Compute limits re-emerge

Artificial intelligence companies are running into a simpler limit than software: not enough chips, power and data-center space to meet demand. (cnbc.com) Nvidia shares rose more than 18% over 10 trading days through April 14, 2026, as investors bet demand for its artificial intelligence chips would stay strong despite broader market swings. Nvidia reported $62.3 billion in quarterly data-center revenue on February 25, up 75% from a year earlier. (cnbc.com) (nvidianews.nvidia.com) The pressure is no longer just about buying processors. Enterprise AI, citing a Wall Street Journal report on April 15, said companies have started rationing access and raising prices as “agentic” tools that can write code or complete tasks drive a new jump in computing demand. (economictimes.indiatimes.com) A data center is the warehouse where these systems run: rows of servers, networking gear and cooling equipment that turn electricity into computing work. The International Energy Agency said global electricity used to supply data centers is projected to rise from 460 terawatt-hours in 2024 to more than 1,000 terawatt-hours in 2030 in its base case. (iea.org) That strain is already showing up in the grid. In Northern Virginia, the world’s largest data-center market, Dominion Energy has warned that proposed facilities could require more electricity than the utility can deliver before the end of the decade. (msn.com) Virginia’s own numbers point the same way. Virginia Business reported on March 31 that Dominion now sees electricity demand growing 5% to 6% a year, up from roughly 1% annual growth projected five years ago, with data centers driving much of the increase. (virginiabusiness.com) The bottleneck has shifted over the past year. In 2023 and 2024, the main question was whether cloud providers could get enough high-end graphics processing units; in 2026, the question is also whether they can find substations, transmission lines, transformers and enough cooling to run them. (cnbc.com) (iea.org) Utilities and energy analysts are now putting hard numbers on that build-out. S&P Global, citing the International Energy Agency on April 10, reported that data-center power use could reach 945 terawatt-hours by 2030, roughly equal to Japan’s current total electricity consumption. (spglobal.com) That leaves technology companies with a narrower set of choices in 2026: pay more for scarce computing capacity, delay product rollouts, or sign longer-term infrastructure deals to lock in supply. The boom in artificial intelligence has not outrun the old constraints of steel, copper and electrons. (economictimes.indiatimes.com) (iea.org)