You may have encountered the assertion that each ChatGPT inquiry utilizes as much water as a water bottle. While this is technically true, it oversimplifies a far more intricate issue.

A recent comprehensive analysis from MIT Technology Review illuminates the actual energy requirements of the artificial intelligence sector — especially services such as ChatGPT. The report details the energy consumption of large language models (LLMs) for each response, ranging from 114 joules to 6,706 joules. To put this into context, that’s the energy equivalent of operating a microwave for a brief moment to roughly eight seconds. Models that utilize less energy generally have fewer parameters, which may also mean their answers are less precise.

Regarding AI-generated video, energy demands surge dramatically. The report indicates that producing a five-second video with a contemporary AI model demands approximately 3.4 million joules — over 700 times the energy required to create a high-quality image. This is roughly equivalent to conducting a microwave for more than an hour.

To exemplify the aggregate impact, researchers estimated the energy required for a hypothetical user session comprising 15 chatbot queries, 10 image generations, and three five-second videos. The total energy usage? Around 2.9 kilowatt-hours — comparable to running a microwave for more than three and a half hours.

The report further emphasizes the escalating energy requirements of the data centers powering AI. Prior to the AI surge, data center energy consumption remained relatively constant due to enhancements in efficiency. However, since 2017, the emergence of energy-intensive AI technologies has led to a doubling of electricity usage in U.S. data centers. Government projections estimate that by 2028, half of all electricity consumed by data centers will be allocated to AI tools.

This report emerges at a moment when generative AI is being woven into almost every facet of digital existence. At its annual I/O event, Google unveiled extensive AI upgrades across its platforms, including Search, Gmail, Docs, and Meet. In the meantime, individuals are increasingly relying on AI for a myriad of tasks, from conducting job interviews to fabricating deepfakes and even cheating on academic assignments.

All this innovation, the report cautions, carries a significant environmental cost.

Disclosure: Ziff Davis, the parent company of Mashable, initiated a lawsuit against OpenAI in April, claiming copyright infringement in the training and functioning of its AI systems.