Utility companies and IT firms have been preparing for an energy crisis as artificial intelligence prompts a rise in data center growth. However, a recent study indicates that small changes in power usage could release enormous system capacity, allaying worries about artificial intelligence’s rising electrical needs.

Researchers found that if data centers and other high-energy users reduced their power draw to 90% of peak consumption for just a few hours at a time—amounting to roughly one day per year—they could free up 76 gigawatts of electricity. That amount exceeds the total energy used by all data centers worldwide. Given the rapid expansion of AI-driven computing, this strategy could provide a crucial buffer without requiring immediate large-scale infrastructure investments.

Industries like manufacturing and universities have long participated in similar demand-response programs, temporarily cutting power usage during peak hours in exchange for financial incentives. Unlike these industries, data centers have always shunned such uptime-maintenance strategies. Nevertheless, the paper argues that AI infrastructure offers a unique level of flexibility that allows participation without disrupting regular operations.

Moving non-urgent computer operations, like training AI models, to off-peak hours is one such tactic. Spatial flexibility is another strategy that allows businesses to shift workloads to data centers in areas where demand is lower. Operators can also combine jobs, which enables them to temporarily shut down servers that aren’t in use.

Alternative energy sources offer yet another option for workloads that need to run continuously. These resources could assist lessen grid dependence without compromising performance, since many data centers currently use backup batteries and on-site power generation for emergencies.

The methods have previously been tried by a number of industry experts. With its carbon-conscious computing platform, Google has reduced energy use, and Enel X has assisted data centers in connecting to uninterruptible power sources to sustain the grid. Meanwhile, PG&E has offered faster grid connections to operators willing to participate in demand-response programs.

While these measures won’t eliminate the need for new power generation, they could significantly reduce strain on the grid. Rather than facing an unavoidable energy crisis, the industry could address AI’s rising electricity demands through strategic and well-coordinated solutions.