AI Energy Use: Is the Outrage Justified? Comparing with Daily Activities

Debate surrounds the energy consumption of artificial intelligence (AI), with critics claiming it rivals that of a small nation. Recent data, however, suggests that the actual energy cost of a typical AI prompt may not be as alarming as some have suggested. Google has released figures indicating that a median text prompt for its Gemini AI consumes approximately 0.24 watt-hours (Wh). This figure allows for a clearer comparison with other everyday energy uses, such as streaming video or gaming.

Understanding AI’s Energy Consumption

To put the energy usage into perspective, a single AI prompt’s consumption is equivalent to 1.5% of a charge for a new iPhone 17 or less than 10 seconds of streaming video on a 55-inch television. Notably, during typical video streaming at home, 99.97% of the total electricity consumed is attributed to the television itself, while the data center’s contribution amounts to a mere 0.03%. This trend holds across devices: 99.6% of the power is used by laptops and 98.4% by mobile phones when streaming.

The figures released by Google reflect only the energy consumed by the data centers. When considering the end devices—like televisions and laptops—the majority of electricity usage is not attributable to the AI operation itself. In a broader context, a user’s daily interaction with AI—averaging between 10 to 20 prompts—translates to about 0.03% of their total daily electricity consumption.

Scaling Up: The Bigger Picture

While individual AI prompts represent a small fraction of overall energy use, the implications change dramatically when considered at scale. OpenAI handles over 2.5 billion prompts daily, which could lead to significant cumulative energy consumption. For heavy users, averaging around 50 prompts a day, the contribution to total electricity use could reach 0.15%, comparable to the energy used by a television in standby mode.

Moreover, Google’s analysis has highlighted that each AI prompt also requires around 0.26 milliliters of water for cooling and generates 0.03 grams of carbon dioxide equivalent emissions. This is a minimal amount, roughly equivalent to the carbon footprint of a single sip of soda, but it illustrates that AI’s environmental impact extends beyond just electricity usage.

These figures raise questions about the real energy costs of AI when juxtaposed against other daily activities. While the energy consumption associated with AI is a valid concern, it appears that the outrage may be overstated compared to other common tasks.

The upcoming second installment of this series will delve deeper into the global impact of AI’s energy usage, providing a more comprehensive overview of how billions of prompts accumulate in terms of resource consumption.

As the conversation around AI continues, it is crucial to consider not just the numbers but also the context in which they are used. AI’s energy footprint, while not negligible, is only one part of a larger picture that includes various aspects of modern life.