There is something almost ironic about a company whose mission is built on the promise of advanced, beneficial artificial intelligence warning that its own future might be tripped up by something as old-fashioned and unglamorous as natural gas prices. But that is exactly what Anthropic did in its initial public offering filing on Monday. The company, one of the most closely watched names in AI, cautioned investors that the rapid growth of data centers across the United States is creating a huge new demand for natural gas, and that many data center operators — including Anthropic itself — are increasingly turning to private, on-site gas plants to power their servers. These facilities are often built behind the electricity meter, meaning they bypass the traditional grid queue and generate power directly next to the data center. On paper, that sounds like a pragmatic solution to a very modern problem: how do you power millions of servers when the electrical grid is too slow, too burdened, or too expensive to expand? But in practice, it means the AI boom is now tied to the volatile, globalized, and often unpredictable market for fossil fuels. Anthropic’s filing warns that this reliance on private natural gas plants is a key near-term risk, and that if enough other companies do the same thing, the resulting competition for gas could drive prices higher in certain regions. That would affect not only Anthropic’s own ambitions, but also the broader data center industry and the communities where these giant facilities are being built.
The scale of this shift is difficult to wrap one’s head around. Energy analyst company Cleanview estimates that behind-the-meter gas projects in development as of October 2026 stood at 110 gigawatts, up from just 20 gigawatts in July 2025. That is a more than fivefold increase in roughly a year, and it represents a sea change in how the technology industry thinks about energy. In the past, data center developers mostly assumed they could plug into the existing electric grid, perhaps building a new substation or negotiating a power purchase agreement. But now, with grid interconnection queues stretching for years, and with demand for electricity from everything from electric vehicles to factories growing at the same time, many companies have decided they simply cannot afford to wait. Instead, they are building their own natural gas plants, sometimes tiny and modular, sometimes utility-scale, right next to their server halls. These facilities are often called “behind-the-meter” because they are on the customer’s side of the electric meter, and they give the operator direct control over power supply. But they also mean that the tech sector is no longer just a buyer of electricity; it is becoming a major consumer of raw natural gas, with all the price risk, supply chain exposure, and climate consequences that come with it. These are not abstract statistics. They represent real decisions being made right now by developers in Texas, Louisiana, Virginia, and beyond, often on rural land where landowners are being offered large sums of money and communities are grappling with the sudden arrival of giant, loud, and pollution-heavy industrial facilities.
Even before data centers arrived, the natural gas market in the United States was already tight. American export terminals are sending record amounts of gas overseas, driven by a mix of geopolitical conflicts, international energy insecurity, and federal support for expanding exports. The result is that gas is no longer just a domestic fuel; it is a global commodity, and its price is influenced by events thousands of miles away. Over the next decade, export demand is expected to double, which means more competition for every molecule of gas produced in the Permian Basin or the Marcellus Shale. Now add data centers to that equation. Even under conservative assumptions, the new demand is staggering. BloombergNEF estimates that grid-connected data centers alone could add an additional 15 billion cubic feet per day of natural gas demand by 2035, a 15 percent increase over current overall U.S. consumption. If you include a conservative estimate of on-site gas capacity in the pipeline, that adds another 3 billion cubic feet per day by the next decade. To put that in perspective, a single billion cubic feet of gas per day is enough to supply millions of homes. The explosion of AI infrastructure is therefore not just a technological story; it is an energy story, and it is happening at a time when the power system is already stretched. It’s a bit like showing up to a crowded restaurant at peak hour and demanding a table for a thousand, then wondering why the kitchen is stressed and the prices on the menu are climbing.
Anthropic’s own journey makes this tension especially visible. For most of its life, the company has relied on renting compute from major cloud service providers rather than owning its own data centers. Amazon, Google, Microsoft, and SpaceX are all named in the IPO filing as partners or suppliers, and the company has rented server space from SpaceX’s Colossus data centers in Memphis, Tennessee. That last relationship is particularly striking, because the Memphis facility is currently facing a lawsuit over its reliance on unpermitted gas turbines. The image of one of the world’s most innovative AI companies running its models on servers powered by contested, backup-style gas generators is a reminder that there is no clean, simple way to power AI at this scale. For a long time, Anthropic was insulated from the physical realities of data center construction, energy procurement, and fuel price volatility because it was renting someone else’s machines. But that insulation also meant dependence, and dependence on a handful of giant technology companies has its own risks. Cloud providers can raise prices, change terms, prioritize their own internal AI projects, or simply run out of capacity during the next surge in demand. Anthropic’s move to reduce that reliance is understandable. Yet in trying to free itself from one kind of dependency, it has walked directly into another: the messy, hard, and increasingly expensive world of natural gas infrastructure.
Over the past year, though, Anthropic has changed course, moving to directly lease and develop data centers in an effort to exert more control over its own destiny. In its IPO filing, the company highlights a $50 billion partnership with Fluidstack, announced last year, which provides data center development and operations services. It has announced data center sites in Texas and New York, and its name has been tied to developing facilities with major on-site gas projects in states as spread out as Louisiana, Texas, and West Virginia. According to developer Nscale’s own IPO filing, the West Virginia project could eventually be among the largest gas plants in the United States. That is a massive bet on a very particular kind of infrastructure. Building your own data centers gives you more control over supply chains, security, and performance, and it offers the promise of being able to design everything from the cooling systems to the electrical layouts exactly the way you want. But it also means taking on the headaches that come with being a developer, a utility, and an energy buyer all at once. You have to negotiate long-term gas contracts, worry about pipeline capacity, face the possibility of fuel supply disruptions, and deal with environmental permitting and community opposition. In its own words, Anthropic says it is exposed to “fuel price volatility and supply disruptions that are outside of our control,” and that competition from other gas-fired data centers could “drive up the price of natural gas in such areas and may adversely affect our ability to contract for firm fuel supply.” This is not just a legal disclaimer; it is a real operational challenge, and it is happening right now across the country.
What does this mean for the rest of us? It means the future of AI is deeply entangled with the future of fossil fuels, at least for the next decade or so. The data center boom is not just a story about silicon, software, and algorithms. It is a story about land, water, pipelines, power plants, and the communities forced to live next to them. Every time an AI model is trained, every time a chatbot answers a question, every time a video is generated or a document is summarized, somewhere a machine is burning fuel to keep the servers running. The industry is working on alternatives, including nuclear power, solar, wind, and battery storage, and some of those efforts are genuinely promising. But in the near term, the numbers point overwhelmingly to natural gas, and that has serious implications for climate change, for local air quality, and for the affordability of energy for ordinary people. If natural gas prices rise because data centers are competing with homes and hospitals for fuel, the burden will not fall equally. It will fall hardest on people who can least afford it, on electric ratepayers, and on rural communities that suddenly find themselves hosting industrial facilities the size of small towns. So when Anthropic warns about the risks of its own dependency on natural gas, it is also warning us about something larger: the cost of building the future, and the uncomfortable truth that some of the most advanced technology humanity has ever created still runs, for now, on the same ancient sunlight that has powered civilization for centuries. We have to choose how to navigate that reality, and we have to make those choices honestly, with open eyes, and with a real understanding of what is being built and at what price.
