In case you missed them, look out for my interviews with Grant Williams and Peter Schiff from New Orleans. Well worth your time. More to come :)
I heard Porter Stansberry tell this story about Alvin Adams and the US railroads on a panel we shared at the New Orleans Investment Conference last week. It was the best story I heard all conference, and it has been rattling around in my head ever since.
Between 1865 and 1873 the United States built, as Porter put it, “the most consequential physical network in its history”, but in doing so destroyed swathes of capital. Track mileage doubled, from 35,000 miles to 74,000, in a decade. Railroad capital reached $4.5 billion, when the capital of the entire US banking system was $720 million and the federal debt was $2.3 billion. In January 1870, 87% of the shares traded on the New York Stock Exchange were railroad shares.
Every argument you hear about AI today was made about the railroads then, boiling down to: they will transform the country. And they did. The railways changed America. The people who financed their construction lost fortunes.
Indeed, the losses caused the Great Depression, as it was known until the 1940s, when what went on in the 1930s usurped it. In 1873 the New York Stock Exchange closed for ten days because of the failure of Jay Cooke & Co. By 1876, 134 railroads were in default on $500 million of bonds. By 1877 a fifth of American track was in receivership. European investors lost around $600 million.
Then competition meant that revenue per ton-mile fell by around 60% between 1870 and 1900. The traffic came, just as promised, but the users got the benefit, not the shareholders.
Some years earlier, having lost his shirt in the Panic of 1837, a businessman in his late 30s by the name of Alvin Adams began carrying packages - letters, cheques, bearer bonds, valuables - between Boston and New York. His business did so well that in 1854 it was incorporated as Adams Express Company, with $1.2 million of capital. It did not own a single mile of track. It bought space on other men’s trains and transported stuff.
By 1875 its capital was $12 million. Porter told the story of how it paid an $8-a-share annual dividend through the Great Depression of the 1870s and the subsequent downturns. Adams Express was able to build a successful business using the railway network without financing the construction of that network itself. In 1929 Adams Express became an investment fund. It still trades in New York today as Adams Diversified Equity Fund (ADX)
By 1890 the express companies, American Express among them (yes, the one that’ll do nicely) , were moving 115 million packages a year over 175,000 miles of routes without owning a single locomotive. It’s like Trip Advisor being the world’s biggest travel company without owning a hotel, or Uber not owning a taxi.
The business that used the railways, rather than build them, was the one that did best in the railway boom and bust.
Cummings, AI and power
Which brings me to this article, AI and Power, by Dominic Cummings. I should say that I find Cummings difficult to read, and I have little patience for writers who don’t put in the hard work required to make life easy for readers. But Cummings is box office, and few people understand the machinery of power the way he does, so he gets a pass.
His case is that AI will impact not just the economy, but the balance of political power. If machines do an ever-growing share of work currently done by humans, the implications for jobs, wages, tax, the welfare state and the relationship between citizen and state are profound. Disruption will arrive faster than institutions can adapt, and power will sit with barely handful of companies. We may get growth. We will also get upheaval and a political settlement that becomes increasingly difficult to hold together.
Which all makes me want to own gold.
If AI weakens the tax base while increasing demands on government, the pressure on already stretched public finances gets harder. Governments will respond with yet more borrowing, printing and repression. Then gold comes into his own. Plus you can’t prompt it into existence.
Elon Musk, meanwhile, has been making the point that producing computing power is one thing, finding profitable uses for it is another. Trillions are going into data centres, chips, power and cooling. Their value depends on what businesses do with the compute, what they will pay for it, and whether the revenue ever justifies the capital.
If compute gets cheaper, usage could explode and businesses could run at a fraction of their old costs. Great for the users, potentially bad for the sellers, who may end up competing to supply a service/commodity at ever lower prices. Revenue per ton-mile all over again.
The question is where the profit ultimately accrues. It almost certainly won’t be with the companies building the infrastructure, or even those selling the computing power. It will be with the businesses that use it to reduce costs, improve services and increase margins. Again, the railway analogy is useful here: the people who owned the track did not make the most money from the networks. It was the people transporting stuff.
If you live in a third world country such as the UK, I urge you to own gold or silver. The pound will be further devalued, as will the euro and dollar. The bullion dealer I use and recommend is The Pure Gold Company. They deliver to the UK, the US, Canada and Europe. More here.
The midterms, mining and oil
The Republicans have taken a hammering in the midterms. Democrats have won control of both the House and Senate, a substantial rebuke to Trump and his administration. That changes the political landscape for the next two years, and it matters for energy, mining, AI regulation and the investment climate generally.
Under Biden, whose administration was pretty green, permitting was an almost insuperable wall for much of the American mining industry. Under Trump, that quickly changed and the US became one of the most attractive mining jurisdictions in the world. But these changes are not permanent. A future Democratic president, if his advisors are sufficiently green, could take policy in a very different direction. Mining projects take years to permit, finance and build, which makes political uncertainty especially consequential.
The Iran conflict has clearly contributed to the Republican defeat, not least through its effect on energy prices and the cost of living. If Trump wants to help his party recover before 2028, he needs to find a way to end the conflict ideally while saving face. Good luck with that. That would take some of the geopolitical premium out of oil, and we saw some of that with the sell-off this week. But Iran was only the trigger for higher prices. The deeper problem is a decade of underinvestment, and I don’t see oil back below $60 on any sustained basis.
The revolution is real. So is the bubble
All of which brings us to Henrik Zeberg’s superb piece, The Revolution Is Real. The Bubble Is Too, from last month. Anyone who uses AI can see how transformative it is. Anyone who studies bubbles can see how precarious the mania around AI is. Both are true, as they were of the railways, of electricity in the 1920s and of the internet in the late 1990s.
Zeberg reckons AI capital expenditure could average 3.6% of US GDP between 2025 and 2032, amounting to some $10.3 trillion over eight years. How much infrastructure can be built, and at what cost, before the revenues generated by its use justify the expenditure? That’s the question.
The railways and the fibre boom at least left something behind. Trains still run on the same lines today, and the fibre that bankrupted its dotcom financiers became the infrastructure of the internet. AI hardware is different. It does not have the same useful life. Track laid in 1870 could still be earning money a century later. A chip bought in 2026 may be economically obsolete by 2030, even if it is still physically functional.
On the other hand, today’s hyperscalers - Google (Alphabet), Microsoft, Amazon, Meta, and Oracle, which together control over two-thirds of global AI compute capacity - are not Northern Pacific. They are cash-generative giants with substantial existing businesses and customers. But if capital expenditure gets too far ahead of demand, financing tightens and returns disappoint. If that spending has been propping up economic growth, AI stops being a stock market problem and becomes a macroeconomic one.
Especially given the concentration of capital in maybe two handfuls of US companies, including those five.
The question, then, is how to position yourself for a technological revolution that may also produce one of the great investment busts. Here’s how I’m approaching it.
