Anthropic's IPO Filing Shows an AI Company Hooked on Fossil Fuels

Anthropic wants investors to know it's betting big on AI. Its own paperwork shows that bet runs on gas.

In the S-1 registration it filed with the US Securities and Exchange Commission Monday ahead of a planned public offering, Anthropic lays out in blunt terms how dependent its data center expansion is on natural gas, the fossil fuel industry's term for what is mostly methane. The company says some of its planned data centers will draw power from on-site natural gas turbines, a workaround for a simple problem: the electrical grid can't keep up with what AI needs.

"Competition for access to electrical supply and data center locations with adequate power availability is intense," the filing states.

That's not a small admission for a company valued on its ability to scale. Anthropic leases most of its computing capacity from hyperscalers, the Amazons and Googles of the cloud world, but it's also building its own footprint. Last year it announced a $50 billion plan to expand data centers in Texas and New York. Getting those sites powered, apparently, increasingly means burning gas on-site rather than waiting for grid connections that may take years.

Anthropic isn't alone in that math. The US already generates more electricity from natural gas than from any other source, at 40 percent of the national mix, and the country is on track to add more fossil fuel power capacity than it has in a decade, driven largely by data center demand. The Trump administration has leaned into that trend, pulling back support for renewable energy projects while easing the path for fossil fuel development, arguing gas and other always-on sources are better suited to round-the-clock data center operations than intermittent wind and solar.

But the gas itself is getting harder to come by, at least in usable form. A shortage of gas turbines is slowing construction timelines across the industry. Some developers are responding by installing smaller gas engines on-site to skip the turbine wait entirely, or switching to simpler turbine designs to dodge manufacturer backlogs. Both workarounds tend to be less efficient and dirtier than the equipment they're replacing, which cuts against the idea that going around the grid is a clean shortcut.

The irony Anthropic can't avoid

Here's where the filing gets uncomfortable for Anthropic specifically. The company flags climate change as a financial risk to its own business, warning that rising global temperatures will push up costs for the electricity and water it needs to run and cool its servers. That's a direct acknowledgment that a warming planet makes AI more expensive to run, even as AI's own power demands lean harder on the fuels driving that warming.

It's a loop Anthropic doesn't resolve in the filing, and can't really be expected to. The company isn't claiming to be a climate-friendly operation; it's disclosing risk to shareholders the way securities law requires. But seeing the dependency and the downside written out in the same document underscores how tightly AI's growth and fossil fuel consumption are now linked, at least for now.

The research backs that up on a broader scale. A recent Texas-based study found that data center electricity demand could erase decades of air quality improvements within five years, pushing up soot, smog-forming pollutants, and carbon dioxide, according to reporting on the study.

Anthropic's own numbers add another wrinkle: it expects to be competing with other AI companies and cloud providers for the same limited gas and pipeline capacity as everyone races to build out data centers simultaneously. "This competition could drive up the price of natural gas," the filing says, which means Anthropic is warning investors that the AI industry's collective gas habit could make its own fuel more expensive.

The backlash is already showing up in the filing

Rising fuel costs aren't the only friction point. Residents near proposed data center sites have successfully blocked dozens of projects across the country, often over fears that new facilities will spike local utility bills. Anthropic addresses that risk directly, warning that if its efforts to secure power are seen as burdening other customers or the broader grid, regulators could respond with standby charges, usage conditions, or other restrictions.

The company also flags a related possibility: if regulators start requiring data center developers to cover more of the cost of new energy infrastructure, that would raise Anthropic's own operating expenses. And without secured gas supply and the pipeline capacity to move it, the filing says, Anthropic may have to delay or scrap data center projects outright.

Taken together, the filing frames energy access, not chips or model performance, as one of the central constraints on Anthropic's growth. As the company puts it, any failure to secure adequate electrical power, or any regulatory action limiting power availability, "could adversely affect our cost structure" and limit its ability to scale compute infrastructure, build models, and serve customers.

That's a notable thing for an AI company to put in writing on its way to public markets. The industry's public narrative tends to center on model capability and compute scale. Anthropic's own disclosure suggests the more immediate bottleneck is something far more mundane: finding enough power, buying enough gas, and surviving the local and regulatory pushback that comes with both.