Google’s greenhouse gas emissions in 2023 were 48% higher than in 2019, according to its latest environmental report. The tech giant attributes this surge to the escalating energy needs of its data centers, driven by the explosive growth of artificial intelligence (AI).

AI-powered services require significantly more computer power—and consequently, electricity—than standard online activities. This has prompted numerous warnings about the environmental impact of AI technology. Google aims to reach net zero emissions by 2030, but it acknowledges that “as we further integrate AI into our products, reducing emissions may be challenging.”

In its 2024 Environmental Report, Google states that the increased emissions are “due to increasing energy demands from the greater intensity of AI compute.” Data centers, essentially massive collections of computer servers, are crucial for AI operations, which demand a substantial amount of computational power. A generative AI system, such as ChatGPT, might use approximately 33 times more energy than machines running task-specific software, according to a recent study.

However, Google’s report also highlights significant global disparities in the impacts of its data centers. Most of the centers in Europe and the Americas derive the majority of their energy from carbon-free sources. In contrast, data centers in the Middle East, Asia, and Australia rely far less on carbon-free energy. Overall, Google reports that about two-thirds of its energy comes from carbon-free sources.

“If you actually go into a data center, it’s really hot and really noisy,” says Tom Jackson, professor of information and knowledge management at Loughborough University. “People don’t realize everything they’re storing in the cloud is impacting their digital carbon footprint.” Prof. Jackson leads the Digital Decarbonisation Design Group, which aims to measure and find solutions to reduce the carbon footprint of data usage.

“Data providers must work closely with large organizations to help them move away from storing so much of their dark data,” he says. Dark data refers to data collected by organizations but used either once or not at all. Even though it is not being actively used, storing it on chips consumes significant amounts of energy.

“On average, 65% of the data an organization stores is dark data,” Prof. Jackson notes. He commends Google’s target of achieving net zero in its data centers by 2030 but acknowledges that it will be “really tough.”

The increasing energy—and water—use of AI has led to a series of warnings, especially as the sector is expected to continue growing rapidly. The head of the UK’s National Grid stated in March that the combination of AI and quantum computing would lead to a six-fold surge in demand over the next ten years.

For more details, you can read the full report here.