When a real technology still creates a bubble: railways, the internet and AI

Can a real technology still create a technology bubble?

Yes. A technology bubble needs two ingredients: a technology that works, and prices that run ahead of what it earns. Usefulness supplies the first ingredient and makes the story credible. It leaves the second open, because price depends on what buyers pay, not on what the technology can do. A reader who separates the two questions can hold a positive view of the technology and a measured view of its price.

This explains why the AI boom question and the AI bubble question can both receive a yes. Real use cases make investors confident, and confidence raises prices, sometimes beyond earnings. The sections below test that sequence on two historical records and then on AI, with one source and one date for each fact.

What pattern does every technology bubble share?

Technology bubbles share a sequence. A real advance opens a market, early investors profit, and the profit draws more capital. Builders then plan more capacity than early demand can fill, since each builder expects a share of the whole market. Supply arrives together, and revenue per unit of capacity falls short of the plan.

The sequence has a useful property: it can be observed before the end. Capacity plans, proposed projects, and funding volumes are public. A reader who tracks them can see whether plans exceed demand. Rising plans do not prove a bubble, and they show where to look for the gap between plans and use.

What did railway mania show about a real technology and a bubble?

Railway mania in Britain ran from about 1840 to about 1850 and peaked in 1846, when 263 Acts of Parliament authorised new railway companies with proposed routes totalling 9,500 miles (Wikipedia, Railway Mania, consulted 22 October 2026). Railways were real and transformed transport. The plans still exceeded what the market could use. Proposed miles measure ambition, and built miles measure what demand supported.

The record shows the outcome. Projects authorised from 1844 to 1846 produced 6,220 miles of built line, and about a third of the authorised railways were never built. The same article links the turn partly to the Bank of England raising interest rates in late 1845. A real technology and a funding squeeze produced a bubble and a reset together.

What did the dot-com crash show about a real technology?

The internet is the closest modern case. The Nasdaq Composite peaked on 10 March 2000 at 5,048.62 and fell 78% from that peak by October 2002 (Wikipedia, Dot-com bubble, consulted 22 October 2026). That fall measured how far prices had moved ahead of earnings, and the technology remained in use afterwards. An index level summarises thousands of company prices, so it shows the mood of the whole market.

Two lessons follow. First, a fall of 78% in an index can coexist with a technology that stays important. Second, timing stayed uncertain even in hindsight: the peak and the low sat more than 2 years apart, so an early reading of prices could not fix the date. Readers who need a date can use scenarios instead.

How does AI compare with railways and the internet as a technology bubble?

The EQOS Asia AI study records valuations up ×4 in 18 months and about $200B a year invested in AI research and deployment. In railway terms, these resemble the authorised routes. In internet terms, they resemble the index level. They measure plans and prices, and neither measures the use that will repay them.

On the use side, the study projects productivity gains of +40% and counts 95% of deployed projects failing in production (MIT 2025). Real use exists, as it did for railways and the internet, and delivered use trails the plans. The study scores the AI market at φ(S) = 0.337, in the fragile zone. Compare both figures with your own sector before you apply them.

What remains of a real technology after a bubble deflates?

The railway record gives a concrete answer. Many failed lines were bought by larger companies, and some became viable under their new owners. The assets survived, and the original investors absorbed the loss. In a bubble, infrastructure often outlives the prices that financed it. The pattern gives a second reading of the same event: a loss for the first owners and an opportunity for the next.

For AI, the parallel assets are data centres, chips, trained teams, and tested products. Which of these keep earning depends on demand, so the useful task is to identify the buyers that would remain after a price fall. Exposure analysis then follows who owns the assets, who lent against them, and who depends on their output.

How can you apply the technology bubble lesson to AI today?

Check four items each quarter. Is the technology working in customers’ hands? Do valuations run ahead of sales? Are builders planning more capacity than current demand? Who would buy the assets after a reset? Record the answer with a date. These items come from the railway and internet records and fit on one page.

The Λ simulator turns your exposure into a first score in 10 minutes, free and anonymous. If your decision involves AI spending in Singapore, ASEAN, or Hong Kong, the EQOS Asia 30-minute pre-diagnostic returns a score and a ranked set of options (EQOS Asia AI study, April 2026).

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