AI’s effect on energy demand may be more nuanced than is reported — and distributed energy and efficiency are part of the reason why, according to a new report by S&P Global.
The report, “AI Won’t Solve Its Own Energy Problem — and That Might Be Fine,” points out four interwoven phenomena that influence AI’s power demand.
First, AI’s appetite for power is clearly growing. By S&P’s reckoning, it is about 19% annually in the US. Virginia’s Dominion Energy alone has 70 GW of pending connection requests, largely tied to data center growth. This figure has tripled since early 2025, according to the report.
Second, data centers are becoming more energy-efficient, which reduces some of that demand. Google, for example, says the energy required for a Gemini prompt fell 33-fold in just one year.
Third, just because AI is becoming more efficient doesn’t necessarily mean it will use less energy. When machines become more efficient, they are cheaper to run. As a result, we end up building more of them. This is called the rebound effect, or Jevons paradox. So efficiency ultimately spurs more energy use, not less. The report cites the rise of efficient steam engines as an example. Lowering the cost of coal-fired generation led to a surge in coal demand.
Yet there is number four—the complex influence of distributed energy resources (DERs). AI’s outsized demand is forcing grid modernization. This leads to more virtual power plants, microgrids, and other DERs that make data centers flexible grid assets. The report argues that flexible load arrangements could ease interconnection bottlenecks. Data centers could participate in grid balancing, rather than simply demanding constant capacity.
AI as a DER accelerant
In other words, DERs allow AI not only to take from the grid but also to give back.
This all makes for a complex supply/demand picture. AI demands more power, becomes more efficient and cheaper, so the industry builds more data centers, which in turn spur greater energy use. Yet, underlying all of this is a grid that works better because AI is modernizing it and providing it with more resources.
And this all gets even more nuanced when you take into consideration that AI’s influence is more regional than national.
“What is not straightforward is whether those savings will materialize fast enough and in the right geographies to keep pace with demand,” the authors write.
Even if AI eventually delivers broad efficiency gains across the economy, those gains may not appear where the largest concentrations of data centers are being built. Northern Virginia, Texas, and other hyperscale markets are already struggling to accommodate load growth.
The report suggests the industry should prepare for sustained load growth while rethinking how data centers interact with the grid.
“Ultimately, the tension between AI’s insatiable energy demand and its increasing efficiency will not be resolved in a simple net-zero equation,” the report says. “Instead, AI is acting as an accelerant, forcing a decade of grid modernization to happen in just a matter of years.”