Researchers continue to shine a spotlight on the energy costs of AI with a new study estimating AI-backed Google search could consume as much electricity as Ireland.
The study published in Joule explains that’s a worst-case scenario, based on generative AI being fully deployed across standard Google searches using current hardware and software.
In that scenario, the energy consumed by Google’s AI alone would total 29 terawatthours (TWh), or roughly equivalent to Ireland’s annual electricity consumption. For comparison, Australia’s annual electricity generation was 272 TWh.
Cosmos has been reporting on the energy and climate costs and opportunities associated with the widespread adoption of generative AI, as researchers warn care needs to be taken to avoid uncontrolled increases in greenhouse gas emissions.
Author Alex de Vries, the researcher behind website Digiconomist, based at the Vrije Universiteit Amsterdam in the Netherlands, considers a range of scenarios and data to estimate the electricity use of widespread AI adoption.
De Vries is best known for work estimating the electricity consumption and unintended consequences of digital trends like cryptocurrencies, and now generative AI.
“Throughout 2022 and 2023, AI has witnessed a period of rapid expansion and extensive, large-scale application,” the paper says.
As generative AI is more energy-intensive than other computing methods in both training and use phases, the technology’s expansion comes with escalating energy costs.
The paper quotes the chair of Alphabet (the company which owns Google) as stating interacting with generative AI could “likely cost 10 times more than a standard keyword search”.
The technology company has announced plans to integrate generative AI into its search technology and workspace products. While other major technology companies, including Microsoft, are embedding AI models across their software.
Google’s total electricity use – across offices and data centres – is already significant, totalling around 18.3 TWh in 2021. And generative AI is adding substantially to electricity demand. In 2021 – well before the AI boom – the technology was already accounting for 10 – 15% of the company’s electricity use, the paper says.
De Vries estimates that by 2027, worldwide AI-related electricity consumption could increase by more than 85 TWh annually based on projections for AI server production.
However, the exact future is hard to predict, the paper notes, as further improvements in the efficiency of AI models or hardware could change the outcome.
“Looking at the growing demand for AI service, it’s very likely that energy consumption related to AI will significantly increase in the coming years,” says de Vries.