The Surprising Contribution of Distributed Energy in Offsetting New AI Demand

The news is full of stories about artificial intelligence (AI) and data centers overwhelming the U.S. electricity grid. And there are mixed reports about how much power we will need to feed the AI beast. An under emphasized part of the answer is a build out of distributed energy resources. This could ease the urgency to build transmission that can be nigh impossible to site and big power plants such as natural gas or nuclear, which are too expensive, slow to build and/or heavy on pollution.  A recent study by Duke University called Rethinking Load Growth: Assessing the Potential for Integration of Large Flexible Loads in US Power Systems emphasized the need for caution to those who want to spend close to a trillion to address growing demand. 

Distributed energy resources (DERs), such as commercial and industrial solar, can help mitigate the increasing electricity demand from AI data centers by providing on-site power generation, reducing reliance on the grid, and enabling flexible load management. DERs can also be aggregated into virtual power plants (VPPs) to provide grid services and balance electricity supply and demand.  An example of how flexibility supplied by DERs can be leveraged by data centers: AI training workloads, which are a large part of AI data center energy consumption, are deferrable and can be scheduled for times when renewable energy is abundant. Data centers can leverage on-site solar to power these tasks during peak solar production hours, reducing the need for grid power at peak times. Similarly, data centers can participate in demand response programs, reducing their electricity consumption during peak periods to provide grid services. They can use on-site generation to on-site generation to support their power needs during demand response events, helping to keep the grid stable.  All of this also works vice versa, i.e. DERs can serve population centers at peak periods, freeing up more power for data centers. 

Goldman Sachs estimates that $720 billion of grid spending through 2030 may be needed to address new AI power demand in the US alone. The Duke report points to efficiency improvements and saying we don’t have time for large-scale infrastructure projects. 

At New Columbia Solar, we see an increase in deployment of community and commercial and industrial solar and all rooftop solar as offering nearer term, lower cost ways to mitigate local power supply constraints and transmission logjams. Collectively these solar plants could alleviate pressure on the grid, reduce local demand, limit the need for massive additional infrastructure investment and provide clean energy rather than keeping coal plants around or building new natural gas. Crucially, these resources are the fastest to put onto the grid, compared with nuclear plants that require the longest development timeline of any resource, taking a decade or more to fully develop. 

These distributed energy assets are exactly the kinds of projects we finance, develop and build. Obviously, no one company will build enough solar or other distributed energy to meet new power demand from AI and data centers. But together companies like New Columbia Solar can meet significant demand and ease the strain on the grid.  According to the U.S. Department of Energy, distributed energy resources, including solar and battery storage, could reach between 80 and 160 gigawatts by 2030. This would account for 10% to 20% of peak electricity demand.  That’s a huge portion of what researchers say is needed to meet growing power demand. The 2024 Long-Term Reliability Assessment of the North American Electric Reliability Corporation says aggregated US winter peak load is forecasted to grow by 21.5% over the next decade, rising from approximately 694 GW in 2024 to 843 GW by 2034. Data centers will account for the largest growth segment.  

The Duke report urges caution on overbuilding transmission lines and old baseload capacity.  “The promise is that the unique profile of AI data centers can facilitate more flexible operations, supported by ongoing advancements in distributed energy resources,” Duke researchers say. “The sheer volume of required new generation, transmission, and distribution capacity forecasted for the United States within a condensed timeframe appears unprecedented,” the report says. “The immensity of the challenge underscores the importance of deploying every available tool, especially those that can more swiftly, affordably, and sustainably integrate large loads.”