Data centres could use up to 9% of the total electricity generated in the US by 2030, according to new research from the Electric Power Research institute on Wednesday (29 May).

The statistic comes as technology companies expand their data centres to fuel the heavy demand for AI.

The estimated annual growth rate of electricity usage from AI ranges from 3.7% to 15% through 2030, depending on the efficiency of newly built centres, the institute said. 

The advanced computing and cooling systems within a data centre require massive amounts  of power. A new data centre can use as much electricity as 750,000 homes, according to energy company earnings calls. 

Wilson Chan, CEO of data company Permutable AI, previously told Verdict the large amount of energy consumed by AI may mean its not sustainable in the long run.

“This could lead to an excessive strain on existing energy resources and potentially contribute to environmental issues such as increased carbon emissions,” Chan said.

OpenAI CEO Sam Altman said in January that an energy breakthrough would be necessary for future development of AI, which he said will require much more power than the industry initially expected. 

Altman said the only way forward for AI would be through the use of climate-friendly nuclear fusion or solar power. 

“There’s no way to get there without a breakthrough,” Altman said. “It motivates us to go invest more in fusion,” he added. 

Isabel Al-Dhahir, principal analyst at GlobalData, told Verdict that AI is already significantly contributing to carbon emissions.

“Sam Altman is not alone in his concerns that the growing diffusion of AI tools will exacerbate the climate crisis,” Al-Dhahir added.

Al-Dahir noted that researchers at McGill University, Mila Quebec AI Institute and Carnegie Mellon are all exploring ways to quantify and mitigate AI’s impact on the climate.