As enthusiasm for artificial intelligence drives a global surge in capital spending, comparisons with earlier technology manias abound. James Knoedler, portfolio manager of the Evenlode Global Equity Fund, examines what today’s AI boom may have in common with the great railroad bubbles of the 19th century — and why history’s lessons on speculation, misallocated capital and misplaced assumptions remain uncomfortably relevant.
The current AI investment cycle has been compared not just to the tech cycle of the late 90s but also to ‘the railway boom’, which was really a series of bubbles in the US and elsewhere which ran through most of the 19th century.
So what are the parallels to today’s capex explosion and the construction of the US transcontinental railroads? In both cases, capital was mobilised on an enormous scale to build existing technology in an unprecedented speculative leap. This rested on the assumption that profitable businesses could be created once the massive infrastructure was in place.
Exact timing of the evolution of the boom is hard to predict, but there are some important principles which can be drawn from the transcontinental experience which remain valid today.
Similar tracks
Firstly, calling a speculative bubble ex ante does not immunise against a speculative bubble. The historian Richard White said that during the frenzied construction of the 1880s, ‘the railroads were like racers at the starting line in a race none of them particularly wanted to run. They would run only because they feared their rivals would run without them’.
I grew up at the edge of the Great Plains, the Front Range of Colorado, and while we were near a big city (Denver) it was just emptiness to the north, east, and south. Most Europeans have no idea how vast and lonely the real high plains in Montana and Wyoming and the Dakotas are.
Contemporary letters and diaries from management and their bankers make it clear how aware they were of the irrationality of the cause: ‘capital [was being] wasted … in wild cat enterprises such as Railroads through deserts – beginning nowhere and ending nowhere,’ said one; ‘we are all going to the devil, and going together’, said another.
They were proved correct. Railroad bonds outstanding grew 27% per annum in 1867-1874 with predictable results; by 1875 40% of total railroad bonds were in default. This devastation was followed by another boom and then another collapse in 1893 when again, a quarter of total US railroads by capitalisation entered receivership.
Secondly, beware wishful thinking. The builders of transcontinentals over-extrapolated from the economics of Eastern railroads which connected large cities and ran through densely settled farmlands. Managements knew that arable farming was impractical much west of the 98th meridian but promoted the myth that ‘Rain follows the plow’, in which increased farming would magically promote increased rainfall.
This theory was endorsed by academics and the US Geological Survey, but it proved flat wrong. Other assumptions about the economic potential and pricing power of the transcontinentals were also wrong. They were structurally uncompetitive with ocean transport for coast-to-coast travel. In intracontinental travel, overbuilding eroded pricing power even as the capital stock of the railroads accumulated, forcing the roads to issue debt to cover interest costs which in turn incentivised railroad managers to keep slashing rates to compete for cash revenue.
AI is in an even worse position. Railroads enjoy very attractive incremental margins – the second ton of freight or second passenger has very modest extra costs over the first one, and so on – whereas AI combines very high fixed and variable costs.
Rose-tinted economics
Other people today who have read White’s book have reached the rosy conclusion that as plenty of people are on ChatGPT, things aren’t as bad as they were in the railroad days. But this misses the point. The transcontinentals could have offered $10 fares from coast to coast and found plenty of takers, but lost money on every passenger. This is functionally the equivalent of what OpenAI is doing. There is an old business saying that the easiest way to get revenue is to sell $2 of value for $1.
Thirdly, when the mine shuts down, suppliers of picks and shovels are left swinging. In the fifteen months after Jay Cooke’s company failed in September 1873, triggering the Panic of 1873, 50% of America’s iron foundries closed down. Nationwide bankruptcies doubled year on year.
Fourthly, don’t count on insider alignment. The apparent recklessness of the insiders who knew trouble lurked ahead makes more sense when we know that great dynastic fortunes were made despite the terrible outcomes for public investors. Complex related party transactions made them rich. Promoters were often tremendously gifted salesmen unburdened by moral purpose or technological savvy.
Fifthly and lastly: the transcontinentals might show that ‘creative destruction’ is a fatalistic myth. It is fashionable when discussing the parallels between past booms and today to take the Panglossian view that bubbles are a necessary condition for growth. Yet it is impossible to identify the value created by building six transcontinental railroads in North America at mad haste.
The economic historian Robert Fogel won a Nobel for his work demonstrating that railroad investment had surprisingly modest overall benefits to US economic growth. Even if this is an exaggeration, the societal value created by having six transcontinental railroads built so far ahead of need is questionable when set against the devastating ecological, human, and capital costs of the roads.
Humans are addicted not just to the Great Man but to the Great Gadget theory of history. Fogel and White’s close empirical analysis of the American railroad booms make a persuasive case that capital was misallocated unnecessarily. The machine of perverse incentives which promoted so much useless track would have rolled on and created even more unviable capacity if not for the efforts of hard-headed financiers like J P Morgan and E H Harriman.
We might all benefit today from paying attention to Fogel’s Nobel citation which concluded that ‘The sum of many specific technical changes, rather than a few great innovations, determine[s] economic development’.
