When megawatts ration the buildout, market share goes to the most tokens per watt
With power, land, and deployment — not chip supply — the binding constraints, hyperscalers now select compute by useful intelligence per watt, and vendors compete by codifying efficiency metrics, benchmarking tokens per megawatt, and opening their designs so rival systems still get built around their stack.
The same conclusion keeps arriving from across the workspace's research — 1 topics independently instantiate this theme. Filter the evidence by where it came from:
With megawatts the binding constraint, the largest hyperscaler redefines its success metric as useful intelligence per watt — and custom silicon is the route it pursues it by.
With megawatts the binding constraint, the largest buyer codifies useful intelligence per watt as an industry-standard selection metric — the efficiency yardstick, not raw GPU counts, now decides vendor share.
With megawatts the binding constraint, the vendor competes by codifying data-moved-per-watt efficiency into open industry specs rather than proprietary benchmarks.
The vendor itself states the law: Nvidia's paradoxically rising share is credited to efficiency-per-watt selection under power constraint, with dollar content per gigawatt more than doubling as it sells the full AI factory.
The customer's substitution challenge is priced entirely in tokens per megawatt against unnamed GPU baselines — share contested on watts, not raw capability.