Study highlights potential savings

Up to 150 billion dollars could be saved in ten years through adjustments in the electricity consumption of data centers, according to a study by the Nicholas Institute at Duke University.

New alliance formed

Google, Nvidia, and the startup Emerald AI announced on September 16 the creation of the AI Energy Management Alliance (AEMA), an initiative aimed at making the electricity consumption of data centers more flexible.

Data center energy use

In the United States, data centers consumed between 177 and 192 TWh in 2024, or 4 to 5 percent of the country's electricity. According to the Electric Power Research Institute (EPRI), this share could rise between 9 percent and 17 percent by 2030.

Flexibility methods

The principle of flexibility involves modulating the consumption of data centers during peak demand on the grid. Several levers exist: postponing non-urgent information tasks to a time when the grid is less strained, transferring calculations to another site located in a less tense area, drawing from batteries installed on site rather than from the grid, or adjusting the operation of cooling systems.

Nvidia's test site

Nvidia is already testing this approach in Virginia, on Aurora, a 96 megawatt site built by Digital Realty. The Emerald AI software there shifts or transfers the most power-hungry artificial intelligence calculations when the grid is strained, before restarting them once demand decreases.

Potential for reduction

According to Arin Kaye, technical director at EPRI, data centers surveyed by the institute estimate they could reduce their peak power demand by 10 to 30 percent depending on the type of installation, a margin even higher for some hyperscalers. He stated this on October 8, 2025, in a statement reported by Reuters.

Current limitations

These reductions are currently limited to pilot projects or case-by-case agreements. OpenAI, on the other hand, has committed to reducing up to 1 gigawatt of power drawn from the grid during periods of tension, out of the 3.2 gigawatts planned for a future site in Georgia.

Financial implications

The financial stakes are significant, with the study by the Nicholas Institute at Duke University estimating that this flexibility could save between 40 and 150 billion dollars in investments in electrical infrastructure over the next decade, whether it be new production capacities, strengthening of lines, or expansion of substations.

Benefits for consumers

Researchers from Duke specify that these savings would especially avoid households and small businesses from paying part of the bill. David Porter, vice president of EPRI, summarizes the idea in an interview with Axios in October 2025: "It's like taking semi-trucks off the road during peak hours."

Government response

American authorities have taken note of the movement. In October 2025, the Secretary of Energy Chris Wright asked the federal energy regulator, the FERC, to accelerate the connection of large consumers, with an expedited process for those willing to reduce their power on demand.

FERC actions

As early as June 18, the FERC had already ordered the six major regional grid managers to review their connection rules, paving the way for faster connections for flexible installations. It also required, the same month, that each manager explain how it plans to ensure enough capacity to supply large consumers, both old and new.

Challenges in implementation

However, not all information tasks can be postponed without consequences. Operators must determine which loads can be adjusted, and to what extent, without degrading performance or disrupting customer services. Large-scale deployment also requires investments in control software, consumption management equipment, and, depending on needs, storage batteries.

Need for incentives

Grid managers and energy suppliers will have to, on their part, offer appropriate tariffs or compensation in exchange for power reductions.