AI and the railway mania
The AI bubble continues. The Magnificent Seven of tech media continue to drive the US stock market, along with the AI companies. The ten largest US companies by stock market value, holding over 40% of the total market cap of the S&P-500 index, have continued to shift upward in price, well above any increase in earnings (profits) recorded.

Recent earnings from major US ‘hyperscalers’ (AI development companies) show that revenue growth remains strong, but free cash flow is being sucked up by accelerating capital expenditure. So these firms are turning to leasing and new debt to sustain the AI development race. The AI investing companies now represent 75% of S&P 500 returns, 80% of earnings growth and nearly 90% of capital-spending growth in the last year. Global AI infrastructure investment is near $400bn in 2025 and by 2030, cumulative spending could exceed $5–7 trn. Roughly 60% of this investment will go on semiconductors and computing hardware, an unprecedented level of investment in a new technology just starting commercial use.

It is not entirely true that infotech investment is the total driver of US economic activity. A lot of the equipment going into data centres is imported, so that means there are offsetting negative contributions to GDP. Even so, the ‘silicon mountain’ continues to erupt upwards.

The AI bubble, and that is what it is, has startling similarities with the so-called ‘railway mania’ in Britain in the 1840s and later in the US in the 1870s. Then railways were also seen as a powerful new technology that could transform transport and travel, and so boost productivity. This led to massive speculation in railway shares as company after company launched a new rail line across Britain during the 1840s – and later in the 1870s right across the US, culminating in the transcontinental rail connection.

In the UK, the mania reached its zenith in 1846, when 263 Acts of Parliament for setting up new railway companies were passed, with the proposed routes totalling 9,500 miles (15,300 km). About a third of the railways authorised were never built—the companies either collapsed because of poor financial planning, were bought out by larger competitors before they could build their line, or turned out to be fraudulent enterprises to channel investors’ money into other businesses.
Between the 1860s and the 1900, the transcontinental rail tracks transformed America. They helped populate the west, and as in Britain, developed a new from of capitalist enterprise, the joint stock company, i.e. publicly owned and financed corporations. Rail helped turn the US into a coast-to-coast dual-ocean superpower and revolutionized modern finance. As the historian Richard White wrote in his history of the transcontinentals, Railroaded, “they created modernity as much by their failure as their success,” by leaving behind “a legacy of bankruptcies, two depressions, environmental harm, financial crises and social upheaval.”
So far, the AI investment boom has not yet reached the size of that 19th century railway investment, which eventually hit 6% of US GDP compared to 1.2% of GDP invested in AI datacenters and 4% of GDP in overall information processing so far. But it’s getting there.

In the ‘Railway Mania’, eventually the stock market bubble burst. In the US, the trigger for the panic of 1873 was the failure of Jay Cooke & Co., America’s premier banking house. Cooke had made his name as the chief financier of the Union army. He agreed to fund the Northern Pacific Railway’s project to connect the Oregon coast with the existing northeastern rail network. But the first transcontinental line had already been completed and concerns about overcapacity, along with increasing distrust of railroad securities drove down Northern Pacific bond prices. Cooke’s firm went bust in September 1873, precipitating a stock market panic and eventually a worldwide depression that extended into the 1890s.
Marx commented at the time that the huge concentration of stock market investment in US rail companies “gave in one word an impetus never before suspected to the concentration of capital and also to the accelerated and immensely enlarged cosmopolitan activity of loanable capital, thus embracing the whole world in a network of financial swindling and mutual indebtedness, the capitalistic form of ‘international’ brotherhood’”. When the rail company stocks fell, the rest of the market fell and an economic slump ensued.
In Britain, the railway bubble burst around 1847. Marx only analysed that bubble some 20 years later in Capital, Volume 3. There he called it the ‘great railway swindle’ to emphasise that the claims made by the railway companies for huge profits to be made were deliberately overstated. Investors naïvely poured capital into schemes that were significantly less profitable than they expected and promoters and directors had promised. The railway mania of the 1840s took place when the average rate of profit on British capitalism was falling – and it continued to do so through the 1840s. Marx noted, “in the railway swindle from summer 1844, railway investors apparently expected much higher than the average profit rate.” Those hopes were dashed by 1847.

It’s the same issue now. If the returns on massive AI investments turn out to be lower than expected and claimed, that will cause a serious stock market correction. In other words, the mechanism by which an AI bust could cause a recession is not through suddenly lower growth, but through a failure to obtain expected returns on investments.
For now, optimism remains among the tech sector. Mark Zuckerberg recently predicted that half of Meta’s code would be written by AI within a year. But so far, most companies are seeing little if any benefit from their initial investments. A widely cited study from MIT found that so far, 95 per cent of generative AI projects produce no return in productivity growth or profits. To justify the required investment, annual data-centre revenues would need to rise from $20bn today to about $2trn. Existing revenues will fall short by $800bn, according to Bain & Company. Even with expected efficiency gains, that gap illustrates how far current valuations depend on unproven revenue streams. That gap will have to be filled by borrowing and raising debt and equity.
Increasingly, investment in AI assets is being financed by loans and borrowing by the AI companies, while the stock investors also borrow more to leverage up their financial bets on AI. OpenAI’s data centre partners are on course to amass almost $100bn in borrowing for investment in OpenAI. So far, SoftBank, Oracle and CoreWeave have borrowed at least $30bn to invest and a group of banks is in talks to lend another $38bn to Oracle and data centre builder Vantage to fund further sites for OpenAI. Investment group Blue Owl Capital and computing infrastructure companies such as Crusoe also rely on deals with OpenAI to service about $28 billion they have in loans. Lenders and bond holders are starting to get worried and taking out increased default insurance on Oracle if it should not be able to service its debt.

Gita Gopinath, former chief economist at the IMF, has calculated that an AI stock market crash equivalent to that which ended the dot-com boom would erase some $20tn in American household wealth and another $15tn abroad, enough to strangle consumer spending and induce a global recession. But the argument goes that, even if there is financial bust and even if there is an ensuing slump, the best companies will survive and the huge productivity gains from the application of AI in all sectors of the economy will eventually deliver a step-change in the growth of the productivity of labour. Output will rise because AI will replace human labour, reducing costs for companies and boosting profitability. After all, even though the 1873 panic led to a market collapse in railway stocks and a deep recession, in the end, the US had a legacy of a rail network across the continent. Similarly in Britain, after the slump of the late 1840s, the subsequent long boom of the 1850s rested partly on the 6,000-mile rail network which then formed the backbone of the country’s transportation system and helped Britain maintain its global hegemony.
Will AI do the same for US capitalism, currently facing increasing rivalry to its global hegemony? Possibly not – after all, the Magnificent Seven may be riding high in the stock markets, but their technological advantage is seriously under threat. Last year, China delivered DeepSeek, a much cheaper, but nearly just as good Large Language Model (LLM) as OpenAi’s ChatGPT. And this year, there have been new launches of Chinese LLMs that perform as well and cost a fraction of the investment made by the US companies.
Mainstream economists remain divided on whether AI will deliver in the same way that railways did in the 19th century or the internet did in the late 20th century. Stanford University economist, Eric Brynjolfsson predicts that AI will follow a ‘J-curve’ in which initially there is a slow, even negative effect on productivity as companies invest heavily in the technology, before they finally reap the rewards. And then the boom comes. The J-curve can be seen in US manufacturing productivity growth, which fell in the mid-1980s and then after the recession of 1991, accelerated sharply until the mid-2000s.

But Daron Acemoglu, an economist at MIT and 2024 Nobel Prize winner, argues the productivity gains from generative AI will be far less and take far longer than AI optimists think. Moreover, AI companies are too narrowly focused on ChatGPT and other AI products that have little relevance to most business sectors. Others point out that despite smart phones and social media and apps such as Slack and Uber, past digital technologies have done little to grow the economy. Next year should reveal who is right.



