The energy consumption of artificial intelligence has become a hotly debated topic. While individual AI interactions may seem trivial, the aggregate energy use of AI agents is substantial and often underestimated. In 2026, understanding the true energy costs of AI agents is crucial for both consumers and businesses.
Recent studies have shown that AI data centers could account for around 12% of US electricity use by 2030. However, the energy impact of individual AI interactions has been downplayed, with estimates suggesting that a single chatbot prompt uses minimal energy. For instance, Google reported that a median Gemini text prompt uses only 0.24 watt-hours (Wh), comparable to watching nine seconds of television. Similar figures were cited by Sam Altman for ChatGPT and by Epoch AI. These numbers suggest that individual AI interactions have a negligible impact on energy consumption.
The Reality of AI Agent Energy Use
The reality, however, is more complex. The fastest-growing use of AI involves agentic workflows where AI agents perform tasks such as writing and executing code, reading results, and iterating on their own. These agents make dozens of model calls per human prompt and engage in complex reasoning chains. A deep dive into the energy use of AI agents reveals a stark contrast to the energy estimates for individual chatbot prompts.
Over an 8-week period, one user typed 1,138 prompts into Claude Code, triggering more than 14,000 model calls that processed 3.2 billion tokens. The estimated energy use for this activity was around 170 kWh of data center electricity, with a range of 70 to 330 kWh. This translates to approximately 150 Wh per prompt, which is roughly 600 times the energy of a median chat prompt. This discrepancy highlights the significant energy consumption of AI agents compared to individual chatbot interactions.
The Impact of AI Agents on Energy Consumption
The energy consumption of AI agents is not just a matter of individual interactions but also of the tasks they perform. A new white paper from Watershed (Bistline et al. 2026) proposes a standardized framework for corporate AI emissions accounting. The paper highlights that electricity per AI task spans more than five orders of magnitude, from thousandths of a watt-hour for text classification to 50-500 Wh for an agentic workflow making 5-50 frontier model calls.
Other researchers have found similar results. Bai et al. (2026) measured coding agents on real software tasks and found they consume roughly 1,000 times the tokens of an ordinary chatbot interaction. These agentic tasks represent the most rapid driver of increased AI usage. Anthropic’s Economic Index found that 97% of their API usage now shows automation-dominant patterns associated with agents.
Comparing AI Agent Energy Use to Common Household Appliances
To put these values into perspective, the energy use of AI agents can be compared to common household appliances. The median Claude Code session uses around 0.6 kWh, which is at the top end of Watershed’s generic agentic usage estimate and fifty times the energy used to charge a cellphone. An average day of Claude Code use (3.0 kWh) consumes more electricity than running two refrigerators.
The energy consumption of AI agents is not just a theoretical concern but a practical one. As AI usage continues to grow, understanding and managing the energy impact of AI agents will be crucial for both individuals and businesses. The energy costs of AI agents are a hidden but significant factor in the
