Sofie Van der Meulen is a former imec researcher who co-founded Neuromorphic Foundries in Leuven in 2058 with fellow imec alumnus Bram Coenraets, building offline, single-instance, locally-hosted neural core hardware against the cognition-cluster rental model. In September 2058, the year of the company’s founding and before any core was offered for sale, she and Coenraets published The Core in the Room, a design and siting guide that was, at the date of publication, the only publicly available specification of what local hosting requires; universities, hospitals, and municipal authorities used it as a planning reference before any hardware existed to plan for.
Van der Meulen’s structural critique of the rental model was direct: “The cluster model is enormously profitable for the companies running it. Renting cognition by the query is a much better business than selling someone a box they only buy once. Efficiency research that threatens the rental model was never going to come from inside the companies built on the rental model. It was always going to have to come from somewhere that didn’t have that business to protect.” The company’s direction was set by two external developments: the Volkonsky et al. (2055) sparse connectomic scaffold paper, in whose efficiency-research tradition she and Coenraets had initially worked, and Kosmograd, whose USTAV system demonstrated that a genuinely narrow, efficient cognitive system could run an entire settlement’s infrastructure on hardware far smaller than anything the commercial cluster operators had shipped.
The company released its architecture and a baseline trained instance under open license in early 2062; within four months a community of modifiers, curriculum-sharers, and independent hardware builders had formed, and Van der Meulen declined to legally contest derivative instances published by third parties: “watching this thing escape the small number of companies that built the rental economy around it is the single most satisfying part of this entire five-year project.” Asked whether a household-scale neural core was realistic in her lifetime, she answered: “Five years ago I’d have told you no, confidently. Two years ago I’d have told you I didn’t know. Today — I’ve stopped betting against this getting smaller.”