In July 2026, a transmission line fault in Ashburn, Virginia – an area known as Data Center Alley because it is the most data-center-dense place in the world – prompted data centers to switch to their emergency backup power, resulting in an almost instantaneous drop of more than 3 GW in grid load.

For people working with data center clients like Wannie Park, the only surprise was that events like this aren’t more common in the United States. “Technology is being built on the fly, especially in the software and services side. There’s going to be growth hiccups,” he said.

Park’s startup Pado AI makes grid orchestration software designed to help data centers become flexible loads while making the most of their white space onsite and their compute per megawatt. Its technology can be applied to both greenfield and brownfield sites, including smaller legacy data centers (30 MW to 50 MW) of which there are thousands in the United States. These players don’t have fat checkbooks like the hyperscalers, but they do have existing grid connections, said Park.

Pado AI spun out of South Korean battery company LG a little over a year ago and is shipping its own vendor-agnostic product. Park is eyeing neocloud opportunities, and he has noticed an industry-wide shift towards inference tasks, which use less energy.

Based on tasks, energy demand and renewable resources on the grid, Pado AI’s technology creates a forecast of how an individual data center interacts with the grid, which utilities can use to create flexibility measures.

Source link