US Department of Energy launches Agora to simulate AI data center strain on the grid as power becomes the binding constraint
Every AI buildout now hits the same wall, the power grid: the US Department of Energy launched Agora, a platform that simulates how hyperscale AI campuses strain regional grids, as Gartner estimates global data center electricity demand passes 1,000 TWh this year, double 2023.
The hardest part of building AI capacity in 2026 is no longer the chips. It is the electricity. The US Department of Energy has launched a platform called Agora that simulates how hyperscale AI campuses behave when they connect to regional power grids, including the volatile, high density loads that current AI hardware creates. The scale explains the concern. Gartner estimates global data center electricity demand passes 1,000 TWh this year, roughly double the 2023 level, and a single Nvidia $NVDA GB200 rack can draw more than 120 kW. The practical effect is that the limiting factor for AI expansion is moving from how efficiently the chips run to whether the local grid can physically deliver the power. For utilities, regulators and anyone siting a data center, that is the variable that now decides what gets built and where.