You don’t need me to tell you how artificial intelligence (AI) is impacting the power grid; you can just ask AI. Claude, an AI assistant created by Anthropic, told POWER, “AI training and inference are driving unprecedented demand for data center capacity, particularly due to large language models and other compute-intensive AI workloads.” It also said, “AI servers, especially those with multiple GPUs [graphics processing units], require significantly more power per rack than traditional servers—often 2–4x higher power density.”
So, what does that mean for power grid operators and electricity suppliers? Claude said there could be several effects, including local grid strain in AI hub regions, the need for upgraded transmission infrastructure, higher baseline power consumption, and potential grid stability issues in peak usage periods. Notably, it said AI data centers tend to cluster in specific regions with favorable power costs and regulations, creating “hotspots” of extreme power demand.
Sheldon Kimber, founder and CEO of Intersect Power, a clean energy company that develops, owns, and operates a base portfolio of 2.2 GW of operating solar PV and 2.4 GWh of storage in operation or construction, understands the challenges data centers present for the grid. As a guest on The POWER Podcast, Kimber suggested the only way to meet the massive increase in power demand coming from data centers is with scalable behind-the-meter solutions.
“These assets may still touch the grid—they may still have some reliance on the grid—but they’re going to have to bring with them an enormous amount of behind-the-meter generation and storage and other things to make sure that they are flexible enough that the grid can integrate them without creating such a strain on the grid, on rate payers, and on the utilities that service them,” Kimber said.
Yet, data center developers have not traditionally kept power top-of-mind. “The data center market to date has been more of a real estate development game,” Kimber explained. “How close to a labor pool are you? What does it look like on the fiber side? What does the land look like?” He said electric power service was certainly part of the equation, but it was more like part of a “balanced breakfast of real estate criteria,” rather than a top priority for siting a data center.
In today’s environment, that needs to change. Kimber said Intersect Power has been talking to data center companies for at least three years, pitching them on the idea of siting data centers behind-the-meter at some of his projects. The response has been lukewarm at best. Most of the companies want to keep their data centers in already well-established hubs, such as in northern Virginia; Santa Clara, California; or the Columbia River Gorge region in Oregon, for example. Kimber’s comeback has been, “Tell us when you’re ready to site for ‘Power First.’ ”
What “Power First” means is simple. Start with power, and the availability of power, as the first criteria, and screen out all the sites that don’t have power. “To date, data center development that was not ‘Power First’ has really been focused on: ‘What does the plug look like?’ ” Kimber said. In other words: How is the developer connecting the data center to the power grid—or plugging in? The developers basically assumed that if they could get connected to the grid, the local utility would find a way to supply the electricity needed.
However, it’s getting harder and harder for utilities to provide what developers are asking for. “The realization that the grid just isn’t going to be able to provide power in most of the places that people want it is now causing a lot of data center customers to re-evaluate the need to move from where they are. And when they’re making those moves, obviously, the first thing that’s coming to mind is: ‘Well, if I’m going to have to move anyway, I might as well move to where the binding constraint, which is power, is no longer a constraint,’ ” he said.