Generative AI tools, from image generators like Stable Diffusion and DALL-E to advanced chatbots, consume a tremendous amount of power—a reality that’s often overlooked when we marvel at their capabilities. A recent study conducted (PDF version) by researchers from Hugging Face and Carnegie Mellon University put this energy usage into perspective, highlighting both the environmental cost and the disparities between AI tasks.

For instance, generating a single image using AI, specifically with Stable Diffusion XL, consumes roughly the same power as charging a smartphone battery. Multiply that by 1,000 images, and the energy usage jumps significantly—equating to the carbon emissions of driving a gasoline car for around four miles. On average, generating 1,000 images required nearly 3 kWh, enough to charge a typical smartphone to 24% per image.

The researchers noted a stark contrast in energy consumption when comparing text generation to image creation. Text generation is significantly less power-intensive, with 1,000 queries consuming the energy equivalent of just three smartphone charges. This difference is due to the inherent complexity of image data compared to text, which makes image generation a more resource-heavy task.

These findings underscore the growing environmental footprint of AI. Giants like OpenAI and Google, which rely heavily on data centers to support AI operations, are witnessing skyrocketing energy demands. The energy used to power AI servers worldwide now rivals the consumption of entire nations. Google’s data centers alone consumed billions of gallons of water last year just to keep their systems cool, reflecting the substantial environmental impact of AI infrastructure.

With AI’s popularity soaring, the need for transparency about its environmental effects is becoming critical. Researchers are urging AI developers to be more upfront about the carbon footprint of their models, especially as we edge closer to global climate tipping points.

This latest research didn’t cover some of the most popular AI image generators like DALL-E, so the exact numbers may vary, but the message is clear: the power costs of AI are significant and warrant careful consideration as the technology becomes more embedded in daily life. The carbon footprint for producing stunning images on-demand might be far greater than many realize, urging a balance between technological advancements and environmental stewardship.