Frugality aims—while maintaining performance and meeting needs—to minimize the environmental footprint of a system or an application, and therefore to minimize the energy consumed and the hardware used.
Frugality will impose itself as an imperative for Artificial Intelligence not only through constraints on energy resources and environmental impact, but also—and this is linked—quite simply for economic reasons.
Large LLM models, most often developed by digital giants, do exist but are no longer the decisive strategic stake. The strategic battle is now being fought on inference, through smaller and specialized models depending on applications or uses—in other words, business domains—including for example connected objects or embedded systems.
Frugality, unavoidable, can only result from a dual-track “hardware/software” co-design, and even a triple-track approach when human resources are included: their intelligence and their knowledge not only of technologies but also of business domains—in short, their very real intelligence is a primary necessity.
Whether it is a matter of companies’ economic survival, national sovereignty, or European eco-geostrategy, AI frugality is not only an ecological choice; it is the energy foundation of any competitiveness in the century of AI.
Actors will have no choice. They will be frugal or they will be overtaken: they must compel themselves to define their frugality trajectory, continuously evolving,in the form of a roadmap—i.e., a trajectory.