A new study challenges previous estimates about ChatGPT’s power usage, suggesting that the AI chatbot consumes much less energy than once believed. However, as AI models become more complex and widely used, overall energy demands may still rise significantly.
A recent analysis by Epoch AI, a nonprofit AI research group, found that a typical ChatGPT query consumes around 0.3 watt-hours—far lower than the commonly cited estimate of 3 watt-hours per query, which suggested ChatGPT was ten times more power-hungry than a Google search.
Joshua You, a data analyst at Epoch AI, told TechCrunch that ChatGPT consumes less energy compared to everyday activities like using household appliances or driving a car. However, he acknowledged that these figures remain approximations, as OpenAI has not disclosed full details on ChatGPT’s power consumption.
Concerns about AI’s rising energy consumption continue to grow. Recently, more than 100 organizations signed an open letter, urging AI companies and regulators to prevent AI data centers from overburdening power grids or increasing reliance on nonrenewable energy sources.
As reported by TechCrunch, the widely circulated 3-watt-hour figure came from an older study that assumed OpenAI was using outdated chips. However, AI hardware has improved significantly, reducing power consumption per query.
Rand’s report predicts that by 2027, AI data centers could consume nearly all of California’s 2022 power capacity (68 GW). By 2030, training an advanced AI model could require as much electricity as eight nuclear reactors (8 GW).
According to Reuters, OpenAI and its partners—including Microsoft, SoftBank, Oracle, and MGX—plan to invest billions of dollars into expanding AI infrastructure to meet rising demand. However, the shift toward reasoning models, which spend more time processing information before responding, could significantly increase energy consumption compared to current AI models.
Despite concerns over AI’s energy footprint, OpenAI has begun developing more power-efficient models, such as GPT-4o-mini. Still, Joshua suggested in his interview with TechCrunch that users concerned about their energy impact should limit AI usage or choose smaller, less power-intensive models.
While improvements in AI efficiency continue, the growing adoption of AI worldwide may counteract these gains, leading to increasing energy demands in the years ahead.
Get the latest tech news, telecom insights, and product launches wherever you prefer.
Add ProPakistani to Preferred Sources and see more of our stories in Google Search and Top Stories.