Natural gas for AI is emerging as the most practical near-term solution to meet soaring energy demands from data centers.
Artificial intelligence is at the forefront of the United States’ new tech revolution—and the focus of a battle against China for dominance in the sector. But despite flashy bets by computing giants like Microsoft, Amazon, Google, Apple, and OpenAI—and the Trump administration all in on unleashing energy to fuel the revolution—the blunt reality is that it will take a lot of new electricity projects to be built to fuel the up-and-coming technology: More than 1,000 terawatt hours by 2035, according to a new report by the International Energy Agency.
After many years of relatively flat demand, the U.S. power system is witnessing significant growth. There is no doubt that the AI revolution is a big part of that paradigm change and excitement, but not all energy sources are equally suited to the task and timelines.
At least in the United States, natural gas for AI applications is likely the short-term winner. Turbines burning highly monitored and low-emissions domestic natural gas can provide stable, dispatchable power, making them the optimal choice for data center operators either as grid backup or in a “behind the meter” play.
In fact, the energy consultant group Enverus predicts up to eighty new gas-powered plants, adding some 46 gigawatts of capacity, will be built in the United States by 2030—more than 20 percent of what was added in the last five years and equivalent to the size of Norway’s entire electrical system. AI center developer Crusoe has already secured some 4.5 gigawatts of natural gas turbines to power the centers.
Other Options Still Face Headwinds
Building small modular reactors (SMRs), meanwhile, is still handicapped by regulatory approval processes, and thus, commercial deployment may not align with data center timelines—given the current race to deploy. In the medium term, however, that calculus will hopefully change dramatically.
Transmission infrastructure required to connect large projects of all kinds often faces lengthy permitting delays. But in several regional systems, there is “room” in the grid in certain locations or with improved system operations. As an example, a new study from Duke University notes, “…the U.S. power system has extensive untapped potential to more quickly add large loads while mitigating the need for costly system upgrades—as long as those loads can occasionally cut their power use when the grid is most stressed.”
One renewable technology also looks likely to prevail—geothermal energy. Due to recent technological advances—often taken from oil and gas directional drilling and fracking—this longstanding technology is seeing a renaissance. As an example, Fervo Energy has already partnered with Google and other tech giants. Their technology is allowing for geothermal to be much more geographically dispersed—a radical departure from its historical applications.
Making the Gas Bet Work
Stakeholders are changing the market as well. Big energy companies like Chevron and Exxon have announced their intentions to supply data center operators with power generation mainly from natural gas and are also mulling the inclusion of carbon capture and storage technology at certain sites.
The biggest challenge for natural gas turbines is timing. Suppliers like GE Vernova, Siemens, and Mitsubishi Heavy Industries are all still facing post-pandemic supply chain crunches that will constrain new deployments. If data center operators and energy suppliers want to rely on natural gas for AI development, they need to act quickly and secure turbine orders now. Delaying decisions risks a scenario where power shortages force suboptimal solutions, such as increased reliance on diesel backup generators—an outcome at odds with sustainability goals.
On the environmental side, natural gas’s main issue is controlling methane emissions, which can significantly erode climate benefits from unchecked leaks during production and transport. But recent advances in monitoring and mitigation have transformed the industry, including satellite tracking, continuous monitoring, and tighter wellhead standards that have been shown to lower methane leakage by up to 80 percent. Colorado’s decade-long regulatory leadership on the topic has proven valuable throughout the industry.
The benefits and access to U.S. gas exports might also impact investment decisions for data centers in partner countries. The United States was quickly able to supply Europe with cheaper and cleaner LNG that is transparently monitored after the bloc largely cut itself off from high-emission Russian pipeline gas. By adopting the same rigorous tracking and certification practices used in LNG exports, U.S. data centers can power their operations with confidence, knowing they are using the cleanest available gas.
Data centers are the backbone of the digital economy, and their power needs cannot be met through wishful thinking. While the long-term decarbonization of the grid is essential, the immediate reality demands pragmatic solutions. In the short term, in the United States, that choice is likely to be natural gas turbines as the way to power the next generation of compute infrastructure.
Morgan D. Bazilian is the director of the Payne Institute and a professor of public policy at the Colorado School of Mines.
Image: Panuwat Phimpha / Shutterstock
