AI VALUATIONS ARE DETACHING

Public and private markets have assigned substantial premiums to companies associated with artificial intelligence. In many cases the implied future cash flows require those companies to achieve market positions and profitability levels that have been reached by only a small number of technology businesses in the past.

The discrepancy between current valuations and the historical distribution of outcomes is large enough to matter for capital allocation decisions — the standing question across our investing coverage.

What does technology adoption history actually show?

Previous waves of general-purpose technology produced a small number of companies that captured durable economic rents at scale. Most participants earned returns closer to the cost of capital or below. The pattern holds across personal computing, the internet, and mobile — the dot-com era is the canonical case: the Nasdaq peaked in March 2000 and took fifteen years to reclaim that level, even though the internet itself delivered on its economic promise.

AI is following a similar trajectory so far. A few companies with early advantages in data, distribution, or infrastructure are pulling ahead, a sorting we examined in who profits from the AI buildout. A much larger set of companies are competing in application layers where differentiation is harder to maintain once base capabilities improve.

Valuations that extrapolate the outcomes of the winners to the average participant ignore this historical distribution.

Where are moats weakening?

Many AI application businesses rely on access to general-purpose models that are becoming more widely available. As the performance gap between the best and second-best models narrows — a compression accelerated by open-source models — the ability to charge premium prices or maintain exclusive features declines.

Data advantages can persist, but only when the data is proprietary, frequently updated, and tied to a workflow that is hard to replicate. In categories where data is public or easily collected, the advantage erodes quickly.

Switching costs are also lower than in previous software waves for many use cases. When the core capability is delivered through an API or a replaceable model, users can change providers with less friction than when they were locked into on-premise software or complex integrations. The classic sources of economic moats have to be re-verified case by case, not assumed.

How should this change capital allocation?

The current environment rewards narrative and early revenue traction more than evidence of structural advantage. This creates opportunities for investors who focus on businesses that actually control scarce resources rather than those that merely use generally available technology.

For builders, the implication is that defensibility must be engineered into the business model from the start. Relying on temporary leads in model performance or user interface is unlikely to produce durable returns as the underlying technology diffuses. The same discipline applies in venture capital, where portfolio construction implicitly bets on the shape of the outcome distribution.

For operators inside larger organizations, the relevant question is whether internal AI initiatives are building proprietary data or process advantages or simply consuming generally available capabilities that competitors can also access.

The Bottom Line

AI is a genuine general-purpose technology with large economic potential. The distribution of returns from that potential is likely to be highly skewed, as it has been in prior technology waves.

Valuations that assume a large share of participants will achieve outlier returns are pricing in an outcome that history suggests is improbable. Capital allocated on that basis carries elevated risk of permanent impairment when the gap between narrative and realized economics narrows.