If you were concerned that the air was getting too clean, the Trump administration has great news for you. On Monday, Environmental Protection Agency head Lee Zeldin announced that the Trump administration had finalized a rule to eliminate a Biden-era policy requiring fossil fuel–fired power plants to reduce their planet-heating carbon dioxide emissions starting in the 2030s. It has also proposed a supplemental rule that—if it is finalized and survives the inevitable legal challenges—would rescind “every remaining greenhouse gas standard for the power sector” and severely limit future administrations’ attempt to revive such rules. Using one of his favorite lines, Zeldin bragged that these changes amount to the “largest deregulatory announcement in U.S. history.” It was easy to miss this news amid all the talk about whether artificial intelligence will imminently “kill all humans,” a conversation sparked by researcher Jacob Coxon, who resigned from Anthropic over fears that it and OpenAI “are racing straight to self-improving superintelligence and gambling with our lives.” The EPA’s announcement is nonetheless a clear win for Republicans and for the polluters who have been attacking rules to limit power plants’ greenhouse gas emissions since at least 2014, when the Obama administration first proposed its plan to cut power plant emissions. (The goal it set was to cut emissions by 32 percent below 2005 levels by 2030.) But Zeldin’s announcement is also a boon for the companies building AI data centers that are being predominantly powered by fossil fuels, including coal from old and revived coal plants. BloombergNEF has found that U.S. data centers are on track to consume more gas than most countries within a decade; 99 proposed gas-powered data center projects tracked by BloombergNEF could raise U.S. power-sector emissions by 20 percent.
It would be a mistake to let the AI hyperscalers’ warnings about “apocalyptic” risk overshadow the ongoing dangers posed by their business models. As Sam Altman, Dario Amodei, and Elon Musk all rally behind proposals to “pace the frontier”—their peculiar way of describing the need to slow down AI development—pundits have marveled over the CEOs’ seemingly novel request for the federal government to regulate them. None of this is entirely new for top AI companies, however, whose employees and top executives have for years argued that the products they’re building could wipe out human civilization. As Amodei wrote in a long blog post over the weekend, Anthropic has “long supported sensible and targeted AI regulation, specifically bills that focus on transparency and on third-party auditing.” Trump dismissed Amodei’s and his colleagues’ warnings, arguing that the only “control” needed to deal with artificial intelligence is a “STRONG AND SMART (High IQ!) PRESIDENT!”
In pouring cold water on the idea of regulating the AI industry, the Trump White House was pulling from a familiar playbook. The administration seems to have a one-size-fits-all approach to supporting its favorite industries, modeled on its generous handouts to the fossil fuel industry. That strategy isn’t especially complicated: Tear down regulatory barriers to growth, shower companies with tax breaks, and justify all of it in terms of national security. The overriding message here is to keep going as fast as possible for as long as possible—drill, baby, drill.
At the EPA especially, fossil fuel and AI boosterism are one and the same. As Hannah Story Brown reported in The American Prospect, turning the United States into the “AI capital of the world” is now a top priority for Zeldin’s EPA. Data center developers have already been able to take advantage of virtually on-demand “presidential exemptions” from parts of the Clean Air Act—in this case, limits on toxins like mercury, arsenic, and benzene—on the basis of “national security interests.” Zeldin has refused to establish nationwide requirements or recommendations for data centers’ air pollution and water usage; has proposed limiting public input on data center approvals; and has permitted data centers to start construction before attaining air pollution permits. A new report from the Environmental Protection Network—representing more than 800 former EPA employees—details 30 such actions taken by the administration. Even without accounting for recent rollbacks, the report projects that data center–driven air pollution could contribute to some 1,300 premature deaths and more than $20 billion in annual public health costs by 2028.
There are myriad reasons for the general public to be concerned about a gargantuan building binge getting carte blanche from the federal government to pollute. Thanks partly to new electricity demand from data centers, emissions from the power sector—the country’s second-largest source of greenhouse gas emissions—rose by 4 percent last year, outpacing those from the rest of the economy. The engines and turbines that power them can emit nitrogen oxides, volatile organic compounds, and particulate matter linked to worsening asthma, strokes, and premature death, and which can harm children’s developing brains and nervous systems. But hyperscalers themselves—who haven’t seemed terribly concerned about all that—could find that a “drill, baby, drill” approach may not be in their best interest, either.
You don’t need to be completely cynical about executives’ mass-extinction warnings to understand the more quotidian benefits they might see in slowing down their debt-fueled arms race—and why Trump’s refusal to let it happen could be a problem. Bloomberg columnist Matt Levine made the case this week for why a collective slowdown in frontier model development could actually be in the industry’s financial interest. The jockeying among AI “labs” to keep building and releasing ever more advanced models is extraordinarily expensive for all of them, requiring “ever-increasing billions of dollars of computing power,” Levine writes. If they all slowed down at the same time, they could spend less on computing power and keep charging “frontier-model prices for a longer time.” Coordinating a slowdown would seem to constitute a fairly straightforward violation of antitrust law, which forbids competitors’ collaborating to limit output. “But if they publish papers about how important it is to slow down, that might have a similar coordinating function,” Levine writes. “And if the government believes those papers, it might help them coordinate. Maybe the government will impose pacing by regulation that the labs could not impose by agreement. Or at least the government will let them get together and agree to slow down.”
Many things can be true at the same time: Amodei, Altman, and Musk really do seem to believe in the mounting existential risks of “superintelligence.” They could also have a financial interest in “pacing the frontier” and slowing down rapid-fire model development in ways that might protect the public too. There are any number of reasons to be skeptical of these executives’ intentions. OpenAI has lobbied extensively against state-level regulations on data centers and AI alike; OpenAI president Greg Brockman and his wife were MAGA Inc.’s largest donors in 2025. As much as Amodei has tried to position Anthropic as the “good guys,” his company is building a Minority Report–style “pre-crime” system that can alert authorities to crimes before they happen. So yes, it’s possible this is all just naked opportunism: Anthropic and OpenAI might just want to get relatively weak rules on the books while Republicans still control Congress and the White House. Maybe they’re banking on nothing passing at all, or just want to save face with a skeptical public.
But you don’t need to trust these guys to appreciate the prospective dilemma they face: Trump does not want them to slow down for any reason. If they really do want to “pace the frontier,” including for purely financial reasons, he could punish them. That would be bad news both for companies who might want the government to stop them all from spending so much money—and for anyone concerned that their products really are dangerous.
There’s precedent for this kind of thing in the oil industry, of all places. As Oklahoma prospectors drilled furiously in the early twentieth century, they begged regulators to stop them from driving down prices to unsustainable levels and cratering the market. In 1931, Texas Governor Ross Stirling declared martial law and sent 800 Texas National Guard troops to pry workers off their pump-jacks in the name of keeping prices high.
Whatever nonsense Trump and Zeldin might spew about “free markets,” industries depend on states to create rules that keep individual firms’ self-interest from undermining the industry as a whole.
Governments also don’t tend to like it when companies try to do that themselves. After years of oil and gas companies bingeing on debt to fund rapid-fire shale drilling, previously patient and generous Wall Street investors started to get nervous about that sector’s profitability. The industry began to practice “capital discipline,” prioritizing efficiency and returns over output. Among the loudest proponents of this approach was Pioneer CEO Scott Sheffield. “Everybody’s going to be disciplined, regardless of whether it’s $75 Brent, $80 Brent, or $100 Brent,” Sheffield said publicly in 2021. “All the shareholders that I’ve talked to said that if anybody goes back to growth, they will punish those companies.” The Federal Trade Commission subsequently accused Sheffield of engaging in “a series of efforts to coordinate output levels to keep production artificially low.” As I wrote at the time, the politics of all this were disorienting. While Republicans accused the Biden administration of waging war on fossil fuels, its FTC singled out Sheffield for limiting production while approving his company’s merger with ExxonMobil; Trump’s FTC eventually dropped the issue last summer.
Given its history and preference for avoiding volatility, Trump 2.0’s “drill, baby, drill” approach hasn’t been wholly popular with the traditionally GOP-aligned fossil fuel industry. Since he returned to office, executives have routinely complained in the Dallas Fed’s quarterly energy survey about the administration’s tariffs driving up their costs, as well as about chaos fueled by the Iran war and other half-baked policy drives. “Golly,” one exploration and production company executive wrote in the latest edition of the survey, released in June. “What could possibly be affecting our business other than a COVID-sized supply gap driven by a war being commandeered by an administration that just cannot tell the truth? They jawbone the price down basically every Sunday evening. If they know Hormuz reopening isn’t likely, it’ll make the medium-term supply issue ten times worse.” Another wrote that the “whiplash from diplomacy-by-social-media has become the single most unpredictable input in our planning. We don’t need certainty about the future, just certainty that policy won’t change between the morning and the afternoon.”
Silicon Valley might learn the hard way that being Trump’s favorite isn’t always a good thing. Hyperscalers have enjoyed the benefits of the EPA’s regulatory carve-outs for data centers and tried to preempt state regulations. They’ve spent years trying to get policymakers to tear down barriers to rapid-fire frontier model development, insisting that they’re the most responsible actors to be building mass-extinction machines. Amodei and Altman are probably not building a Machine God. They might reasonably want to stop spending hundreds of billions of dollars a year making machines they don’t fully understand, which they also believe could kill everyone. If these executives really do want frontier model development to be regulated, they should have to follow all the other lifesaving rules they’ve been skirting too.