AI is the new gold, but you can’t mine it without energy. And that could offer an edge to retail investors eager to buy into the trend. Many investors are now leery of the high share valuations of AI-focused stocks like Nvidia or the so-called hyperscalers—the large cloud-services providers like Amazon, Microsoft, and Google that manage the computing power used by AI models. But companies across Big Tech’s energy pipeline could turn out to be more attractive buys. 

Utilities have never been sexy, and certainly don’t offer the hockey-stick growth coveted by AI investors, though most pay decent dividends. The S&P 500 Utilities Index has delivered annualized total returns of 6.1% over the past five years, with about half of that coming from dividends; the broader S&P 500 has gained more than 15% a year over that stretch. 

But the AI gold rush is turning the otherwise staid sector on its head. The reason is simple: The large language models employed by AI need massive quantities of electricity to train their systems, with data centers set up with special processors for the task. Every time you type a novel question into ChatGPT or one of its rivals, the process of answering, called “inference,” also requires a power surge. According to Goldman Sachs analysts, AI will grow data-center power demand by 160% by 2030. “Data-center expansion is a generational growth opportunity right now for utilities,” Travis Miller, an energy and utilities strategist for Morningstar, tells Fortune. 

To accommodate that boom, most utilities are going to have to invest considerably in infrastructure, from energy storage to grid expansion, at a time when many are already coping with overexerted grids. Some have imposed new rules to adapt to the demands of new technologies. Many Bitcoin miners, for example, have adopted a practice called curtailment, where they turn off mining during peak times. 

As Dan Thompson, a principal analyst at S&P Global, explains, data centers don’t have the same luxury, especially when it comes to inference demands. “This is very real-time, and it can’t be curtailed,” he says. And the demand is just going to grow as data centers replace rows of older chips with AI-focused GPUs: “It’s what has everybody concerned, excited, interested—whatever word you want to use.” 

While everyone’s attention has been on AI, energy demands are up across the board. Andrew Obin, a managing director of equity research at Bank of America, predicts that total U.S. electrical load will go up 2.8% annually through 2030, after rising just 0.4% a year from 2013 to 2023. While he says that data centers are an important source of growth, he also points to electric-vehicle adoption and “reshoring” of manufacturing. “AI is getting a lot of attention because it’s the shiniest thing out there,” he says.

For utility companies, new demand will hardly equate to pure profit. Everything depends on whether they have the cash on hand to invest in infrastructure development. Large-cap companies are best positioned for this, with Miller pointing to operations like Southern Co., Duke Energy, and Entergy. Two big utilities—Vistra and Constellation—have seen their prices rise comparably to Nvidia’s in 2024, boosted by their focus on clean energy, which many tech companies and regulators are requiring wherever possible.

“The more challenging side will be for some midsize utilities,” Obin says. One example is NiSource, which is headquartered in Indiana. The company recently divulged that it has had over 30 inquiries for new data centers over the past year—but that adding just two or three data centers by 2035 could double the load on its system. Utilities that can’t manage the stresses of such increased demand will face outages—and potential share-price hits.

Utilities can always solve demand spikes by pumping in more gas to their power plants. And that means so-called midstream companies, which focus on the storage and transport of petroleum products, could play a particularly crucial role in the AI boom. Stephen Ellis, an energy strategist at Morningstar, says that midstream companies, like utilities, are facing infrastructure challenges. They’re adapting to serve data centers by rerouting gas to new destinations and building new pipelines. 

Two companies that Ellis flags as well-positioned include the Texas-based Energy Transfer, which has been focused on connecting its pipelines to utilities, and TC Energy, a Canadian firm working on a gas pipeline to northern Virginia, which has one of the world’s highest concentrations of data centers.

And then there are data-center companies themselves, along with the electrical equipment providers, HVAC firms, and software outfits serving them. Obin says those companies might be in the best position to absorb the windfall. As the market booms, they can price their gear more profitably. Obin says that charging extra for critical pieces of equipment, like transformers and switch gear, is a “pretty widespread practice” already. Two companies poised to benefit include Eaton and Vertiv, two of the biggest U.S. electrical equipment providers. 

The last burning question: What happens if AI turns out to be a bubble, and all the new infrastructure was built for nothing? Thompson says that with the largest tech companies spending billions on the sector, AI will likely not lead to another dotcom burst. “We’re asking if Microsoft is too big to fail,” he says. “If for no other reason than just brute force, they’re going to find a way to make this work.” 

This article appears in the August/September 2024 issue of Fortune.