GPT-5, OpenAI's latest model, exhibits extraordinary power but comes at a significant energy cost - reportedly demanding daily electricity consumption equivalent to that of 1.5 million American households.
In a recent development, OpenAI has launched the highly anticipated GPT-5 AI model, boasting sophisticated capabilities in coding, writing, and healthcare. However, the energy demands of this advanced model are raising concerns.
According to research by the University of Rhode Island and independent analyses, GPT-5 consumes an average of 18.35 watt-hours (Wh) of electricity per query, with longer and more computer-intensive responses potentially using up to 40 Wh. This is a substantial increase compared to its predecessor, GPT-4, which used about 2.12 Wh per query.
This rise in energy consumption means that if GPT-5 handles OpenAI’s reported 2.5 billion daily requests, the total energy usage would reach approximately 45 gigawatt-hours (GWh) per day. To put this into perspective, this energy consumption is comparable to the output of two to three nuclear power reactors.
It's important to note that OpenAI has not officially disclosed energy consumption metrics for GPT-5. These figures come from research based on hardware power profiles and data center efficiency assumptions.
Compared to other models, GPT-5's energy use is notably high. For example, Google's Gemini reportedly consumes only about 0.24 Wh per median text query.
The increased power consumption supports faster, more thorough processing of complex queries and dynamic resource allocation depending on query difficulty, unlike previous fixed-power operation models. OpenAI CEO Sam Altman has compared the energy consumption of GPT-5 to an oven using energy in a little over one second or a high-efficiency lightbulb using energy in a couple of minutes.
However, users have expressed dissatisfaction with the recent updates, claiming that the new model has ruined and degraded ChatGPT's user experience due to glitches, bugs, and unresponsiveness.
OpenAI has not yet addressed these concerns, but has announced new updates to improve the model's user experience, including increasing the rate limit to 3,000 messages/week with GPT-5 Thinking.
In conclusion, while GPT-5 offers considerably more powerful AI capabilities, its energy consumption is a significant concern. With its daily energy demands equivalent to the electricity demand of 1.5 million homes in the United States, it's clear that the energy cost of advancements in AI is something that needs to be carefully managed.
Microsoft could consider investing in renewable energy solutions to power its Windows and Xbox hardware, reducing the environmental impact of running future software updates that might utilize AI models like GPT-5.
The surface of the new hardware might require redesigning to better manage heat dissipation, given the increased power consumption of advanced AI models such as GPT-5.
In an attempt to minimize energy consumption, developers could potentially create software optimizations for AI models like GPT-5, ensuring they operate more efficiently on Microsoft's hardware.
The high energy consumption of GPT-5 raises questions about the long-term viability and sustainability of AI advancements in the technology sector, necessitating further research and innovation in energy-efficient hardware and software technologies.
