3 articles
New York's data center moratorium, Utah's forced downsizing, and capital fleeing to India and orbit show permitting, not GPUs, now gates AI scale.
Sparse models, MoE, quantisation, and better local runtimes have changed what is possible on modest hardware. Local AI is no longer only for people running dual RTX 3090 rigs — but it is still very much for developers who can supervise the machine when it starts confidently sawing through the floorboards.
AWS swapped hierarchical fat-tree fabrics for quasi-random flat meshes with passive optical ShuffleBoxes, cutting routers 69% — and the logic will spread.
