OpenAI Pledges to Cut Data Centre Power Use During Grid Strain, Starting in Georgia

The company announced the programme on Monday, November 2. OpenAI says it will agree with each utility how much load a campus can give up, how quickly, and for how long, and that the terms will differ from site to site. Some campuses may also feed electricity back to the grid from on-site generators or batteries. "Our goal is to make campus flexibility a resource that utilities can call on when communities need it most," the company said.

The programme is OpenAI's answer to a growing political problem for the AI build-out. Rising household power bills have stalled data centre projects across the US, and opposition to new facilities was a campaign issue in this week's midterm elections. OpenAI is arguing that a campus able to throttle back during heat waves and cold snaps can help utilities, not just add to their costs. "Families and small businesses should not be asked to subsidize a compute campus's electricity needs," the company said.

The Georgia Pilot

The Effingham County site, known as Project Camellia, runs under a 25-year power agreement with Georgia Power that was approved in August. Under that deal the utility can cut up to 1,000 megawatts from the campus at times of peak demand. OpenAI says that is about the average demand of 700,000 homes, and Tech Times calculated it at roughly 31 per cent of the campus's peak load.

OpenAI says it will pay the full cost of the power lines and service the campus needs, as well as its baseline electricity use. According to The Next Web, Georgia Power says the flexibility means it can build less new generation, and the utility expects agreements with large customers such as OpenAI to save its other customers about $950 million a year from 2029.

John McCarrick, OpenAI's head of energy policy, said: "The initiative reflects a shift in how the industry thinks about reliability."

What Flexibility Can and Cannot Do

OpenAI's case rests partly on recent trials. In a five-day test in December 2025, National Grid worked with Emerald AI, the research group EPRI, cloud provider Nebius and Nvidia on a GPU cluster near London. The cluster ran more than 200 simulated grid emergencies. In results published in March 2026, it cut power use by up to 40 per cent. In the fastest scenario, it shed 30 per cent of its load in under 40 seconds, and it held a 10 per cent reduction for 10 hours in a row. Emerald AI's software keeps latency-sensitive inference running while it pauses or slows deferrable work such as batch training.

OpenAI acknowledges the approach has limits. The company said campus flexibility "cannot replace investment in new energy supply and transmission." The commitment covers campuses OpenAI develops directly, which leaves out computing capacity it rents from cloud partners. OpenAI has not named any site beyond Georgia, and it has not said how much load it would commit to cutting elsewhere.

Other companies are working on the same problem. Google and Nvidia have formed their own alliance to make data centres more flexible in their power use, and the White House has asked builders to sign a voluntary pledge to cover their own power costs.

The Effingham campus is still being planned. Utilities, regulators and local communities will get their first real test of OpenAI's promise when Georgia Power calls on that 1,000 megawatts during a demand peak.