Trump Lifts Pollution Limits on Coal Plants to Power AI Data Centers

AI's appetite for power is starting to reshape more than the electricity market. It is now changing environmental policy in the world's largest economy. Donald Trump's administration has announced it will repeal the Mercury and Air Toxics Standards, a set of Obama- and Biden-era rules that regulated emissions of mercury and other toxic pollutants from coal-fired power plants. The decision comes as US electricity demand grows at an unprecedented pace, and the main driver of that growth is the construction of massive data centers for training and running AI models.

Some context helps explain the scale of the problem. MATS was introduced under Obama and tightened under Biden. The rules limit emissions of mercury, lead, arsenic, and other heavy metals that coal plants release into the atmosphere. Coal generation accounts for roughly half of all mercury emissions in the US. Mercury is a potent neurotoxin: even relatively small doses, with chronic exposure, are linked to birth defects, impaired cognitive development in children, and damage to the kidneys and nervous system. Environmental groups estimate the standards saved thousands of lives and prevented tens of thousands of cases of illness.

The Trump administration, however, follows a different logic. The current White House's sweeping deregulation push has a single goal: build out energy capacity as fast as possible. The reason is simple.

Tech giants — Microsoft, Google, Amazon, Meta, and Elon Musk's xAI — have announced data centers with a combined capacity of tens of gigawatts. Each large cluster for training language models uses as much electricity as a small city. Existing US energy infrastructure can't keep up with this demand, and building new capacity, from nuclear plants to solar farms, takes years.

Coal plants, many of which were slated for closure, have become an unexpected lifeline: they already exist, they're already connected to the grid, and they can ramp up output quickly.

This is where economic pragmatism collides with public health. Repealing MATS effectively frees coal plant operators to run at full capacity without expensive filtration and emissions-control systems. For utilities, that means lower costs and the chance to sell more electricity to data centers at attractive prices. For people living near coal plants, it means higher concentrations of toxic substances in their air and water. Consider the Kingston Fossil Plant in Tennessee, a 1.4-gigawatt coal plant on the shore of Watts Bar Lake. Facilities like this will be the first to benefit from deregulation, and also the biggest sources of pollution.

For the AI industry, this creates a serious reputational problem. Tech companies have spent years building a narrative around sustainability, carbon neutrality, and green energy. Google and Microsoft publish environmental footprint reports, and Amazon invests in renewables. But in reality, the rapid growth of AI energy consumption has already increased these companies' carbon footprints, and now it threatens to drive up toxic emissions as well. When a model generating text or images runs on electricity from an unfiltered coal plant, corporate green pledges ring hollow.

There's a geopolitical dimension, too. The US-China race for AI leadership is turning into a race for energy resources. The Trump administration is openly betting on speed: build faster, launch faster, train the next generation of models faster. In this logic, environmental standards are an obstacle, a bureaucratic barrier that slows progress. That mindset isn't new, but the scale of the consequences is unprecedented. Deregulation used to affect individual industries; now it's directly tied to a technology widely described as defining the 21st century.

Critics point to an obvious paradox: AI is pitched as a tool for solving global problems, including environmental ones. Yet its development, in its current form, worsens the very problems it supposedly exists to solve. Mercury in the water and air won't become any less toxic because it was emitted to train yet another language model.

Still, reducing this to "AI versus the environment" would be an oversimplification. The real problem is the lack of a long-term energy strategy that balances technological growth with public health. Nuclear power, geothermal energy, and new energy storage technologies could all supply data centers with clean electricity, but they take time and investment. Repealing MATS is a choice in favor of the quick and dirty fix. And the people who will pay for that choice aren't the tech companies, but the people who live next to coal plants.

*Meta is designated an extremist organization and banned in Russia.