As artificial intelligence (AI) technology advances at lightning speed, the demand for electricity in the United States is rapidly rising. Tech companies, data center developers, and power utilities are increasingly concerned that this surge in demand could overwhelm the electrical grid. However, a new research paper offers a promising solution that could ease the strain without compromising progress.

Unlocking 76 Gigawatts Through Modest Curtailment

The recent study suggests that if data centers and other heavy electricity users slightly reduce their power consumption — even just 10% of their maximum for a few hours a year — the United States could unlock as much as 76 gigawatts of grid capacity. This amount exceeds the total electricity used by all data centers worldwide and represents approximately 10% of peak U.S. electricity demand.

Remarkably, these changes don’t require massive infrastructure investments. Instead, minor behavioral and operational adjustments by data center operators could generate huge capacity gains for the energy grid.

The History and Potential of Demand Response

The concept of demand response — where big electricity users voluntarily reduce their consumption during peak times in exchange for incentives — isn’t new. For decades, utilities have partnered with shopping malls, factories, universities, and other institutions to help stabilize the grid during high-demand periods, such as hot summer afternoons.

Participants in these programs receive billing credits in return for temporarily decreasing their usage. Typical actions include turning down air conditioning systems or pausing energy-intensive machines. Despite the proven success of this approach, data centers have historically opted out, focusing instead on delivering 24/7 performance and uptime for their clients.

Why Data Centers Are Ripe for Demand Response Integration

This new research argues that data centers may be ideally suited for demand response programs. Their operations offer both temporal and spatial flexibility, two key attributes that allow energy use to be curtailed without degrading performance.

Temporal Flexibility: Rescheduling Low-Priority Tasks

One major opportunity lies in rescheduling tasks to periods of lower grid demand. For example, AI model training is a huge energy consumer, but it doesn’t typically need to happen at specific times. By shifting these tasks by a few hours, data centers could significantly reduce their demand during peak periods without sacrificing results.

Spatial Flexibility: Shifting Workloads Across Regions

Data centers can also balance loads by moving computing tasks to other locations where demand is lower. For instance, an East Coast data center might transfer workloads to facilities in the Midwest during a heat wave. Load balancing and task migration capabilities inherent in modern cloud infrastructure make this sort of regional flexibility increasingly viable.

Supplemental Power: Using Batteries and Alternative Energy

When certain tasks cannot be delayed or shifted, data centers can turn to alternative energy sources to maintain operations during curtailment periods. Batteries, in particular, are attractive because even modest energy storage systems can provide several hours of backup power almost instantaneously. These solutions not only enhance resilience but also enable participation in demand response programs without service disruption.

Real-World Examples: Pioneers Lead the Way

Some big players have already started experimenting with demand response strategies tailored to data centers. Google, for example, has developed a carbon-aware computing platform that was initially designed to reduce carbon emissions. This platform now allows the company to dynamically adjust data center operations based on grid demand and carbon intensity, improving flexibility while supporting the grid.

Another example is Enel X, which works with data centers to tap into batteries in their uninterruptible power supply (UPS) systems. These batteries, typically reserved for emergency backup, are being repurposed to help balance the grid during peak times.

Additionally, Pacific Gas and Electric (PG&E) recently introduced a program that accelerates grid connection for data centers that agree to participate in demand response efforts. These types of incentives align energy priorities with business goals, making participation even more appealing.

Managing the AI Power Boom Without a Crisis

While these strategies won’t completely eliminate the need for new power generation as AI technologies continue evolving, they can significantly reduce the risk of grid overload. If left unaddressed, the spike in AI workloads could lead to catastrophic power shortages, potentially rendering half of all new AI servers ineffective.

Instead, through smart energy management policies, innovative technologies, and active industry participation, the United States can accommodate rapid AI growth without sacrificing service reliability or requiring massive infrastructure overhauls. By embracing even limited curtailments and flexible adjustments, data centers can not only support the grid, but also gain cost advantages and bolster their sustainability goals.