hedged-versus-unhedged choice from Issue 7, with a fifteen-year bill attached.

The index didn't save you either

This is the part that should bother anyone whose answer to AI risk is "I just hold an index fund."

The railway losses were mostly a stock-picking problem: people put everything into one or two companies and some of those companies didn't survive. The Nasdaq isn't one company. It was a broad basket of exactly the businesses that were supposed to win the internet, and a lot of them did. Diversification protected you from any single one going to zero. It didn't protect you from waiting fifteen years to get back to even if you bought near the top. For someone five years from retirement in March 2000, "the index recovered eventually" was true and useless. That's the sequencing risk Issue 19 covered, just arriving through a theme rather than a market cycle.

It matters here because most NZ investors already own the AI buildout without having chosen to. The ten biggest companies in the S&P 500 now make up more than a third of the index, and Nvidia alone is around 7-8% of it. Issue 21 made the point that the infrastructure layer is largely Microsoft, Amazon, Google and Meta, all sitting near the top of any cap-weighted global or US fund. If your KiwiSaver growth fund or your S&P 500 PIE fund has done well over the past three years, a good chunk of that is AI. That's not a reason to sell. It is a reason to count it before you add an AI satellite on top and end up with twice the exposure you thought you had.

What I don't know

Whether AI turns out like electrification, which reshaped the whole economy and paid patient, broad investors over decades, or like 2000, where the investors were right about the destination and still waited fifteen years to break even. Nobody knows. Anyone telling you confidently either way is guessing, whichever side they're on.

What history is consistent about is the mechanics. The people who came through with their money were the ones sized so that no single bet could sink them, and who built up over years instead of all at once on whatever headline felt most urgent that month. That's the satellite sizing from Issue 20 (a few per cent of the portfolio per position, not a quarter of it) and the dollar-cost averaging from Issue 13, applied to a theme rather than an exchange rate.

If it's electrification, being early and patient pays. If it's 2000, being early and patient means a very long wait. I don't know which one this is. I'm still building the position, in small amounts, on purpose.