— RICHMOND, Va.The growth of electricity demand from data centers in three of the largest hyperscaler corridors in the United States has now exceeded the most aggressive forecasts utility planners issued just two years ago, and grid operators in those regions are quietly redrawing their long-term capacity plans to accommodate a load curve that few in the industry believed plausible until very recently.

Internal documents from three regional transmission organisations, reviewed for this report, paint a strikingly consistent picture. In northern Virginia, the largest concentration of data centers in the world, expected peak demand has been revised upward by nearly 40 percent for the second half of this decade. In central Ohio and the Phoenix metropolitan area, the revisions are smaller in absolute terms but no less startling on a percentage basis.

"We are not talking about an incremental change," one senior planning engineer at a regional grid operator said, asking not to be named because the discussions are commercially sensitive. "We are talking about a structural break."

The drivers are familiar in outline if not in scale: the buildout of generative AI compute infrastructure, the accelerating digitisation of corporate workflows, and the relentless growth of streaming and consumer applications. Together, they have transformed what was, less than a decade ago, a quietly compounding 1 to 2 percent annual demand curve into a step-function whose precise contours are still being negotiated, contract by contract, between utilities and the largest cloud companies.

The implications stretch well beyond the technology sector. Utilities have begun to slow the retirement of coal-fired generation, accelerate the procurement of natural-gas peakers, and, in a few quietly important cases, sign decades-long power-purchase agreements for the output of nuclear plants. Environmental groups, who not long ago believed they had won the argument over the trajectory of the American power mix, are now relitigating it.