Markets Have Never Known What to Do With a Revolution in Progress

There is a particular kind of vertigo that grips financial markets when a genuinely transformative technology arrives. Not the ordinary uncertainty of a new product cycle or a shifting interest rate environment, but something deeper: a fundamental confusion about what the rules of valuation even are anymore. Artificial intelligence is producing exactly that sensation right now, and the disorientation is not a bug in how investors are behaving. It is, historically speaking, entirely predictable.

The core problem is that markets are extraordinarily good at pricing things that resemble things they have priced before. A new airline, a new bank, a new retailer: analysts have frameworks, comparables, and decades of margin data to draw on. A technology that may restructure the economics of nearly every industry simultaneously offers none of that comfort. The honest answer to "what is this worth" is "we genuinely do not know yet," and markets are institutionally allergic to that answer.

This is not the first time the financial system has found itself in this position. The arrival of the railroad, the electrification of industry, and the commercialization of the internet all produced versions of the same phenomenon: a period of wild mispricing in both directions, followed eventually by a more sober reckoning with which companies actually captured value and which ones simply rode the excitement. The internet bubble is the most recent and vivid example. The underlying technology did transform the world, spectacularly so, but the majority of companies that attracted capital in the late 1990s did not survive to participate in that transformation. The technology won; most of the investors did not.

What makes AI particularly difficult to price is that the signals investors normally rely on are arriving scrambled. Revenue is growing at some AI-adjacent companies, but it is often unclear whether that growth reflects genuine productivity gains being passed through the economy or simply the initial wave of enterprise experimentation, the corporate equivalent of buying a gym membership in January. Capital expenditure on AI infrastructure is enormous and accelerating, which could mean that the technology is delivering returns that justify the investment, or it could mean that a competitive panic among large technology companies is producing spending that will look, in retrospect, like overbuilding.

The feedback loop here is worth examining carefully. When major cloud providers and chipmakers report strong earnings tied to AI demand, it validates further investment in AI infrastructure, which produces further demand for chips and cloud capacity, which produces further strong earnings. This loop can run for quite a long time before the market receives clear information about whether the underlying enterprise customers deploying all this infrastructure are actually becoming more productive. By the time that signal arrives, capital will have been allocated on the assumption that it would be positive.

There is also a second-order effect that receives less attention than it deserves. As investors struggle to price AI companies directly, they are also repricing everything else in the economy through the lens of AI exposure. Companies in sectors from healthcare to logistics to legal services are being evaluated partly on the question of whether AI will compress their margins or expand them, and partly on whether they will be disruptors or the disrupted. This shadow pricing is happening with even less information than the direct AI investment decisions, because it requires not just a view on what AI can do, but a view on how quickly it will be adopted across specific industries with specific regulatory environments and specific labor dynamics.

Historical analogies suggest that the resolution to this uncertainty takes longer than markets want it to. Electrification of American industry began in earnest in the 1880s but the productivity gains it enabled did not show up clearly in economic data until the 1920s. The lag was not because the technology was failing; it was because organizations needed time to restructure themselves around the new capability. A factory that simply replaced a steam engine with an electric motor captured modest gains. A factory redesigned from the ground up around the flexibility that electric power enabled captured transformative ones. The technology's value was real, but it was locked inside a process of organizational learning that could not be rushed.

AI may follow a similar arc, which would mean that investors are currently trying to price a revolution whose most significant economic consequences are still embedded in a future that organizations have not yet learned how to inhabit. The companies that will ultimately capture the most value from this technology may not be the ones attracting the most capital today. They may not even exist yet. That is an uncomfortable thought for a market that prices things every millisecond, but it is probably the most honest framing available.