Anthropic’s Data Centres Lean on Methane as AI Race Heats Up

Anthropic’s S‑1 documentation lays bare the energy profile of its next‑generation language models, showing that a sizable fraction of its compute load is powered by gas‑fired generators situated in regions with abundant methane supplies. The company argues that the choice offers a stable and cost‑effective power source amid fierce competition for renewable capacity, yet the disclosure highlights a paradox where cutting‑edge AI development fuels greenhouse‑gas emissions.

Regulators in the United States and Europe have begun to scrutinise the environmental footprints of large‑scale AI training, proposing metrics that would require firms to disclose the carbon intensity of each model iteration. Environmental groups warn that without decisive policy action, the sector could lock in a generation of high‑energy consumption patterns that outpace current climate mitigation pathways. Investors are also weighing the reputational risk of backing firms whose growth hinges on fossil‑fuel energy.

In response, Anthropic has signalled an intention to diversify its power mix, exploring partnerships with renewable‑rich data‑centre operators and investigating emerging cooling technologies that reduce overall energy demand. The company’s trajectory will likely influence how other AI enterprises balance performance ambitions with sustainability imperatives, potentially reshaping the economics of future model training.