“A typical large-scale AI training cluster consumes 50-150 megawatts in a single location, whereas a distributed inference setup might consist of multiple 4-5 megawatt clusters across different data centers.”