Anthropic, the artificial intelligence company founded by former OpenAI employees, is facing a reality that few in the industry want to admit aloud: the path to training increasingly capable AI models is paved with natural gas. In documents filed with the U.S. Securities and Exchange Commission (SEC) for its initial public offering (IPO) process, Anthropic explicitly acknowledged that its data centers are becoming increasingly dependent on natural gas-fired turbines — a choice that puts the company in direct conflict with the sustainability promises that have defined its brand.
The disclosure, first published by The Verge, is one of the clearest early signals of how deep the AI industry's commitment to fossil fuels has become. Unlike other tech companies that try to hide this dependency in footnotes of sustainability reports, Anthropic placed it at the center of its securities registration filing — exactly where investors need to read it.
The scale of the problem
What Anthropic describes individually is part of a much broader trend. According to data from Global Energy Monitor, natural gas-fired power generation capacity in the United States dedicated to data centers nearly doubled in just six months, rising from 252 gigawatts to 378 gigawatts in the first half of 2026. Texas leads this growth with 122 gigawatts in development, representing a 51 percent increase in just half a year.
According to Goldman Sachs, approximately 60 percent of data center energy demand will be met by natural gas by 2030, with the remaining 40 percent coming from renewable sources. This represents a significant shift from the initial commitment of many technology companies to operate exclusively on renewable energy.
Anthropic's projects
Anthropic has been particularly active in seeking natural gas capacity. In August 2026, the company announced a $50 billion investment in U.S. AI infrastructure, building data centers with Fluidstack in Texas and New York. More recently, in October, it signed a lease agreement with EverWind for capacity in Point Tupper, Nova Scotia — Anthropic's first data center agreement in Canada.
But the most revealing project is in Bastrop County, Texas, where Anthropic is negotiating to lease a data center campus backed by a 2,842-acre natural gas-fired power generation facility. The project, involving BlackChamber Partners and Pacifico Energy, will offer 710 megawatts of power using simple-cycle natural gas turbines to support 490 megawatts of IT capacity.
Environmental impacts
The effects go beyond carbon emissions. According to a report published by the Houston Advanced Research Center, nitrogen oxide emissions in Texas could rise 35 percent by 2031 — even in a "net zero" scenario where data centers connect to the grid and deploy local solar and wind power.
The Texas Tribune report highlights that pollutants generated by gas turbines include not only carbon dioxide, but also fine particulate matter (PM2.5), tropospheric ozone, and other contaminants associated with respiratory problems, cardiovascular disease, and premature deaths. In Memphis, Tennessee, where xAI operated 41 unauthorized gas turbines to power its Colossus 1 data center, the local community has already recorded a significant increase in asthma cases and respiratory illnesses.
The brand paradox
The irony is that Anthropic has positioned itself as a "safe" AI company from the beginning, with its declared mission to ensure that artificial intelligence benefits humanity. On its website, the company states that AI development should be "transparent, safe, and beneficial to all of society." But by relying on gas turbines to power its data centers, Anthropic is contributing directly to the degradation of air quality in already-vulnerable communities.
The company has tried to mitigate this perception by announcing in 2026 that it will cover electricity price increases that consumers face from its data centers — a generous financial promise, but silent on actual emissions. According to Anthropic's website, the company committed to paying 100 percent of the grid upgrade costs, but has not committed to eliminating the use of natural gas itself.
What this means for the future of AI
The question is not just ethical: it is operational. If Anthropic itself — a company positioned in the middle of the spectrum between the more aggressive OpenAI and the more conservative Google on regulation — is turning to natural gas on a large scale, what does that say about the inevitable environmental cost of AI progress?
The answer may lie in the very nature of training. A single frontier model can require gigawatts of energy for training. And while Anthropic has significantly reduced its use of internet connectivity for training its most recent models — cutting internet access to prevent data leaks — the energy cost of this isolation is even higher, requiring even more intensive local storage and processing infrastructure.
What becomes clear is that the energy transition for artificial intelligence will not be as clean as many investors and enthusiasts hoped. Natural gas may be the "transition fuel" the industry needs in the short term, but the emissions it generates are being counted now — and the communities living next to these data centers are the ones with the least voice in this conversation.
Sources: The Verge, Fortune, The Texas Tribune, Goldman Sachs, Global Energy Monitor
✓ Independent sources cross-checked and verified before publishing
