Anthropic's filing reveals heavy reliance on natural gas for AI growth

The fossil fuel reality

As Anthropic prepares to go public, its S-1 filing with the US Securities and Exchange Commission reveals a significant dependency on fossil fuels. The company acknowledges that data centers are increasingly reliant on natural gas, which is primarily methane. This reliance is a direct result of the intense competition for electrical supply and locations with enough power to support AI.

To keep up with demand, some of Anthropic's planned data centers will use electricity generated by natural gas turbines located on-site. This move allows the company to bypass long wait times to connect to existing power grids. However, using these on-site engines can be less efficient and more polluting than other methods because of current manufacturing backlogs.

The United States is currently aiming to construct a higher volume of fossil fuel power generation than it has in the last ten years to meet this surge in demand. This shift is being driven by the need for power that can run around the clock. While solar and wind generate power intermittently, fossil fuels provide a steady stream of electricity required for constant AI computing.

A climate-related catch-22

Anthropic's filing highlights a paradoxical situation where the company's growth contributes to the very problems that threaten its bottom line. The company admits that rising global temperatures, which are a symptom of burning fossil fuels, will increase its operating costs. Specifically, higher temperatures mean the company will have to pay more for electricity and water to cool its servers.

This creates a cycle where the energy used to power AI models contributes to a warming planet that makes running those models more expensive. The company notes that it expects increased competition for natural gas and pipeline capacity from other hyperscalers. This competition from other cloud service providers could drive up the price of natural gas for everyone in the industry.

The company also notes that data center electricity demand could wipe out decades of gains in air quality over the next five years. A recent study from Texas found that this demand could lead to higher levels of soot and smog-forming pollutants. These environmental impacts are becoming a primary concern for regulators and the general public alike.

Infrastructure and regulatory hurdles

The path to building these data centers is fraught with logistical and social obstacles. Anthropic mentions that local residents who are angry with the projects have successfully stopped dozens of proposed data centers across the United States. These projects often face pushback due to concerns about their impact on the local environment and the broader power grid.

Regulators may impose specific conditions or standby charges if they perceive that data centers are negatively affecting other utility customers. Anthropic warns that if developers are forced to shoulder more of the costs for new energy infrastructure, it could raise their operating costs. These administrative and legal hurdles are now a formal risk factor for the company's public offering.

The company also faces a physical shortage of gas turbines, which is currently slowing down construction. Without a secure supply of gas and the pipelines needed to transport it, Anthropic admits it might have to delay or abandon planned projects. These constraints could ultimately stunt the company's ability to scale its compute infrastructure and service its customers.

Impact on the AI economy

The reliance on natural gas makes the growth of generative AI vulnerable to the volatility of the fossil fuel industry. Because gas makes up 40 percent of the nation’s power generation, any fluctuation in fuel prices or supply chain constraints will directly affect Anthropic's costs. This means the price of AI services could be influenced by global energy markets as much as by technological progress.

Furthermore, the company's S-1 filing makes it clear that these energy challenges could limit its ability to develop and deploy new models. If the costs of compliance or energy become too high, the company may find it difficult to maintain its current pace of expansion. This creates a ceiling for how quickly AI can be scaled in the current energy landscape.

As more companies rush to build their own data centers, the scramble for resources will only intensify. Anthropic's transparency about these risks suggests that the AI boom is facing a physical bottleneck. The transition to a more sustainable energy source remains a significant hurdle for the industry to overcome while trying to meet the massive demands of modern computing.