OpenAI launches plan for data center power cuts during grid stress

OpenAI launched Grid Protectors on November 2 to negotiate power reductions at future campuses it develops during grid stress. Utilities would agree locally on the reduction amount, response speed and duration.

The pilot is a planned Effingham County campus. Georgia Power’s filed contract summary specifies 3,210 megawatts of ultimate demand and up to 1,000 megawatts of required reductions when called upon, providing a model for future arrangements.

Georgia Power can curtail demand if the customer fails to reduce it voluntarily. Frequency, duration and magnitude are limited, with values undisclosed in the summary. Full minimum bills remain payable during reductions.

The financial obligations sit alongside that operational flexibility. In its July announcement, Georgia Power said OpenAI would pay the full infrastructure and electric-service costs needed to supply the facility. The agreement runs for 25 years and includes financial assurances intended to protect other customers. The utility described both a long-term customer commitment and the ability to reduce power delivered during high-demand periods. Georgia Power said the approach follows large-load rules first approved by the state Public Service Commission in January 2025, designed to protect existing customers as it adds infrastructure for new demand.

The utility’s customer-protection pledge spells out its general framework for new users requiring at least 100 megawatts. They pay infrastructure costs, sign 15-to-25-year contracts, pay monthly minimums even without consuming power, post collateral and owe substantial early-termination payments. These are Georgia Power’s large-customer protections; the national program still leaves individual campus terms to negotiation.

The savings figures cited in the current announcement belong to a broader utility forecast. On August 26, Georgia Power projected approximately $950 million in annual customer savings starting in 2029 from its portfolio of large-load agreements and expected growth. It estimated at least $15 a month for a typical household using 1,000 kilowatt-hours monthly. Those figures describe an expected benefit across the portfolio, with a specified usage benchmark and future start date. The company’s projection spans 2029–2031 and includes other announced large-load projects and expected future growth, alongside the OpenAI agreement approved that August.

The project FAQ puts phased development in 2028–2032, with initial capacity potentially online in 2028. Final scale and timing depend on permits, infrastructure, financing, community engagement and construction. These are future development milestones for the planned 3.2-gigawatt campus.

OpenAI leads design and development and is competitively selecting construction and operating partners. The FAQ also promises public power-use data, with those operating arrangements still to be finalized.

The project overview promises an annual audit by an independent firm, with the results released publicly, to check compliance with its commitments. The assurance sits alongside the project’s community and operating promises. The overview describes a future accountability process; it supplies no completed annual audit results or measured demand-response performance from an operating campus.

Further campus commitments will be disclosed as finalized. OpenAI announced no uniform national reduction capacity or implementation date.