All Aboard The AI Railway

At the entrance to Pistolmakers Row in Doune, you could once catch a train. It took two attempts to get one there. In 1846, at the height of Britain’s Railway Mania, Parliament approved a line from Dunblane through Doune to Callander. The promoters couldn’t raise enough money. A second attempt succeeded, and the line opened in July 1858. I think of those twelve years when I read about the money going into AI. Being right about the future of a technology and being right about the finances of a particular project are two different achievements. (douneanddeanston.com, railscot.co.uk)

For anyone who remembers British Rail, it’s easy to read the history backwards. Britain’s railways began as competing private businesses. Companies proposed routes, investors put up money and Parliament decided which schemes could proceed. In 1846 alone, more than 700 railway bills came before Parliament. Westminster had to put up temporary wooden rooms to house all the committees. The familiar national network emerged over generations. Around 120 companies were grouped into the ‘Big Four’ in 1923, and public ownership of the mainline railways followed in 1948. (parliament.uk, lordslibrary.parliament.uk, researchbriefings.files.parliament.uk)

The debt was real. By the end of 1848, UK railway companies owed about £43.7 million in interest-bearing debt, alongside £156.5 million that shareholders had paid in. One reason the boom could gather pace so quickly was that investors could subscribe for shares with a small deposit. The company could call for the balance in instalments as construction progressed. Someone who had backed several promising railways could find that they all wanted cash at much the same time. Doune’s second railway Act shows how closely the two sources of money were linked. It authorised £60,000 in shares and up to £20,000 of borrowing, but the company could borrow only once investors had subscribed for all the shares and paid in at least half. (upload.wikimedia.org, cambridge.org, legislation.gov.uk)

What made it a race was the cost of standing still. A railway director might doubt whether a new branch would pay its way and still prefer to build it before a rival could divert traffic from the company’s existing line. Economic historians have found evidence that protecting territory helped drive expansion. Listen to Alphabet’s Sundar Pichai explaining the company’s AI spending in 2024 and you can almost hear a Victorian railway board at work. For Alphabet, he said, “the risk of underinvesting is dramatically greater than the risk of overinvesting”. Each board can have a sound reason to keep spending. Between them, they can still build more capacity than customers will pay for. (pure.qub.ac.uk, s206.q4cdn.com)

The railway shareholders found out what that could mean. An index of railway share prices fell by more than half between its August 1845 peak and 1850. Yet by 1910 Great Britain had 20,000 miles of railway. The investors who foresaw a country transformed by rail had read the future remarkably well. Plenty still paid too much for a share in it. Today, the US Federal Reserve puts debt issued for advanced computing at more than $680 billion in the first ten months of 2026, up from about $130 billion in the whole of 2023. And an AI investor has a complication a Victorian railway investor might have envied them less for. A railway cutting could serve generations of trains. A data centre may need refitting for a new generation of chips while its original debt still has years to run. (mpra.ub.uni-muenchen.de, parliament.uk, federalreserve.gov, bankofengland.co.uk)

The British Empire adds a harder question. Whose money was at risk? In colonial India, early private railway companies raised capital in London with a guarantee of a 5% annual return on the capital invested. When railway earnings fell short, the colonial government made up the difference from Indian tax revenues. Some lines did fall short, and Indian taxpayers paid. The railways also reduced the cost of moving goods and increased trade between regions. Both facts belong in the story. A railway could benefit the people it served while the terms on which it was financed protected investors thousands of miles away. (cambridge.org, benny.aeaweb.org)

That question brings us back to Scotland. In August 2026, DataVita secured roughly £300 million in borrowing to expand one data centre and build another in North Lanarkshire. CoreWeave, an AI cloud company, contracted for the capacity under a fifteen-year lease. The UK’s National Wealth Fund provided a £202 million guarantee covering part of the banks’ lending. The Indian guarantee protected a shareholder’s return. The Scottish one protects lenders against part of a potential loss. The resemblance lies in what the public promise makes possible. A bank can lend against a signed lease and a government guarantee years before anyone knows how much customers will ultimately pay for the computing power. (umbraco-prod.nationalwealthfund.org.uk, cambridge.org)

In Doune, people could say quite plainly what their railway had done for them. Coal became cheaper. Farmers could get lime for their fields more cheaply. Passengers could reach the next town faster. When that fifteen-year lease in North Lanarkshire is up, what will people there be able to say became cheaper, easier or possible because those data centres were built? Given the public money standing behind the loan, that seems a fair question to ask before the first bill falls due. (douneanddeanston.com, umbraco-prod.nationalwealthfund.org.uk)

Next
Next

Speaking Up and Staying in the Room