The Big Power Use of GPT-5 A Coming Energy Problem
OpenAI's next GPT-5 model may bring big leaps in AI skill, but it could also bring a huge hidden cost its need for power. A new look suggests GPT-5 might need a lot more power than the one before it, showing a key test for the future of AI.
The Huge Amount of GPT-5's Expected Power Use
While OpenAI keeps its model power uses secret, work from the University of Rhode Island's AI lab gives a stark guess. The data hint:
- A Big Jump in Need GPT-5 might use up to 8.6 times more power than GPT-4.
- Per-Ask Power Use One normal GPT-5 ask might need over 18 watt-hours (Wh) of power, more than 2.12 Wh for GPT-4.
- Daily Power Use If all of ChatGPT's said 2.5 billion asks per day used GPT-5, the full day power use could hit 45 gigawatt-hours (GWh).
To show what 45 GWh means, it's about the daily power from two or three new nuclear power places enough power for a small country. This is a big possible pull on world power goods.
A Big Note These Are Guesses Built on Guesses
We must know that these numbers are not sure facts. The researchers made key guesses as OpenAI has not shared its setup. The lab took these steps:
- Guessing the Tech The group thought that GPT-5 uses Nvidia DGX H100 or H200 systems on Microsoft Azure. If OpenAI has newer, better tech like Nvidia's Blackwell, these guesses would be wrong.
- Using Reply Time They used the time it takes the model to reply and the guessed power use of the tech to find out the watt-hour numbers.
This missing hard data means the solid numbers might be off. But, the way they show a big rise in power need for better AI models is a point to watch.
The Wide View Skill vs. Lasting Use
With this look, a key push-and-pull in the AI field shows itself the race for more skill against the need to keep power use low. Models like GPT-5, with deep settings like "reasoning mode" that might raise power use by 5 to 10 times for a single answer, stretch what AI can do. But they also test what our power grids can take.
Rising power bills for AI data spots all over the U.S. are now a fact. The way laid out by the GPT-5 guesses tells us this power need will only get worse, making us face a hard talk Is the power cost of new AI in the long run too much?
The Unspoken Issue A Need for Openness
The most important lesson is that bigger openness from big AI labs like OpenAI is needed. Without real data on power use, water use, and tech set-up, researchers, leaders, and the public must make guesses. As AI models become key parts of the world's setup, knowing their real world and power effects is a must it's needed for smart steps forward.
While the exact power use of GPT-5 is still not shared, the way it shows is clear. The AI world is on a path that will need a lot of power, and a hard look at better use and lasting ways is coming.
