The backlash came first. Across the United States, massive data-center proposals have increasingly encountered residents who say projects were advanced with little local input and with too little clarity about their effects on power systems and the environment. In Utah, opponents protested a planned facility expected to consume more than twice the state’s 2025 peak electricity demand and expel hot air into a desert valley. States including Tennessee, Florida, Oklahoma and Oregon have also moved to limit the ability of AI companies to shift energy costs onto residents, while research warned that proposed Texas projects could erase decades of air-quality gains within five years.¹

Against that backdrop, OpenAI announced Monday that it would test its “Grid Protectors Program” at a planned campus in Effingham County, Georgia. The company says it could cut demand by as much as one gigawatt during extreme weather or other grid emergencies—roughly the electricity use of 700,000 homes.²

The company’s argument is that the industry must move beyond broad assurances and make curtailment operational: agreeing with utilities on “how much demand can be reduced, how quickly and for how long.”² John McCarrick, OpenAI’s head of energy policy, portrayed the shift as evidence that the sector is retreating from the old assumption that data centers require near-perfect uptime: “I don’t think anybody’s doing that now.”²

OpenAI also contends that steady AI demand can spread grid-infrastructure costs over more electricity sales, while emergency reductions can ease peak strain. Its critics are unlikely to see that as enough on its own. Their objection is not merely to what happens during a crisis, but to the scale of projects built before communities have accepted their everyday costs. The Georgia pilot therefore offers a test of whether flexible operations can rebuild trust—or simply rebrand an unpopular expansion.