Cohere — AI supply-chain exposure

Cohere is an enterprise-focused model lab that builds its own large language models — the Command family — and sells them through a platform, North, aimed at companies that want private, deployment-flexible AI. Unlike the consumer-facing labs, its pitch is that a business can run these models inside its own environment rather than sending data to a public API. It is both a producer of the underlying models, through pretraining, and of the enterprise knowledge assistant layer that sits on top of them.

Its structural hook is that it competes not on being the single smartest model but on control, privacy and deployment flexibility — the things regulated and data-sensitive enterprises care about. That positioning is defensible but crowded, and it depends on the same scarce compute and accelerators as every other lab to train competitive models. The model places Cohere mid-chain as a demand-side model maker: a driver of compute consumption whose leverage comes from enterprise distribution rather than raw frontier scale.