It’s way more than you think.
Power Hungry
It’s an open secret that generative AI tools like OpenAI’s ChatGPT eat up an astronomical amount of power.
Even generating images from a text prompts with a tool like Midjourney or OpenAI’s DALL-E is immensely power-intensive.
As first spotted by The Register, a team of researchers from AI developer Hugging Face and Carnegie Mellon University recently shared a yet-to-be-peer-reviewed paper on how much power AI tools need to perform a variety of tasks.
Their results are — and highlight the very real carbon footprint of turning to AI instead of a human artist for imagery needs.
Strikingly, they found that the “least efficient generation model” they studied — Stable Diffusion’s open source XL, which was released in July — used almost as much power per image as that required to fully charge a smartphone.
Coming up with 1,000 images using this model generates the carbon emission equivalent of “4.1 miles driven by an average gasoline-powered passenger vehicle,” according to the paper.
Hidden Costs
However, due to the “large variation between image generation models,” that number can also be smaller. Overall, across all models the researchers tested, generating 1,000 images took an average of 2.907 kWh, roughly the equivalent of charging a phone’s battery to 24 percent per image.
Generating text, however, is a seemingly far less power-hungry process and only used as much as the equivalent of three smartphone charges for 1,000 queries, the researchers found.
But extrapolating the data to a global scale reveals an even uglier truth. Companies like OpenAI and Google are already facing rapidly growing energy bills as they attempt to keep their generative AI tools online.
Recent estimations suggest AI servers on a global scale use the equivalent of what an entire country like Argentina uses in a year.
Just cooling these servers alone has an astonishing environmental footprint. According to Google’s 2023 Environmental Report, the company used an astronomical 5.6 billion gallons of water last year, a 20 percent increase over its 2021 usage.
In short, the AI industry’s carbon footprint will continue to be a big problem, especially as the world creeps ever closer to a climate catastrophe.
The latest research serves as a reminder that even on an image-by-image basis, the energy costs of using these generative AI tools can be considerable.
It’s unclear, though, how these results compare to more commonly used AI image generators like Midjourney or OpenAI’s DALL-E, which weren’t part of the study
More on generative AI: AI’s Electricity Use Is Spiking So Fast It’ll Soon Use as Much Power as an Entire Country
Share This Article