Don't Fear the Singularity

In the 1840s, the British railway mania produced one of the most spectacular speculative collapses in financial history. Hundreds of railway companies were formed. Capital poured in from across the social spectrum — clergymen, shopkeepers, servants, novelists. Shares in companies that had never laid a single mile of track were trading at multiples of their nominal value. Parliamentary authorisation was sought for 650 new railway schemes in 1845 alone.

By 1850 the collapse was complete. Three-quarters of the companies formed during the mania had failed or been absorbed. Fortunes built in months evaporated in weeks. The human cost — in ruined savings, broken businesses, and outright fraud — was significant.

And then something remarkable happened. The railways that survived were built. The network expanded across Britain and then across continents. The transformation of commerce that the mania had promised — faster movement of goods, people, and capital; the integration of regional markets into national ones; the urbanisation that followed — arrived exactly as advertised. Just twenty years later, and at prices that bore no resemblance to the mania valuations.

The investors who understood this sequence — who survived the bust with capital intact and deployed it when railway shares were despised — generated generational returns. The investors who bought at the peak of the mania and held through the collapse either recovered eventually, or didn't recover at all, depending on when they had entered and how long they could wait.

The pattern has repeated across every transformative technology since. Not because markets are irrational — they are collectively processing genuine information about genuine opportunities — but because the collective processing consistently overshoots, and the overshoot is funded by credit, amplified by social contagion, and ultimately corrected by the weighing machine reasserting itself against the voting machine.

Every technology cycle has its argument for why this one is different. In 1999, the argument was that the internet had changed the rules of valuation — that traditional metrics like earnings and cash flow were irrelevant to businesses whose value was in network effects and user growth. That argument was partially correct. The internet did change valuation frameworks — but not in the way the bulls imagined, and not at the prices they paid.

The argument for AI being genuinely different deserves serious engagement rather than reflexive dismissal, because it contains real substance.

Previous technology disruptions were largely additive. The railway added speed to commerce without eliminating the commerce itself. Electrification added productive capacity without displacing the workers who used it. The internet added a distribution layer without fundamentally changing the economics of labour. AI is different in a specific and important way: it is the first technology that directly competes with the cognitive labour of white-collar workers at scale — not in specific narrow tasks, but across the entire spectrum of knowledge work simultaneously.

This is not hyperbole. In 2026, a competent developer working with agentic coding tools can replicate the core functionality of mid-market enterprise software in weeks. What once required teams of engineers and months of development time has been compressed by an order of magnitude. The competitive moat of the incumbent software business — built over years at enormous capital cost — can be replicated by a challenger who launched last quarter.

Almost. The time function in development has been greatly compressed. And the same compression is coming to any workflow-based task.

This is the crucial element that the market is struggling to price in: the feature that makes AI truly paradigm-changing — its potential to remove friction and cost in cognitive work. AI disrupts the relationship between intellectual capital and value. The technology that boosts corporate margins by eliminating workers is the same technology that destroys consumer spending by eliminating earners. These are not separate effects. They are the same effect, observed from two different points in the economic chain.

This is what makes the pattern more dangerous in the AI cycle than in previous ones — not because AI is less real, but because the economic feedback loop it creates is more rapid and more systemic than anything a railway or a power grid or even a search engine produced.

The most underappreciated element of the AI disruption story is not the displacement of labour. That story is visible and is being priced, however imperfectly. The less visible story — the one with deeper implications for equity valuations across a much wider set of businesses — is what happens when AI removes the friction on which trillions of dollars of enterprise value quietly depends.

For the past fifty years, the US and global economy built an enormous rent-extraction layer on top of a specific set of human limitations. Things take time. Patience runs out. Brand familiarity substitutes for diligence. Most people accept a worse price to avoid the cost of finding a better one. Inertia is commercially valuable — the subscription that passively renews, the insurance policy that doesn't get re-shopped, the financial adviser whose client can't be bothered to check whether the fee is justified.

Trillions of dollars of enterprise value has been built on these constraints persisting. And AI agents — running continuously in the background, optimising every consumer transaction, never getting tired, never accepting the easiest option out of habit — are removing those constraints systematically.

What makes this mechanism particularly important for equity investors is that many of these businesses currently trade at valuations that assume their moats are durable. The stock market has identified that AI will disrupt some businesses — the "long tail of SaaS" has been heavily sold. What is less priced is the possibility that the disruption extends far beyond the obvious targets, into businesses whose competitive advantage was always, at its core, the exploitation of human limitations that AI is systematically removing.

The reflexivity of the situation compounds this. The companies most threatened by AI become AI's most aggressive adopters, because the alternative is to sit still and die slowly. Each company's individual response is rational. The collective result — a simultaneous reduction in white-collar employment across the economy, feeding into reduced consumer spending, feeding into weaker revenues for the very AI-adopting companies that cut the workers — creates a feedback loop that the individual stock-by-stock analysis consistently underestimates.

What follows is a scenario exercise, not a prediction. The historical pattern — genuine opportunity, speculative excess, correction, realisation — describes a sequence. This is outlined in greater detail in our Crash Dynamics piece. It does not describe a timetable. The duration of each phase is unknowable in advance. The specific trigger of the correction is unknowable in advance. What the pattern does provide is a coherent description of the stress pathway — the mechanism by which the current environment resolves itself if the historical sequence holds.

The most important thing to understand about the historical technology cycle is not the bust. It is what happens after the bust.

After the railway mania of the 1840s collapsed, the tracks remained. The infrastructure installed at mania prices, funded by the capital of speculators who lost everything, became the foundation of British commercial expansion for the next century. The investors who were right about railways — who understood the genuine transformation the technology would produce — were right. They were just wrong about the price and the timeline.

The dot-com collapse of 2000-02 was devastating to investors who had paid mania prices for internet businesses. But the network they funded — the fibre optic cables, the server infrastructure, the protocols — remained. Amazon launched its cloud business on that infrastructure. Google built its search dominance on it. The entire digital economy of the 2010s was built on the bones of the 1990s mania, purchased at distressed prices by investors who had either survived the bust or arrived afterward.

AI will almost certainly follow the same sequence. The models being developed now are genuinely remarkable. The agentic tools being deployed are genuinely transformative. The friction removal from economic intermediation, described in the previous section, will happen — the question is when and at what speed, not whether. The white-collar labour disruption will happen. The productivity gains, that companies are currently struggling to extract from AI, will ultimately be real.

The issue is entirely one of price and timing. The investor who pays peak-mania prices for AI exposure will likely see those prices correct before the genuine transformation arrives. The investor who navigates the correction with capital intact — who can deploy at the distressed prices that follow the speculative excess — is positioned for what may prove to be the most significant investment opportunity of the coming decade.

The specific opportunity in the aftermath of an AI correction is likely to be concentrated in two areas. First, the infrastructure layer — the compute, the energy, the physical assets that AI requires regardless of which applications succeed. These tend to be less speculative than the application layer during the mania and less destroyed during the bust. Second, the companies that used AI as a genuine operational tool to build sustainable competitive advantages — lower cost structures, superior data assets, network effects built on AI capability rather than on the label — and whose valuations reflect the post-correction pessimism rather than the pre-correction enthusiasm.

Identifying those companies, at those prices, requires exactly the kind of framework thinking this series is designed to build. Not pattern-matching on what was expensive in the mania. Not assuming the most beaten-down names are necessarily the most attractive. But applying fundamental analysis — what is the business actually worth, what is the sustainable competitive advantage, what would a rational owner pay for this cash flow — to an environment where the emotional context has shifted from euphoria to despair.

That shift in emotional context, when it arrives, is the opportunity. Recognising it when it comes — and having the pre-committed structure to act on it rather than being paralysed by the prevailing pessimism — is the decision intelligence challenge that will define who benefits from the AI era and who pays for the mania that preceded it.

This piece is one of the case studies the framework is built around — applied to a mania while it is still running, not after the fact. The Decision Intelligence framework does not produce a specific portfolio recommendation from this analysis. What it produces is a set of structured disciplines — things to have in place before conditions change rather than improvised responses to conditions after they have.