The power grid has an AI problem

Read original article

AI data centers are sprouting up across the United States at a rate not seen since fro-yo took over strip malls in the 2000s. And their staggering growth is causing alarm that the country’s power grid doesn’t have the electricity capacity to absorb them without breaking.

At a major energy conference in Houston last week, the No. 1 agenda item—of all things—was artificial intelligence, with leaders including Bill Gates warning that the top constraint to AI development was finding sufficient power sources to feed data centers.

How we got here

Over the past few decades, US demand for electricity has been pretty flat. But the horizontal chart went fully vertical as the excitement around AI set off a land grab for data centers, which are massive buildings housing the computing infrastructure that trains large-language models, like the one behind ChatGPT.

• Earlier this month, Oracle Chairman Larry Ellison boasted that he’s building an AI data center that could fit eight Boeing 747s nose to tail.

The power problem: Airplane hangar-sized data centers training AI systems require “exponentially more power” than traditional data centers, per the Washington Post, causing utilities to dramatically upgrade their demand projections.

• Oregon’s Portland General Electric just doubled its forecast for new electricity demand in the next five years.

• Georgia now predicts 17x more demand for industrial power than recent estimates. “This has created a challenge like we have never seen before,” the chairman of Georgia’s electricity regulator told the WaPo.