As AI demand strains electricity grids worldwide, power supply—not computer chips—is becoming the scarce resource that will determine which companies profit most from the AI boom.

What happened: Goldman Sachs Research projects global data center power demand will surge up to 165% by 2030 compared to 2023 levels, and hyperscalers are now asking utilities for hundreds of megawatts on three-year timelines. The grid was built for electricity demand growth of 1–2% per year, with decades of warning; utilities are responding that they cannot deliver the power AI companies need.

Why it matters: The pattern mirrors NVIDIA's rise—when the entire economy depends on something scarce, the company controlling it gains pricing power. A single ChatGPT query consumes roughly 10 times the energy of a Google search, and training next-generation large language models requires power equivalent to small cities. Industry forecasts put AI data center capital expenditure at roughly $5.2 trillion（約830兆円） between now and 2030, meaning the shortage of electricity supply could redirect enormous value to whoever controls it.

What to watch: The mismatch between desperate demand and constrained electricity supply is the lever. Unlike chip supply—which NVIDIA largely solved—power delivery cannot be quickly ramped up, creating a structural bottleneck that may persist through the end of the decade.