On September 25, Atlas Energy Solutions, a Texas company best known for the sand pumped into oil and gas wells during fracking, told investors about a new customer. A "leading frontier AI lab" was backing Atlas's purchase of 283 megawatts of Caterpillar natural-gas generators for a data centre project, and planned to buy the electricity they produce.
One of the people who oversee Atlas is Mark P. Mills, a director since the company listed in March 2023. He is also a leading voice on how much electricity AI will need. His latest forecast, written for an energy institute at Oklahoma State University, argues that AI will require a 10% to 20% rise in American natural gas production. The institute treats it as the baseline for its AI energy work.
Nothing we found shows that Mills bends his forecasts to suit Atlas, and his enthusiasm for fossil fuels is decades older than his board seat. But forecasts like his help decide which power plants get built, and households pay for them for decades. Readers deserve to see the seat beside the numbers, especially since Mills has made this argument before.
When the internet needed coal
In May 1999, Mills and Peter Huber published a Forbes article titled "Dig more coal, the PCs are coming." The research behind it, The Internet Begins with Coal, was published by the Greening Earth Society, a public relations group created by the Western Fuels Association, a consortium of coal-based utilities. Archived copies of the society's website list Mills among its "scientific advisers" from 1998 to 2000.
The report claimed that the internet already used 8% of American electricity and would use 50% by 2008. Within months, scientists at Lawrence Berkeley National Laboratory, a federal research lab, found that some of his estimates were more than ten times too high. Correcting them cut his total about eightfold. A decade later, data centres used a couple of percent of US electricity.
Fourteen years later came a sequel, The Cloud Begins With Coal, sponsored by the National Mining Association and the American Coalition for Clean Coal Electricity. Its best-known claim, which researchers said rested on outdated and overstated figures, was that an iPhone, counting its share of networks and data centres, used more electricity in a year than a fridge.
From fracking sand to data centres
Atlas's main business is proppant, the fracking sand that holds open cracks in rock, mostly in the Permian Basin of Texas and New Mexico. In January 2025 it agreed to buy Moser Energy Systems, which operates about 212 megawatts of gas-fired generators, for $180 million plus shares. The deal, approved by Atlas's board, took the company into "distributed power", generators that sit beside a customer instead of feeding the public grid.
Early this year Atlas signed a framework deal with Caterpillar for 1.4 gigawatts of gas generators through 2030 and told investors this put it "squarely in the crosshairs for data center deployments." Then came the AI lab.
What he told investors
In December 2024, weeks before Atlas announced the Moser deal, Mills appeared on Financial Sense, an investment podcast. He said most of the new electricity America needs would come from burning natural gas, and repeated his phone-versus-fridge comparison. Before discussing Chris Wright, then Trump's pick for Energy Secretary, he disclosed that he knows and likes him. Asked about investments, he said he was personally drawn to the infrastructure that powers AI, and that the undervalued "low-hanging fruit" was in oil and gas. The published transcript contains no mention of Atlas.
Four months later, Mills testified at a House Oversight hearing on AI and electricity. To his credit, the biography he gave the committee lists his Atlas directorship. His profile page at the National Center for Energy Analytics, the think tank he runs, does not.
What the forecast asks for
Mills' report, The Rise of AI, was published in November 2025 by the Hamm Institute for American Energy, named for shale oil billionaire Harold Hamm. It forecasts that digital uses could need up to 1,000 terawatt-hours of extra electricity a year by the early 2030s, roughly 15% to 20% of all US demand. The report notes that the International Energy Agency's outlook implies about a quarter of that growth for American data centres, though Mills also counts networks and chip factories.
Most of that new power, Mills writes, "will necessarily come from burning natural gas." One immediate step, he argues, is to postpone or cancel plans to retire about 100 gigawatts of conventional power plants.
That step is already being taken. Wright, who has cited the AI race as a reason to keep coal running, used emergency powers in May 2025 to stop Michigan's J.H. Campbell coal plant from closing, days before its retirement. Michigan's regulators and the regional grid operator had both approved the closure. In September a unanimous federal appeals court threw out the order, calling the emergency power a narrow, last-resort backstop. The department says its orders prevented blackouts.
Later orders, which the court has yet to rule on, keep the plant running until November 14. Consumers Energy has reported $295 million in costs from May 2025 to June 2026, which it is seeking to recover from electricity customers across the Midwest.
Wright took part in the summit where Mills' report was launched. The first Campbell order came six months before that report, and nothing we found shows the report shaped later ones. Mills cites the Energy Department's own grid study, so influence may run the other way. But the argument and its author keep turning up where these decisions are made.
A record that points one way
Mills' defenders have a fair case. He was right that digital technology would become a large and growing consumer of electricity, and data centres used 4.4% of US electricity in 2023. His AI forecast starts from projects already under construction or permitted. A board seat does not make an argument wrong.
But direction was never the hard part. A forecaster's job is magnitude, because magnitude is what gets built and billed, and on magnitude Mills' record points one way: large numbers, published for coal and oil-and-gas institutions, recommending more fossil fuel. His think tank was launched by the Texas Public Policy Foundation, and we could not find a public list of the "generous donors" Mills says helped start it. His biography says he has briefed state utility regulators, the officials who decide which plants customers pay for.
Mills is entitled to his views. His audiences are entitled to see his interests. Anyone presenting electricity demand forecasts to Congress or a utility regulator should state who funds their work and which energy companies pay them, and their past forecasts should be checked against what happened.
The Late City emailed Mills, through his think tank's press office, and the Hamm Institute in early October with detailed questions, including about investments that would benefit from higher gas demand. Neither replied by our October 7 deadline.
Atlas's September filing about the AI lab is a few paragraphs long, and there is no reason it should mention what one of its directors tells Congress. The reports that carry his forecasts are another matter.
