AI Agents Are Thirsty for Power

Many people are puzzled by the massive investments in data centers that tech companies are making as artificial intelligence becomes more advanced. Simple chatbot queries are becoming less relevant as AI increasingly relies on agents to make autonomous decisions and carry out tasks. These agents, which are based on large language models, can autonomously generate hundreds of prompts based on a single user request, like building a website.

Recently, OpenAI's agents solved a complex math problem, though some mathematicians questioned the company's claims. Solving such problems required significant processing power, likely costing tens of millions of dollars. AI companies often avoid disclosing environmental metrics, but some believe agents use far more energy than individual queries.

For instance, OpenAI CEO Sam Altman suggested an almond uses the equivalent of 38,000 ChatGPT queries, though this claim was disputed. Few details are available about the energy consumption of agents, which can range from minor tasks to full days of coding involving multiple "helper" agents. The potential for a large number of AI agents to operate in the background is enormous, posing a challenge for understanding and measuring energy use.

While personal use of AI agents may currently have a relatively small environmental impact, the widespread adoption of personal AI agents could lead to significant emissions.