Artificial Intelligence

General-purpose technologies pay out on a lag. Their value is realised not through deployment in isolated applications but through the systematic reorganisation of the processes and knowledge flows that determine what an organisation can do.

It helps to think in layers: compute, models, tooling, applications, and finally organisational redesign. Capital today is concentrated at the top of that stack, in the picks-and-shovels of compute and foundation models. But the durable surplus accrues lower down, where a firm re-architects how it actually works around a new marginal cost of cognition. Value migrates down the stack over time; price has not.

The electric dynamo was commercially viable by the 1880s, and Edison's Pearl Street station opened in 1882. Yet electric motors provided less than 5% of mechanical drive power in US factories as late as 1899, and the productivity gains did not arrive until the 1920s, roughly a forty-year lag.

The delay was not technological. Early factories simply kept their centralised line-shaft and belts and swapped the steam engine for a dynamo. The gains came only when owners adopted “unit drive”, a motor on each machine, and re-laid the factory floor around it.

Most AI today is the equivalent of swapping the engine: bolting copilots onto workflows designed for humans. The re-rating comes with unit drive, redesigning the organisation around cheap cognition, and that requires co-invented, intangible investment in process, data and human capital that near-term earnings and national statistics both miss.

Consensus overpays for the obvious beneficiary and underprices the co-invention lag. The market thinks the winners are those who spend the most on AI; we think the winners are those who compress their cost-to-serve and can defend the saving.

So we look for three things: incumbents whose moats deepen as cognition gets cheap; the second-order beneficiaries of a margin shift from labour to capital; and, most of all, the gap between a company's AI narrative and its AI operating leverage in the actual numbers.