Tool-using AI makes energy accounting harder than chatbot counts
The mechanism is simple. One user request can become many model calls, tool steps, and background helpers. That does not create one fixed energy multiplier. It does make one-question comparisons too small for judging the data-center buildout behind these products.
What the source reports
WIRED's Molly Taft connects the data-center buildout to agentic AI. An agent is an AI system that uses tools to carry out tasks instead of answering once and stopping. The article quotes Maxwell Zeff saying a website-building agent may run for hours. It may ask itself dozens of smaller prompts to make pages, menus, and datasets. The story also names OpenAI's math swarm and Meta's Muse as examples. These workloads can grow with task difficulty, not just with user count. The strongest evidence is about mechanism and disclosure. Companies often cite simple one-query costs, while task sizes vary and reliable energy data is scarce.
Original source
- Title
- AI Agents Are Thirsty for Power
- Author
- Molly Taft
- Publication
- WIRED
- Date
- Sunday, September 13, 2026