The Dirty Bargain: How Trump’s Coal Plant Deregulation Is Fueling AI’s Insatiable Energy Appetite

On February 20, 2026, the Trump administration quietly dismantled one of the most consequential public health protections of the modern era—the Mercury and Air Toxics Standards (MATS). The move, which effectively greenlights coal-fired power plants to pump dramatically higher levels of mercury, arsenic, and acid gases into the atmosphere, wasn’t just another regulatory rollback. It was a signal flare, illuminating a deeply uncomfortable truth about the future of technology: the machines we’re building to think for us are making the planet dirtier.

This isn’t a story about coal nostalgia. It’s about the brutal arithmetic of energy demand. As artificial intelligence systems scale from experimental curiosities into the backbone of global infrastructure, the electricity required to train and run them is exploding. And in the United States, that demand is being met by the oldest, filthiest workhorses on the grid. The repeal of MATS is the Trump administration’s answer to a question the tech industry has been reluctant to ask: What are we willing to burn to keep the data centers humming?

The Regulatory Guillotine: What the MATS Repeal Actually Unleashes

To understand the magnitude of this decision, you have to understand what MATS was designed to stop. Enacted during the Obama-Biden era, the Mercury and Air Toxics Standards were a landmark set of rules requiring coal-fired power plants to install pollution control technologies—scrubbers, activated carbon injection systems, fabric filters—to capture heavy metals and toxic gases before they escaped into the air. These weren’t aspirational targets; they were enforceable limits that forced the industry to clean up or shut down.

The Trump administration’s repeal doesn’t just loosen the screws; it removes the entire mechanism. Coal plants can now operate without those expensive scrubbers, slashing operational costs while dramatically increasing emissions of mercury—a potent neurotoxin linked to developmental delays in children and cardiovascular damage in adults—along with arsenic, hydrogen chloride, and other hazardous air pollutants. For communities living downwind of these facilities, particularly in the Ohio River Valley and Appalachia, this is a public health catastrophe in slow motion.

But the timing is what makes this decision so perverse. The repeal comes at a moment when the U.S. power grid is under unprecedented strain. Data centers consumed roughly 4% of all U.S. electricity in 2024, and that figure is projected to double by 2030. The training of large language models alone can consume megawatt-hours in a single session—enough to power hundreds of homes for a day. As companies race to deploy open-source LLMs and proprietary models alike, the grid is being asked to deliver more baseload power than it has in decades.

Coal, for all its environmental liabilities, offers something that renewables currently struggle to provide at scale: reliable, dispatchable power around the clock. The MATS repeal effectively lowers the cost of that reliability by externalizing the health costs onto the public. It’s a subsidy for dirty energy, paid for in asthma attacks and neurological damage.

When Silicon Valley Meets the Smokestack: The Unholy Alliance of AI and Coal

The irony here is thick enough to cut with a GPU. The same industry that prides itself on disruption, efficiency, and futuristic vision is now functionally dependent on the most retrograde energy source available. Every time you query a chatbot, generate an image, or run a code completion, you’re drawing power from a grid that, in many regions, still relies on coal to meet its base load. The MATS repeal ensures that those electrons will be dirtier than they were a month ago.

This isn’t a hypothetical. Major tech companies have been scrambling to secure power purchase agreements, and in many cases, they’re finding that the only available baseload capacity comes from coal plants that were scheduled for retirement. The Trump administration’s deregulatory push effectively extends the economic life of these plants, making them more attractive to data center operators who need guaranteed uptime.

Consider the geography: the largest concentration of data centers in the U.S. is in Northern Virginia, a region that draws significant power from coal-fired generation in the Midwest and Appalachia. The repeal of MATS means that the pollution from those plants will increase, directly impacting air quality in communities that have no say in the matter. The tech industry’s carbon neutrality pledges, meanwhile, rely heavily on carbon offsets and renewable energy certificates—accounting tricks that don’t change the physical reality of what’s being emitted from the smokestack.

This dynamic creates a feedback loop that environmental economists have long warned about. Deregulation lowers the cost of coal power, which makes it more competitive relative to natural gas and renewables. That, in turn, discourages investment in cleaner alternatives, locking in a high-carbon infrastructure for years to come. And because AI’s energy demands are growing faster than the grid can decarbonize, the net effect is a net increase in emissions, even as individual companies tout their green credentials.

The Public Health Calculus: Mercury Doesn’t Care About Your Data Center

It’s easy to get lost in the abstract numbers—megawatts, gigawatt-hours, teraflops—but the human cost of the MATS repeal is brutally concrete. Mercury is a persistent bioaccumulative toxin. Once released into the environment, it settles into waterways, enters the food chain, and concentrates in fish. For pregnant women and children, even low-level exposure can cause irreversible neurological damage. The EPA’s own analysis, conducted during the Biden administration, estimated that MATS prevented up to 11,000 premature deaths, 130,000 asthma attacks, and 5,700 hospital visits annually.

Those numbers are now in jeopardy. Without mandatory pollution controls, coal plants will emit more mercury, more arsenic, and more acid gases. The communities most affected are disproportionately low-income and non-white, a pattern that has held true for decades of environmental injustice. The Trump administration’s decision effectively tells those communities that their health is less important than keeping the lights on for AI servers.

And let’s be clear: this isn’t about energy security in any meaningful sense. The U.S. has ample natural gas reserves, a rapidly expanding renewable energy sector, and a growing fleet of grid-scale batteries. The choice to double down on coal is ideological, not practical. It’s a signal to the fossil fuel industry that the regulatory pendulum has swung back in their favor, and that the costs of pollution will continue to be externalized onto the public.

For the tech industry, this creates a moral hazard. By continuing to build energy-intensive AI infrastructure without demanding cleaner power sources, companies are implicitly endorsing a system that prioritizes short-term profits over long-term health. The AI tutorials and model documentation that celebrate efficiency gains often gloss over the energy cost of training and inference. The reality is that every optimization in model architecture is dwarfed by the scale of deployment.

The Global Precedent: What America’s Coal Pivot Means for the World

The United States has long been a trendsetter in environmental regulation, for better or worse. The MATS repeal sends a clear signal to other nations that the world’s largest economy is willing to sacrifice air quality for energy availability. This is particularly consequential for developing countries that are already struggling to balance industrialization with environmental protection.

China, for instance, has made significant strides in reducing its coal dependence, but it still operates the world’s largest fleet of coal-fired power plants. If the U.S. signals that coal is acceptable—even preferable—for meeting the energy demands of cutting-edge technology, it undermines the moral authority of those pushing for cleaner alternatives. The same logic applies to India, Southeast Asia, and parts of Eastern Europe, where coal remains a politically sensitive but economically entrenched energy source.

The Trump administration’s broader deregulatory agenda, which includes rolling back vehicle emissions standards and weakening the Clean Water Act, creates a permissive environment for high-pollution industries. The MATS repeal is just one piece of a larger puzzle, but it’s a particularly revealing one because it directly ties environmental policy to the future of technology. If the U.S. can’t figure out how to power its AI revolution without poisoning its citizens, what hope is there for countries with fewer resources and weaker institutions?

This isn’t just an environmental story; it’s a geopolitical one. The race to dominate AI is often framed in terms of compute, data, and talent. But the underlying energy infrastructure is the invisible enabler. Nations that can generate cheap, reliable, and clean power will have a structural advantage. Nations that cling to coal will find themselves locked into a high-cost, high-pollution trajectory that becomes harder to escape with each passing year.

The Uncomfortable Question: Can We Afford the AI We’re Building?

The MATS repeal forces a reckoning that the tech industry has been avoiding. For years, the narrative has been that AI will solve our biggest problems—climate change, disease, resource scarcity. But the tools themselves are resource-intensive in ways that are only now becoming apparent. Training a single large model can emit as much carbon as five cars over their lifetimes. Inference—the process of running a model to generate responses—is even more energy-intensive at scale.

The vector databases and retrieval-augmented generation systems that power modern AI applications require constant computation. Every search, every recommendation, every automated decision draws on a vast infrastructure of servers, networking equipment, and cooling systems. The energy demand is not a bug; it’s a feature of the architecture. And as AI becomes more embedded in everyday life, that demand will only grow.

The Trump administration’s answer is to make coal cheaper and dirtier. But that’s not a solution; it’s a deferral. The health costs will be borne by communities for decades. The climate costs will be borne by the entire planet. And the tech industry will continue to operate as if these externalities don’t exist, because the regulatory framework that would force them to account for them has been systematically dismantled.

The question that remains—and it’s one that no press release or earnings call has adequately addressed—is whether the benefits of AI are worth the cost. Not the financial cost, but the human and environmental cost. If the price of progress is a generation of children with elevated mercury levels and a planet that’s harder to decarbonize, then the bargain looks less like innovation and more like exploitation.

The Path Forward: Regulation, Innovation, and Accountability

None of this is inevitable. The MATS repeal can be challenged in court, and environmental groups are already preparing lawsuits. But legal challenges take time, and the damage from increased emissions is cumulative. The real solution lies in decoupling AI’s growth from fossil fuel dependence, and that requires both policy and technology.

On the policy side, the next administration—or a future Congress—could reinstate MATS and strengthen it. But more importantly, the U.S. needs a national energy strategy that accounts for the coming surge in data center demand. That means streamlining permitting for renewable energy projects, investing in grid modernization, and creating incentives for tech companies to co-locate with clean power sources.

On the technology side, the AI industry has a role to play. Model efficiency is improving, but not fast enough to offset the growth in deployment. Hardware innovations, from specialized chips to liquid cooling, can reduce energy intensity. And better vector databases and retrieval systems can reduce the computational overhead of inference. But these are incremental gains, not silver bullets.

The deeper challenge is cultural. The tech industry has to stop treating energy as an externality. Every model release should include a transparency report on energy consumption and emissions. Every data center contract should prioritize clean power. And every executive should be asking not just “Can we build this?” but “Should we build this, given what it costs?”

The MATS repeal is a wake-up call. It reveals the ugly truth that the AI revolution is being powered by the same dirty energy that has been poisoning communities for generations. The question now is whether we have the will to change course—or whether we’re willing to let the machines burn whatever they need to keep running.

---

References

[1] Rss — Original article — https://www.theverge.com/science/882288/trump-ai-data-center-power-plant-pollution-mercury-mats

[2] Wired — They Bet Against Trump’s Tariffs. Now They Stand to Make Millions — https://www.wired.com/story/they-bet-against-trumps-tariffs-now-they-stand-to-make-millions/

[3] TechCrunch — The creator economy’s ad revenue problem and India’s AI ambitions — https://techcrunch.com/video/the-creator-economys-ad-revenue-problem-and-indias-ai-ambitions/

[4] The Verge — SCOTUS rules Trump’s tariffs are illegal — but the fight is far from over — https://www.theverge.com/policy/882227/scotus-trump-tariffs-ruling-imports-small-businesses-refunds