If data centers can be flexible and limit their power demands for a handful of hours each year, the U.S. could add some 100 gigawatts of new load without expanding generation, according to a new report from Duke University.
The findings published this week add a wrinkle to the dire warnings that the load from data centers training artificial intelligence models could crash the grid. It’s been forecast that electricity use from data centers could double or triple by 2028. According to the North American Electric Reliability Corp., tech demand could drive up peak load by 20 percent over the next decade.
The grid, however, has “existing headroom,” the study found, since operators try to plan enough generation to meet the hours where demand is at its highest with a small reserve on top. Outside of those few peak periods, that headroom would be sufficient to accommodate constant new loads. That, in turn, means that more data centers could go online without needing to build new fossil fuel plants.
Researchers from Duke’s Nicholas Institute for Energy, Environment and Sustainability modeled a number of curtailment scenarios and found that if data centers limited their load even 0.5 percent of the time — less than 44 hours — the grid could handle 98 GW of new load. That’s roughly equivalent to powering about 70 million homes.
