AI’s Energy Hunger: Driving a Coal Comeback?
The rise of artificial intelligence is reshaping the digital world—and now, the energy system that powers it. As demand for AI accelerates, so does the need for electricity to run the vast data centers behind it. While tech companies are investing in renewables at record levels, the sheer scale and immediacy of AI's energy appetite are forcing a rethinking of what powers the grid.
In some regions, that rethink includes a return to coal. From China's coal expansion to delays in U.S. coal plant retirements, fossil fuels are being called back into service to guarantee reliable electricity for AI's growing workload. This shift is stirring debate about whether coal is a necessary stopgap—or a dangerous step backward at a time of climate urgency.
AI’s Electricity Demand Is Surging
Artificial intelligence doesn't just run in the cloud—it runs on power. Training large models and serving billions of user queries daily requires vast amounts of electricity. The International Energy Agency (IEA) estimates that global data center electricity consumption could more than double by 2030, reaching 945 terawatt-hours (TWh) in its base case. That's roughly the current electricity consumption of Japan. In a higher-growth "Lift-Off" scenario, where AI expands rapidly with few constraints, consumption could reach 1,260 TWh by 2030 and 1,700 TWh by 2035.
Even with advances in energy efficiency, electricity use from data centers is still projected to grow significantly, with a lower-bound estimate of 800 TWh by 2030 in the IEA's most optimistic scenario. AI, quite simply, is becoming one of the fastest-growing demands on global power grids.
The Reliability Gap: Why Coal Is Back in the Mix
While renewable energy is expanding rapidly, it hasn't yet solved one key challenge: reliability. AI workloads demand stable, always-on electricity. Solar and wind, by nature, are intermittent and often constrained by storage, transmission bottlenecks, and slow permitting processes.
This has led the world's two most powerful AI competitors—China and the United States—to lean, once again, on fossil fuels to fill the gap. China, the world's largest energy consumer, began building 94.5 gigawatts (GW) of new coal-fired power capacity in 2024—the highest annual total in over a decade. Another 3.3 GW of previously suspended projects also resumed.
The United States is following a similar path. A July 2025 White House executive order has accelerated permits for data center energy infrastructure, explicitly prioritizing "dispatchable baseload" sources like coal, natural gas and nuclear power. The order streamlines environmental reviews for projects exceeding 100 megawatts of demand—roughly equivalent to powering 80,000 homes—while repurposing federal lands for energy infrastructure.
The Limits of Renewables—For Now
Despite these setbacks, AI is also helping to accelerate clean energy deployment. Major tech firms are among the biggest buyers of solar and wind power, signing long-term contracts that support new project development. Many are also investing in grid innovations and energy storage.
Still, the reality is that large-scale batteries, long-duration storage, and upgraded transmission lines aren't yet in place everywhere. Permitting and interconnection delays can stretch years, and clean baseload alternatives like geothermal or hydropower are limited by geography.
Waiting on the Next Wave of Solutions
Looking ahead, the executive order’s definition of "dispatchable baseload" notably includes next-generation nuclear (Sec. 2b-ii)—a recognition that small modular reactors (SMRs) could offer carbon-free reliability. This aligns with tech giants like Microsoft, which recently signed a deal to power an AI data center with Natrium SMRs by 2028.
Regional disparities highlight alternative paths. While Virginia delays coal plant retirements to support Data Center Alley, Iowa’s wind-rich grid already powers server farms with 90% renewable energy—proof that geography shapes solutions. Even in fossil-dependent regions, efficiency gains soften the blow: Nvidia’s new Blackwell chips perform AI tasks using 25x less energy than 2018 models.
Yet these innovations can’t yet offset AI’s exponential demand. The executive order’s focus on immediate capacity—coal, gas, and conventional nuclear—reflects a bet that bridging technologies will arrive too slowly to meet the 100+ MW projects now being fast-tracked.
The debate over AI’s soaring power needs has divided stakeholders—from utility CEOs and Silicon Valley architects to policymakers and prominent online voices—into rival camps, each marshaling evidence to defend their vision for tomorrow’s grid.