Anthropic filed its S-1 with the SEC on Monday, and buried in the hundreds of pages of financial disclosures and risk factors is a stark admission: the AI boom runs on natural gas, and there isn’t enough electricity to go around.

“Competition for access to electrical supply and data center locations with adequate power availability is intense,” the company wrote. It’s the kind of line that sounds like boilerplate risk disclosure until you realize what it means. Anthropic, which has built its brand partly on being the responsible AI company, is now telling prospective investors that its business depends on burning methane to keep the servers running.

The filing doesn’t dance around it. Data centers are “becoming increasingly reliant on ‘natural gas,’” Anthropic states, using the fossil fuel industry’s preferred term for what is mostly methane, a greenhouse gas roughly 80 times more potent than CO2 over a 20-year period. This isn’t a future problem or a hypothetical concern. It’s happening now, at scale, as Anthropic and its competitors race to build out the infrastructure needed to train and run increasingly large AI models.

The timing is blunt. Anthropic is going public just as the infrastructure costs of the AI race are becoming impossible to ignore. The company reported soaring revenue but mounting losses, according to coverage of the filing. It has major enterprise adoption. It also has an energy problem that money alone can’t solve.

The power crunch is real enough that it made it into the risk factors section of a securities filing. That’s significant. Companies don’t volunteer information about competitive disadvantages in S-1s unless they’re legally required to disclose material risks. Anthropic is telling investors that access to electricity is now a limiting factor for growth, right alongside things like regulatory uncertainty and model safety.

The local backlash

The data center buildout isn’t just an energy policy problem. It’s becoming a voting issue. Communities across the country are pushing back against the massive facilities and the infrastructure projects that come with them, and it’s showing up in the 2026 midterms.

Data center opposition is “uniting America,” according to reporting on congressional and gubernatorial races where the issue has become central. The backlash crosses typical political lines. People don’t like massive construction projects that stress local power grids and water supplies, regardless of party affiliation.

At the same time, tech money is flooding into these races. The Verge reported that Leading the Future, a super PAC backed by venture capitalist money, has been active in many of the contested districts. The dynamic is straightforward: AI companies need data centers, data centers need power and real estate, and communities where they want to build are saying no.

This isn’t a typical NIMBY fight over traffic or property values. The scale is different. A single large data center can consume as much electricity as a small city. When Anthropic talks about “intense” competition for power supply, this is what it means on the ground. Every AI company is looking for the same thing: locations with available electricity, water for cooling, and fiber connectivity. There aren’t that many places that have all three.

What the filing means

Anthropic’s S-1 also reveals “an unusual governance structure designed to preserve founder control and its AI safety mission,” according to TechCrunch’s coverage. That’s interesting in context. The company is going public while acknowledging that its ability to execute on that mission depends on access to infrastructure that requires burning fossil fuels.

The contradiction isn’t lost on anyone paying attention. Anthropic has positioned itself as the AI company that takes long-term risks seriously, including existential risk from advanced AI. But the near-term risk, the one detailed in the S-1, is more mundane: can we get enough electricity to keep training models?

Some analysts aren’t satisfied with what the filing reveals. “Anthropic owes investors more information,” one told WIRED, which reported that the prospectus leaves out quite a few financial details despite running hundreds of pages. The missing numbers matter because they would show just how much the power dependency actually costs.

What happens next

The S-1 filing is just the beginning of the IPO process. Anthropic will face more questions from investors and regulators, and the energy issue isn’t going away. If anything, it’s getting worse. Every AI company is in the same race, competing for the same constrained resources.

The legal and regulatory response is still taking shape. Data center permitting is mostly a local and state issue, which is why it’s showing up in midterm races rather than as federal policy. But the energy question is broader. When AI companies tell investors that their growth depends on natural gas, they’re also telling policymakers that the AI boom has locked in fossil fuel demand for the foreseeable future.

That creates tension with climate commitments, and eventually it creates legal exposure. Anthropic’s competitors face the same calculation. The question is who gets stuck with the bill when the infrastructure costs and climate risks start showing up in shareholder lawsuits or regulatory enforcement actions.

For now, Anthropic is being honest about the dependency. The company that talks the most about AI safety is also the one telling investors, clearly and on the record, that it can’t build what it wants to build without burning methane. That’s not a risk factor that gets resolved with better alignment research or more careful deployment. It’s a constraint that shapes everything else.