When Stargate’s first AI-focused data center goes live next year, the 900-acre complex will require enough electricity to continuously power 300,000 homes. But getting that much energy isn’t as straightforward as simply tapping into the local utility.

Unlike its competitor Meta Platforms Inc.—which just revealed a 20-year agreement to purchase electricity from a nuclear facility in Illinois—Stargate is turning to a power source that had largely fallen out of favor: small, single-cycle natural gas turbines.

That’s because the explosion in AI-related computing has sharply increased electricity demand, outpacing the construction of new power plants and the timelines required for grid connections.

To meet the urgent power needs of its Texas-based supercomputer hub, a joint venture involving OpenAI, Oracle, and SoftBank is assembling a cluster of smaller generators. Other major AI infrastructure efforts, such as Elon Musk’s xAI and the Carlyle–Schneider Electric joint venture Alphastruxure, are employing similar approaches.

This shift has unexpectedly brought single-cycle gas turbines—once sidelined for being inefficient and environmentally harmful—back into play. It also highlights a growing disconnect between the rapid rise of AI and the capacity of the U.S. energy system. Over the next few years, AI infrastructure is expected to scale massively, and companies that once committed to green energy are now racing to construct data centers the size of small cities. At the same time, the outdated U.S. power grid and slow nuclear buildouts mean that renewable and nuclear alternatives are often unavailable or insufficient.

“There’s a real urgency across the industry to find power quickly,” said Cully Cavness, COO and co-founder of Crusoe, which is developing the first Stargate facility. “We’ve had to get innovative with our energy strategies for data centers.”

Stargate and its peers say they’re also evaluating carbon-free sources like solar power, batteries, and nuclear energy to meet the enormous, continuous electricity requirements of AI operations. This week, Constellation Energy Corp., the nation’s largest nuclear operator, signed a deal to provide Meta with nuclear energy, and last year announced plans to restart the Three Mile Island plant to supply electricity to Microsoft.

However, many data centers—Stargate included—have limited access to nuclear energy, and concerns about the reliability of renewable sources have driven many developers to opt for gas-fired solutions that can run 24/7.

As a result, demand for more advanced, combined-cycle gas plants—which are larger and more efficient—has surged, creating production bottlenecks that can delay delivery by three to five years, with an extra year for installation and connection. To avoid those delays, developers are turning to smaller, faster-to-deploy single-cycle turbines, typically generating 100 to 200 megawatts.

Turbine manufacturers like Mitsubishi Power Americas and Siemens Energy AG report a spike in orders for compact units, often under 70 megawatts, with expected delivery dates in 2025 or 2026. GE Vernova is also experiencing heightened interest in simple-cycle turbines, which can later be upgraded to more efficient systems.

“It’s the biggest rush for gas turbines in the U.S. since the Enron era,” said Rich Voorberg, President of Siemens Energy North America. Even demand for the smallest turbines is picking up, with developers purchasing off-the-shelf jet engine turbines that generate as little as 5 megawatts.

“Many developers prefer to build incrementally—it gives them more flexibility,” said Shannon Miller, CEO of Mainspring, which makes micro-generators so small that it takes 100 of them to produce just 25 megawatts.

Once Stargate and similar projects are eventually connected to the regional power grid, the smaller turbines will likely be repurposed for backup use and grid-stabilization services, Cavness noted.

However, the growing dependence on gas-fired generation raises significant environmental concerns. Big tech companies made bold climate commitments years ago under pressure from staff and the public, but now Amazon, Google, and Microsoft admit that the rush to develop AI and new data centers could complicate their sustainability goals.

Single-cycle turbines release an average of 1,389 pounds of carbon per megawatt-hour, compared to 839 pounds for more efficient combined-cycle units, according to BloombergNEF. Furthermore, many tech companies try to offset emissions by purchasing clean energy in some areas, while situating high-power-consuming data centers in regions dominated by fossil fuel-based electricity.

This imbalance means that the growing demand for AI infrastructure is still heavily reliant on natural gas and coal, potentially slowing the overall decline in U.S. carbon emissions, BloombergNEF researchers warn.

For the moment, the priority remains simply obtaining enough power—megawatts now, gigawatts later.

“The world’s not going to wait,” said Sebastian Bonneau, a partner at McDermott Will & Emery who leads the firm’s data center practice. “You’ve got to lock in your energy supply.”