AI’s Power Hunger Is Pulling the Grid Back to Gas

The US is building more fossil fuel infrastructure than it has in ten years, and AI data centers are a big reason. The Verge AI reports on a new study from the Environmental Defense Fund and research firm Atlas Public Policy. It finds that gas-fired plants are quickly taking share of the country’s power development pipeline while renewable growth slows.

The report uses US Energy Information Administration data, and the numbers move fast. Fossil fuel projects made up just 5 percent of planned new capacity in 2024. By the second quarter of 2026, that share hit 30 percent.

⚡ What’s Changing

Electricity demand is climbing faster than it has in more than a decade. Data centers, factories, and electric vehicles are all adding load. The Trump administration’s answer is “drill, baby, drill.” It treats fossil fuels as the preferred power source for energy-hungry data centers and works against wind and solar through stop work orders, blocked permits, and the removal of tax incentives.

As detailed in The Verge AI, the shift already shows up in the pipeline:

  • Planned fossil capacity could reach 68 gigawatts by 2030. That’s a 45 percent jump from construction plans just six months ago.
  • Fossil fuels could make up 70 percent of the development pipeline by the end of the decade if the trend holds.
  • Renewables are being canceled faster. 4.9 percent of renewable projects are getting scrapped, compared with 2.9 percent of fossil projects.
  • Solar and wind still get built, but many of those farms are the last wave of projects that qualified for Biden-era incentives.

“The U.S. power sector is at a new inflection point, with decisions made today shaping the cost, reliability, and emissions profile of the grid for decades to come,” the report says.

💰 The Cost Paradox

What stands out to me is that the market is heading toward the more expensive option. New utility-scale solar and wind cost $68 to $69 per megawatt-hour on average. New combined-cycle gas plants cost around $90. Globally, solar is now the cheapest power source in most of the world.

The momentum before this year was real, too. Wind and solar together generated more electricity than coal for the first time in 2024. Solar made up 81 percent of new capacity added that year. Even so, the US still gets 60 percent of its electricity from fossil fuels.

Gas also brings risks that a cost-per-megawatt figure doesn’t capture. Prices have become more volatile, especially since the US-Iran conflict began. Gas and coal plants also produce air pollution and greenhouse gases, and those turn into health costs. The authors warn that if every gas project in the pipeline gets built, the country could be stuck with more pollution and higher electricity bills for decades.

🤖 Why This Matters for AI

The AI industry is now a political force in energy policy. Hyperscalers want power quickly and reliably, and gas plants can promise steady output in a way that intermittent renewables without storage can’t. Policymakers have taken that demand and used it to justify a broader fossil buildout.

There’s a big caveat, though. The report notes that many of these plans could still fall through, particularly if AI data center demand forecasts turn out to be overblown. If that happens, ratepayers could end up paying for gas plants that weren’t needed and are expected to run for 30 or 40 years.

🔮 The Next 1-3 Years

Here’s what I expect to see:

  1. Electricity costs become an AI story. Bills are already rising as utilities upgrade grids. Communities will blame data centers more and more, and local pushback on new facilities will grow.
  2. Energy strategy sets companies apart. AI firms that lock in long-term clean power deals, on-site storage, or nuclear agreements will have more stable costs than those relying on volatile gas markets.
  3. Demand forecasts face scrutiny. If AI efficiency gains slow the growth in compute demand, expect hard questions about stranded gas assets.

📋 What You Should Do Now

  • If you’re planning compute capacity, model energy prices with a gas volatility scenario included, not just today’s rates.
  • If you’re choosing cloud or colocation providers, ask about their power sourcing and how they plan to handle price swings.
  • If you report on sustainability, expect closer attention to Scope 2 emissions (emissions from the electricity you buy) as grid mixes shift.
  • If you’re an investor, keep an eye on the gap between announced data center demand and what actually gets built.

The grid decisions being made right now will outlast the current AI hype cycle by decades. Whether this gas buildout turns out to be smart planning or an expensive overreaction depends on demand forecasts that nobody has proven yet. The full breakdown of the EDF and Atlas Public Policy findings is available at The Verge AI.

Scroll to Top