Melton Chang, Schneider Electric Power Systems | CASPA CEO Summit 2026
summary
GPU and semiconductor generations turn over in ~1 year while utility power cycles run 6–10 years, creating a mismatch that forces power to go distributed as AI moves to edge and inference. AI loads swing from 30% to 150% in milliseconds, causing voltage, frequency and resonance failures. A 1 GW build puts electrical at 5–10% of cost yet a one‑month delay can cost $600–700M. Solutions: digital twins, simulation modeling, prefabrication, mass procurement and software‑defined wiring; skilled labor remains the chokepoint.
AI-written summary · about
