Once AI is to be deployed, government entities and enterprises adopt it with care for the environment and for resources: choosing the simplest model that meets functional needs; prioritising model compression, pruning, quantisation and efficient architectures; training with optimised routines, tuned batches, scheduled training windows and hardware that uses fewer resources; drawing on transfer learning, on synthetic data where it fits and on targeted datasets instead of oversized corpora; requiring every AI proposal to assess energy use, compute requirements and carbon footprint; favouring cloud services, data centres and hardware accredited for energy efficiency or powered by renewables; tracking runtime costs, electricity and compute use through the lifecycle; and selecting higher-complexity models only where the performance gain is justified and no frugal option meets accuracy, safety or fairness needs.
The graph holds this control, the 0 it maps to, and the evidence behind each claim, over MCP and REST.