Zendoric
← Back to the day · October 28, 2026

Anthropic's S-1 admits its AI depends on natural gas: AI's bottleneck is energy, not chips

🕒 Published on Zendoric: October 28, 2026 · 08:22

✨ AI-generated · how it's made

In its IPO filing, Anthropic acknowledges that its data centers will rely more and more on natural gas and that global warming will raise its operating costs. This regulatory candor shows where the real limit on AI growth lies.

🎧 Listen to the analysis (in Spanish) · synthetic voice

By The Verge · October 26, 2026.

Anthropic has filed its S-1, the document that precedes a stock market listing, with the U.S. Securities and Exchange Commission (SEC). According to The Verge, which analyzed the filing, the company acknowledges that its data centers increasingly rely on "natural gas" (the commercial term for methane, its main component), leaving its expansion exposed to fuel prices and supply chain constraints.

The S-1 text quoted by the outlet is blunt: "competition for access to power supply and to sites with adequate power is intense." Anthropic adds that some of its planned facilities will use electricity from gas turbines generated on site. The company rents capacity from major cloud providers (hyperscalers), but last year it announced a $50 billion plan for data centers in Texas and New York.

The filing also reflects a paradox: climate change, which gas itself worsens, will raise the cost of electricity and of the water used to cool its servers. It also anticipates that competition from other hyperscalers for gas and pipelines "could raise the price of natural gas."

The S-1 further identifies a regulatory and social risk. If its power supply is perceived as harming other grid customers, regulators could impose conditions or additional charges. According to The Verge, disgruntled residents have already blocked dozens of data center projects in the U.S. Anthropic warns that without secured gas and pipelines it could delay or abandon projects, and that any failure to access energy or any cost increase would limit its ability to scale.

As context provided by the outlet: gas accounts for 40% of U.S. electricity generation, the country plans to build more fossil capacity than it has in a decade, and the turbine shortage is forcing some developers to install simpler engines and turbines on site, which are less efficient and more polluting. A Texas-based study cited by The Verge argues that the electricity demand of these facilities could erase decades of air quality improvements within five years.

**Our take.** What matters is not that an AI company uses gas; that was already known. What matters is that an S-1 forces a company to state in writing, under legal liability, what could hold back its business. And among the things Anthropic lists there, explicitly, are megawatts. AI's bottleneck is shifting from silicon to the power grid.

That redefines who wins. The big cloud players are competing for the same gas and the same grid connections, and whoever secures energy first wins: capital and access to infrastructure matter more than the best model architecture. It is another route to concentration of power, the kind worth watching in the short term. Residents who fear higher bills also lose, and with political justification: social license is now a cost variable.

We are optimistic in the long term, and this fact does not contradict that, but it demands honesty about the transition. The promise of eradicating diseases and generating abundance requires abundant, clean energy; building that foundation today on makeshift, more polluting turbines is a real debt. We distinguish aspiration from demonstrated capability: AI has not yet shown that its contribution offsets this environmental cost.

The sensible way forward is to accelerate clean, firm supply (nuclear, geothermal, renewables with storage) and to have data centers pay for their own infrastructure so the cost is not passed on to households. If gas is a bridge, it must have an end date. That Anthropic acknowledges this before the SEC is, paradoxically, a useful first step: what is declared as a risk can be measured, demanded and governed.

🔗 Related on Zendoric

Sources & references