A data centre should be able to give capacity back to the grid

OpenAI announced the national Grid Protectors program on 2 November for its future compute campuses. The idea is to agree in advance with utilities and regulators how quickly, for how long and by how much a site can cut demand when the grid reaches a peak. OpenAI presents its planned Effingham County, Georgia campus as the first model. Georgia Power says the project represents about 3.2 GW of new demand, while a 25-year agreement provides up to 1 GW of controllable demand response. OpenAI has also committed to pay the infrastructure and electricity-service costs required for the project. The campus is still in development: this is a contracted operating model, not evidence from years of production use.

The key question is how the promised flexibility is measured

Controllable demand can help a grid: urgent computing can continue while deferrable jobs, cooling within safe limits or battery charging are rescheduled. But a megawatt ceiling alone guarantees little. A credible program needs a baseline load profile, notice time, required duration, annual call limits, penalties for non-performance and independent verification of the actual reduction. Flexibility also does not create new electricity or replace transmission, generation and storage; OpenAI explicitly says those investments remain necessary. The useful test for future campuses is therefore straightforward: do they publish measurable commitments and results from real demand-response events, rather than only the site's total planned capacity?

Sources