The cloud taught us to think of computation as weather: ambient, inevitable, owned by no one in particular. But the cloud is just somebody else’s factory. Every prompt sent to a frontier model enters a system of metering, moderation, surveillance, and rent.
A MACHINE YOU CAN TURN OFF
A local model changes the relation. You choose the hardware, the model, the context, the retention policy, and the moment the machine goes dark. This is not perfect autonomy—chips, electricity, weights, and tooling all have supply chains—but it is a meaningful shift in where control sits.
THE CLOUD IS NOT WEATHER. IT IS A FACTORY WITH AN API.
OWNERSHIP IS A VERB
“Seizing the means of production” was never only about holding an object. It was about gaining the practical capacity to make, repair, adapt, and decide. Local inference offers a modest version of that capacity: not universal liberation in a download, but room to experiment beyond an API’s terms of service.
THE SMALL MODEL POLITICS
The most interesting local systems are not miniature attempts to recreate the biggest lab. They are situated tools: a model tuned to a language community, a private archive, an artist’s odd workflow, a cooperative’s shared knowledge. Smaller can mean less universal—and more accountable to an actual place.