Railway mania vs the AI bubble: infrastructure booms compared

What was railway mania, and why does it echo in the AI bubble?

Railway mania was a British investment boom that ran through the 1840s. Parliament authorised hundreds of new railway companies, 263 Acts in the peak year of 1846, and investors rushed to buy shares. Many lines were built, and many authorised lines never were. The boom funded a network that the country used for generations.

It echoes in AI because the structure looks alike: a general-purpose technology, heavy up-front spending, and a promise that demand will fill the capacity. The boom shows that a useful network and a bad investment can coexist. Both facts held at once, and the AI debate needs both in view.

How did railway mania financing compare with AI bubble financing?

Investors could buy railway shares with a deposit of 10%, and the company could call the rest later. A small sum therefore controlled a large position, and leverage magnified the fall when prices dropped. Falling bond yields pushed money toward shares, and later a rise in rates sent money back to bonds.

For AI, ask how the build-out is financed: company cash, bank loans, bonds, or private credit. The EQOS Asia AI study records about $200B a year invested in AI (EQOS Asia AI study, April 2026). Cash funding leaves room to adjust. Debt that falls due on a fixed date gives less room if revenue arrives late.

Do AI chips and railways age at the same speed?

A railway track lasted for decades, and its value came from carrying goods and people year after year. An AI chip loses value faster, because newer chips arrive every year and make older ones less competitive. A data centre building ages slowly, while the equipment inside it ages quickly. The mix matters for the return on the whole investment.

Short asset life raises the bar for revenue. If equipment must be replaced every few years, the revenue must repay it before replacement. Ask each AI infrastructure company for the useful life it assumes in its accounts. A long assumed life on fast-ageing equipment flatters profit today and shifts the cost to later years.

Who gained after railway mania, and what does that say about the AI bubble?

When railway share prices fell, many early investors lost money, and larger companies bought failed lines at low prices. Some lines that failed as independent companies became viable under new owners. Users gained throughout: the network existed, and the cost of moving goods fell. The gain moved from the first investors to the later owners and the customers.

AI infrastructure may follow the same path. Capacity built at high prices can serve customers at lower prices later, and buyers of that capacity gain. For an investor the point is the timing of entry: the early buyer carries the risk of the overbuild, and the later buyer collects the benefit. Know which role your position plays. A supplier, a lender, and a shareholder each sit at a different point on this timeline, and each carries a different share of the loss if capacity arrives before customers do.

Which five checks separate a useful AI bubble build-out from an overbuilt one?

Check demand first: are paying customers filling the capacity? Check financing: does cash or debt fund it? Check asset life: how many years until replacement? Check ownership: does one group of companies own the capacity it also rents to others? Check fallback use: if the first customers leave, can the capacity serve other buyers?

A build-out that passes all 5 checks resembles the lines that were used for decades. A build-out that fails 3 or more resembles the third of authorised lines that never opened. Record the answers each quarter, and you will see whether the AI build-out moves toward use or toward excess.

How does the AI build-out reach Asia, as railway mania once reached suppliers?

Railway mania lifted iron, coal, and engineering firms along the supply chain, and a fall in orders reached them quickly. AI follows the same pattern in Asia, where chip makers in Taiwan and Korea, and data centre builders in Singapore and Malaysia, sell to AI spenders. A cut in orders at the top reaches these suppliers within quarters.

Suppliers should therefore watch the spending plans of their largest customers and the share of revenue that comes from them. A supplier with one dominant customer shares that customer’s risk. Diversifying the customer base is a clear defence, and it can start before any slowdown appears in the headlines. Contracts with minimum purchase terms add a second layer of protection, because they turn a hope of future orders into a commitment that both sides can measure.

How can EQOS Asia help you size the railway mania risk in AI?

The EQOS Asia AI study scores the enterprise AI market at φ(S) = 0.337, in the fragile zone where a shock can tip the system, and records 95% of AI projects failing in production (EQOS Asia AI study, April 2026). Weak buyer results are the main risk to the capacity being built. The EQOS Asia method page shows how the score is built.

If your company supplies or buys AI infrastructure in Singapore, ASEAN, or Hong Kong, the Λ simulator estimates a score for your case in 10 minutes, free and anonymous. The result gives you a first reading of your exposure before you commit to a larger analysis.

Frequently asked questions

About the author

Stéphane Jambu

President, EQOS Forecast Asia

Stéphane Jambu leads EQOS Forecast Asia and works as a consultant in e-commerce, search visibility, and AI. LinkedIn profile