AI’s massive appetite for energy will make or break climate-change goals

“What the climate and energy sector is facing with AI is not unlike what we’re facing across the board with AI. It’s extremely complicated, and it’s just a question of, ‘Will the benefits outweigh the harms?’” Austin Whitman, CEO of nonprofit The Climate Change Project, told Fortune.

The key issue is AI’s massive appetite for energy. AI will command a huge scaleup in power-generation capacity, a transition that’s already begun. Microsoft, the very firm powering the AI revolution with its $10 billion investment in OpenAI, signed a colossal $10.5 billion deal with renewable energy provider Brookfield last week. Other AI leaders are investing in energy startups, pitching everything from thermal batteries to modular nuclear reactors. 

After years of low-to-no growth, power utilities are facing skyrocketing increases in demand from data centers that host and train AI models—industry insiders have floated potential ninefold spikes in some regional markets. Aging energy infrastructure simply isn’t equipped to handle so much change so fast, raising questions as to whether it’s even possible to feed AI’s appetite for energy.

The demand surge means grid operators need solutions, fast—meaning climate goals sometimes have to take a backseat. After Meta announced it was opening a new data center in Kansas, the local power utility announced it would be pushing back the scheduled retirement date of one of its coal plants by five years, to 2028. Regulators have nixed data center operators’ repeated petitions to use diesel generators for backup power instead of green alternatives. 

“[AI innovation] is happening at a pace where even the even when the tech [companies] are telling the utilities what they would like in order to be able to meet [their emissions] requirements, utilities are pushing that aside in favor of what [they think they] can do at the fastest scale,” Sierra Club Senior Strategy Advisor Jeremy Fisher told Fortune. “If you’re not doing a lot of planning ahead of time, the turnkey solution for utilities that have operated in the space for many years are comfortable with is build new gas [plants,] or keep a coal plant around a couple years longer.”

But some experts say that when it comes to the energy grid, AI could actually be the perfect solution to the problems it’s creating. Power utilities focus on minimizing wasted energy and optimizing the grid to make it as efficient as possible—a task for which AI seems perfectly suited. Power companies are racing to implement generative AI’s massive potential to unlock efficiency gains in how power is stored, transmitted, and managed. 

“You’re talking about an industry that does not move very fast. And now, all of a sudden, they aren’t just kind of jumping on a bandwagon—they truly do see the applicability in a number of different dimensions for [AI],” Schneider Electric North America CTO Scott Harden told Fortune.

Making the grid, not breaking it

Optimistic observers frame AI as a tool that will be vital in making the energy grid greener and more efficient. A recent Department of Energy report highlighted AI’s use in forecasting electrical demand and future production from renewable sources such as wind and solar, which can fluctuate depending on the weather. 

It also pointed to potential generative AI applications in applying for federal permits, a painstaking process that’s a key roadblock for grid operators who want to scale up their operations fast: Equipment lead times and regulatory red tape have stopped many new energy projects in their tracks, with wait times to get new capacity onto the grid as long as multiple years. 

AI has underscored the need for reliable power. Because solar and wind generation is dependent on the weather, they generate electricity in bursts completely uncorrelated with when customers actually need it. That’s generated concerns that renewables are inherently less reliable than fossil fuels, making sources such as coal, petroleum or natural gas preferable for mitigating blackouts and outages. 

“The way grid planning decisions happen is, first and foremost, it’s about reliability…The worst case scenario is that power grids start to go down,” Whitman said.

But technological breakthroughs in energy storage mean that these concerns are mostly misplaced: Companies utilizing everything from heat batteries to hydropower dams are making it easier and cheaper to store excess renewable energy and release it when the wind’s not blowing or the sky is cloudy.

Operators hope that with the help of these storage tools, AI will be able to deal with discrepancies between supply and demand, minimizing waste and making the grid more efficient by monitoring it in real time and balancing generation and usage.

At a basic level, AI won’t be doing anything radically different from the processes that grid operators already have in place. Utilities have relied on algorithms and simple forms of AI for years, Harden said, and they already employ human experts to make predictions about future supply and demand. But Harden said AI’s promise lies in its ability to take in much larger data sets and work much faster than existing methods.

“AI is quite agile, and capable of aggregating and analyzing massive amounts of data and responding in real time. The ability to learn and execute far exceeds human capacity,” Schneider Electric Chief Public Policy Officer Jeannie Salo told Fortune. “We’re already seeing the outcomes of AI as it helps stakeholders better forecast electricity needs and flows and their operations.”

At an event hosted during San Francisco’s SF Climate Week last month, a Google grid researcher said that the company is already using AI to run grid simulations.

“We’re wondering if AI can make the grid instead of breaking the grid,” Page Crahan, the Google employee, said.

The need for more power

The potential efficiency gains of an AI-managed grid can’t hide the fact that the energy industry simply needs more capacity. Providers are racing to get new projects online, and experts are split on the question of whether a spree of investment will push climate goals forward or continue to lean on coal and other dirty sources.

As reliance on coal has diminished over the past 20 years, it’s in large part been replaced by natural gas. A recent Goldman Sachs report estimated that natural gas would supply more than half of AI-related energy demand. While natural gas emits fewer greenhouse gasses than coal or petroleum, it’s still a fossil fuel that’s far worse for the environment than fully renewable energy sources—Fisher said that natural gas contributes roughly half as much greenhouse gas emissions as coal.

“This rush to build as quickly as feasible is really swamping all of our other long-term considerations,” Fisher said. “All this infrastructure that’s going in right now, particularly the new natural gas [infrastructure], that’s 30-year infrastructure. So we have to be able to make good decisions right now about what the future of the electricity sector is going to look like.”

But recent declines in the price of wind and solar energy are giving some analysts optimism that energy utilities keeping an eye on the bottom line will naturally turn to green solutions.

“What gets built is typically, what’s cheapest to build,” Whitman said. “Renewables are, in many cases, cheapest to build. So I would expect that we continue to see a lot of really solid clean energy buildout…All signs are pointing in favor of a grid that’s really heavily, actually continuing to move in the direction that it’s been moving for about the last 15 years, which is toward clean energy.”