AI TRENDS | OpenAI Pledges to Cut U.S. Data Center Power Use During Extreme Weather

OpenAI has pledged to reduce electricity use at data center campuses it directly builds in the United States when extreme weather or other emergencies strain the power grid. According to Sina Finance, the company’s Grid Protectors Program aims to extend nationwide a commitment made for a planned campus in Effingham County, Georgia, where OpenAI said it could cut up to 1 gigawatt of load when the grid needs relief—roughly equivalent to the electricity use of 700,000 homes. OpenAI said it would explain in a blog post later today that it aims to make this load flexibility a core part of how future data center campuses are built and operated. The company said it would work with utilities to determine the amount of load that can be reduced, how quickly it can respond, and for how long. Plans will vary by campus, with details to be announced once finalized. OpenAI Head of Energy Policy John McCarrick said the initiative reflects changes in industry thinking and technology, as well as adjustments made in response to community feedback. He said that about a year ago, another major cloud provider was still insisting on the “five nines” reliability standard, but that no one is insisting on it now. The term refers to 99.999% continuous uptime for data centers. AI and technology companies, including OpenAI, have pledged to be more grid-friendly and avoid raising electricity bills for residential customers. OpenAI described Monday’s announcement as turning a vision into practical operating rules. The company also said steady AI electricity demand could spread grid infrastructure costs across more users, while flexible load could help reduce peak demand. The Federal Energy Regulatory Commission has asked grid operators to explain or reform the process for connecting large customers, with provisions encouraging load flexibility. A Senate permitting bill would explicitly allow states to make peak-hour power curtailment a condition of data center grid connections.