Tech companies, data center developers, and power utilities are expressing concern over the anticipated increase in electricity demand in the U.S., driven largely by the rapid growth of artificial intelligence (AI). However, a recent paper suggests that these concerns may be mitigated if data center operators and large electricity consumers reduce their power consumption slightly.
The study indicates that by limiting power usage to 90% of the maximum for a few hours a year, it may be possible to unlock up to 76 gigawatts of capacity in the U.S. This amount exceeds the total electricity consumption of all data centers worldwide, accounting for roughly 10% of peak demand in the country. If data centers were to reduce their consumption further, even more capacity could be made available.
Historically, utilities have encouraged significant electricity users—such as shopping malls and factories—to lower their power use during peak demand periods, typically in exchange for billing credits. Despite this, data centers have generally avoided participating in such demand-response programs, prioritizing uptime and performance for their customers. The study suggests that data centers could be well-positioned for demand-response initiatives due to their inherent flexibility.
The research outlines several methods data centers can employ to decrease their power consumption. Temporal flexibility allows operators to shift computing tasks to less busy times, such as rescheduling AI model training. Spatial flexibility involves relocating tasks to regions facing lower demand or consolidating workloads by shutting down certain servers. If essential tasks cannot be rescheduled, operators may also utilize alternative power sources, such as batteries, to bridge any temporary power shortfalls.
Some companies are already testing these strategies. Google has integrated demand-response capabilities into its carbon-aware computing platform, while Enel X collaborates with data centers to use batteries for grid stabilization. PG&E is also incentivizing data centers to engage in demand-response programs to expedite their connection to the grid.
While these adjustments may not entirely eliminate the requirement for new power sources, they could significantly alleviate the challenges posed by increased electricity demand associated with AI.
Source: https://techcrunch.com/2025/02/13/data-center-tweaks-could-unlock-76-gw-of-new-power-capacity-in-the-u-s/