The question of how much energy artificial intelligence consumes when generating a single response has long been one of the most elusive — and urgent — metrics in the debate over AI’s environmental impact. Earlier this year, journalists Casey Crownhart and a colleague at MIT Technology Review spent six months investigating the climate and energy burden of AI. Their search for this “white whale” number underscored a broader issue: critical data on AI’s electricity use remains locked away by the very companies driving its growth.
Despite the high stakes — projections suggest AI could consume as much electricity as 22% of all U.S. households within the next three years — AI firms have been reluctant to share detailed figures. Crownhart and his team repeatedly reached out to major players such as OpenAI, Google, and Microsoft, only to be met with silence. “It’s like trying to measure the fuel efficiency of a car without ever being able to drive it,” they reported, relying instead on educated guesses.
However, following their report’s publication in May, some movement emerged. In June, OpenAI CEO Sam Altman disclosed that an average ChatGPT query consumes 0.34 watt-hours of energy. Google followed in August, revealing that Gemini uses about 0.24 watt-hours per query. These figures, while in line with earlier estimates, were a start.
Yet experts caution that the data only scratches the surface. OpenAI’s figure, for example, appeared in a blog post, lacking details about which model it referenced or how the measurement was conducted. Google’s estimate reflects the median usage, not the more energy-intensive interactions that involve complex reasoning or longer outputs. These numbers are also limited to chat-based interactions, excluding the growing footprint from video, image, and other AI-driven services.
“As AI becomes more embedded in creative and technical workflows, we urgently need transparency across modalities,” said Sasha Luccioni, AI and climate lead at Hugging Face.
Experts also note that while the energy cost of a single query is relatively small — comparable to running a microwave for a few seconds — it’s the cumulative demand that could strain energy systems. Microsoft, for instance, reported a 23% increase in emissions since 2020, largely attributed to AI workloads, even as it pursues carbon-negative goals by 2030.
Ketan Joshi, an analyst covering climate and energy, argues that the data center expansion fueling AI growth demands stricter scrutiny. “The rate of data center growth is inarguably unusual,” he said. “Companies should be subject to significantly more oversight.”
Some in the tech industry justify the energy burden by pointing to AI’s potential to accelerate climate solutions — from optimizing energy efficiency in buildings to discovering new materials for batteries. But no substantial evidence yet confirms that AI’s benefits outweigh the environmental costs.
Meanwhile, uncertainty looms over AI’s broader adoption. OpenAI reports that ChatGPT fields 2.5 billion queries daily, and projections suggest this could soar further. But recent setbacks, such as the underwhelming launch of GPT-5 and findings from MIT that 95% of businesses see no returns on AI investments, have fueled skepticism about the sector’s long-term viability.
Ultimately, the biggest unknown isn’t how much energy a single query consumes, but whether AI’s adoption will scale as companies anticipate — or whether the hype-driven expansion will collapse before it reshapes global energy consumption.
As MIT Technology Review’s investigation demonstrates, unlocking these answers will require more than corporate transparency; it demands sustained inquiry, cross-sector collaboration, and regulatory oversight to ensure that AI’s growth aligns with global sustainability goals.
