"It's really a win-win-win solution," Josh Parker, Nvidia's head of sustainability, said ahead of the launch.
Data centers typically draw large amounts of electricity continuously. The alliance argues that AI facilities can operate differently by temporarily pausing less urgent computing, moving workloads elsewhere, using batteries or relying on nearby generation when electricity supplies become constrained.
Google already has utility agreements covering 1 gigawatt of demand that can be reduced when necessary, according to Tyler Norris, the company's head of advanced energy market innovation.
Emerald AI is developing software around the same idea. Its technology links utilities with data centers so computing demand can respond to conditions on the grid. The company says these types of flexible operations could help additional data centers connect without requiring every facility to be planned around maximum electricity demand at all times.
The newly formed group is a revival of the Advanced Energy Management Alliance, which began in 2014 around demand-response companies before becoming largely inactive. Emerald AI CEO Varun Sivaram said the relaunched organization is now specifically targeting the electricity needs of AI data centers.
Its policy push will extend to federal and state regulators, utilities and regional grid operators. The alliance wants data centers that can reliably lower their consumption to receive faster or larger grid connections and potentially reduce the need for some infrastructure upgrades.
Sivaram said those benefits should depend on whether promised reductions in power use can actually be demonstrated, describing the required flexibility as "verifiable and enforceable."
The proposal arrives as public concern around data center electricity use continues to rise. An AP-NORC and University of Chicago Energy Policy Institute poll released Wednesday found that 84% of Americans are concerned about the effect data centers could have on local electricity prices.
More than half said they were extremely or very concerned about the environmental effects of AI, compared with 41% a year earlier. The poll also found support across both major political parties for requiring data center developers to cover necessary grid upgrades, including 79% of Democrats and 76% of Republicans.
Those concerns give the alliance another challenge beyond proving the technology works. The participating companies also need to persuade communities that flexible computing can reduce the strain associated with adding large new electricity users.
"Power-flexible data centers are a wonky topic," Sivaram told Axios.
His broader goal is to make data centers into "good grid citizens" by having them lower consumption during periods of stress while contributing revenue to the electricity system.
Demand response itself is not new. Utilities have long paid large industrial customers to reduce electricity use during peak periods. The alliance is trying to adapt that model to AI computing, where some workloads can be delayed or shifted between facilities.
Federal regulators opened another potential path in June by directing regional grid operators to examine new connection arrangements for large customers capable of adjusting their electricity demand.
Whether those rules ultimately give flexible data centers the faster grid access the alliance is seeking will determine how far the model can move beyond individual agreements and into broader AI infrastructure development.
This analysis is based on reporting from Axios.
Image courtesy of Nvidia.
This article was generated with AI assistance and reviewed for accuracy and quality.
