AI data centers could become flexible grid assets by slowing non-urgent computing when electricity gets tight. In Texas, data-center power demand far exceeds what the grid can supply: ERCOT received requests totaling 474 gigawatts, with 90 percent from data centers, while the consumption record reached only 91,089 megawatts. An experiment conducted in Texas by Luxor Energy and Bentaus showed that a single Nvidia B200 chip could reduce its power draw to approximately 25 percent of normal capacity within half a second, with no loss of work in progress. According to Lawrence Berkeley National Laboratory, data centers could consume between 9.5 percent and 15.3 percent of US electricity by 2030. Emerald AI, a startup specializing in energy flexibility for data centers, raised 150 million dollars at a valuation exceeding one billion dollars.
Source: Read the original article
