The Hidden Cost of Smarter AI: GPT-5’s Hefty Power Bill

As OpenAI rolls out its new GPT-5 model, the focus is squarely on its groundbreaking intelligence, from crafting intricate websites to tackling advanced scientific problems. However, a less-publicized but critical side of this technological leap is its massive power consumption. Without any official disclosure from OpenAI, a group of dedicated researchers is working to shed light on the environmental footprint of this new AI, and their findings are concerning. The enhanced capabilities of GPT-5, they argue, come at a steep and non-negotiable energy cost.
To quantify this, a research group at the University of Rhode Island’s AI lab has put a number on it. Their dashboard indicates that an average medium-length response from GPT-5 consumes just over 18 watt-hours. This is a significant leap from previous models like GPT-4o and is a cause for alarm among environmental advocates. When you consider that a single query could use as much electricity as an incandescent bulb in 18 minutes, and that billions of such queries are processed daily, the cumulative effect is staggering. The total energy demand could be equivalent to powering 1.5 million US homes, highlighting the scale of the issue.
The dramatic increase in energy usage is a direct consequence of the model’s sheer size. While OpenAI hasn’t released the parameter count for GPT-5, experts widely believe it to be several times larger than its predecessors. This theory is supported by a study from French company Mistral, which established a “strong correlation” between a model’s size and its resource consumption. The larger the model, the greater its environmental impact. This principle, applied to a model believed to be orders of magnitude larger than GPT-3, suggests a future where AI’s hunger for energy could become a major global issue.
Moreover, the new features of GPT-5 add to its energy demands. While it employs a “mixture-of-experts” architecture to improve efficiency, its new reasoning mode and ability to process video and images create a greater computational burden. This means the model computes for a longer duration before providing a response, significantly increasing its power footprint. With companies like OpenAI pushing for increasingly large models in pursuit of Artificial General Intelligence (AGI), the call for transparency and official data on energy and resource usage is becoming more urgent than ever.