The next infrastructure layer is cognitive

The history of computing reads as a stacking of infrastructure layers, each invisible once laid, each indispensable to everything built above. Compute first — from mainframe to cloud, power became a commodity rented by the hour. Storage next — data ceased to be a problem of room and became an asset one queries. Network — every machine linked to all the others, transport abolished as friction. Then transactional trust — certificates, encryption, identity — which let the economy circulate over those pipes.

Each layer followed the same trajectory: first a feat, then a product, then an obviousness no one mentions any more — the final fate of any successful infrastructure is invisibility. And each layer answered the same question: what, at this moment in history, has become too critical to remain artisanal?

The question of our decade is no longer to produce intelligence. It is to be able to trust it.

For here is where we stand: machine intelligence has become abundant. Herbert Simon anticipated it as early as 1971 — the abundance of information creates the scarcity of attention; the abundance of generation today creates the scarcity of a still more precious resource: reliability. Any organization can produce a thousand pages of analysis in an hour. None can say, page by page, claim by claim, what in them is true, where it comes from, and whether the verdict will hold in three years before an auditor. Reasoning has become too critical — it enters diagnoses, contracts, regulated decisions — to remain artisanal in its verification.

This is the exact definition of an infrastructure layer waiting to be born: the cognitive layer — the one that preserves knowledge with its provenance, qualifies every claim that crosses it, traces every reasoning that relies on it, and makes the whole replayable. Neither a productivity tool nor one more model: what stands beneath the models, as the network stands beneath the applications. It already has the three canonical properties: invisible in use (one sees the verdict, not the mechanism), continuous (it operates at every exchange, not on demand), and increasing-returns — every preserved piece of knowledge and every rendered verification raises the value of all the next.

Christensen taught a generation of investors to spot disruptions. But infrastructure layers do not disrupt — they found. They do not replace the incumbents: they become what everyone, disruptors and incumbents alike, is obliged to build upon. Compute had its giants, storage its own, network and transactional trust theirs. The cognitive layer will have its own — and they are being decided now, while most are still watching the models.