How Is Google Managing the Environmental Impact of AI?

The expansion of artificial intelligence is presenting a challenge for technology companies, as the models promising to address climate issues are also driving a major increase in electricity demand.

Around the world, data centres are consuming more power to support sophisticated algorithms that raise questions about the industry's ability to meet its environmental targets while advancing AI.

In its 10th annual environmental report, Google has detailed a 12% reduction in emissions from its data centres in 2024, achieved alongside a 27% jump in electricity consumption.

Google states this was accomplished through a mix of clean energy procurement, hardware innovation and improvements in infrastructure efficiency.

“We’re proud to release our 10th annual Environmental Report, which details how we’re working to address the increased energy demands of AI to enable this positive impact, while also showcasing how AI can be used to build a more energy-efficient and resilient world,” Google says.

Clean energy purchasing and carbon-free operations

In 2024, Google signed contracts for over 8 gigawatts of new clean energy generation, its largest annual total and double the volume from the previous year.

This brings Google's total procured clean energy since 2010 to over 22 gigawatts across more than 170 agreements.

This capacity is roughly equivalent to Portugal’s total renewable energy capacity in 2024. During the year, 2.5 gigawatts of new clean energy came online from projects contracted in previous years.

For the eighth consecutive year, Google maintained a 100% renewable energy match on a global basis.

Its carbon-free energy percentage for data centres and offices also increased from 64% to 66% on an hourly basis.

This means facilities matched their electricity consumption with clean power from the same grid during the same hour.

According to Google, its clean energy purchases avoided over 8.2 million metric tons of carbon dioxide equivalent emissions in 2024. Cumulatively from 2011 to 2024, this avoidance is estimated at over 44 million metric tons.

Luke Elder, Lead Sustainability Reporter at Google, explains Google's approach to reporting.

“Transparency, accuracy and rigour are the foundation of sustainability reporting,” he says.

“As the volume and complexity of data and strategies grow, we’re innovating our processes to meet rising expectations without compromising on these core commitments," he adds.

"This year, we integrated generative AI into our workflow to support the development of Google’s 2025 Environmental Report. We’ve learned a tremendous amount from this first implementation and are already building more ways to leverage AI for future reports.”

Nuclear and geothermal energy advancements

Google is also looking beyond traditional renewable sources, signing the first corporate agreement to purchase nuclear energy from small modular reactors (SMRs) through a partnership with Kairos Power. The deal could bring up to 500 megawatts of clean energy to US grids by 2035.

SMRs differ from conventional nuclear plants as they are smaller, factory-built units assembled onsite.

Kairos Power uses a molten salt cooling system and ceramic pebble fuel, which allows the reactor to operate at low pressure.

Google has also extended its partnership with Fervo Energy for enhanced geothermal projects in Nevada.

This involves a contract for a 115-megawatt project, increasing the geothermal generation enabled by Google by nearly 25 times compared to its first pilot, which became operational in 2023.

Data centre and hardware efficiency

Technological advancements are a key part of Google's strategy. Its data centres now deliver over six times more computing power per unit of electricity than five years ago.

Google's average annual power usage effectiveness (PUE) for its global fleet fell to 1.09 in 2024.

A PUE figure closer to 1.0 indicates that almost all energy is used for computing equipment rather than overheads like cooling.

Custom-built processors designed for AI and machine learning applications have contributed to these gains.

The seventh-generation Tensor Processing Unit (TPU), named Ironwood, operates nearly 30 times more efficiently than the first Cloud TPU from 2018. The sixth-generation TPU, Trillium, delivers 67% more energy efficiency than its predecessor.

According to the report, five of Google’s products enabled customers to collectively reduce an estimated 26 million metric tons of greenhouse gas emissions in 2024.

For example, Nest thermostats helped save over 25 billion kilowatt-hours of energy, while fuel-efficient routing in Google Maps enabled over 2.7 million metric tons of emissions reductions.

“AI isn’t just a tool – it’s a catalyst,” Google says. “It’s helping people make smarter decisions faster, and its potential to help manage emissions in key sectors – like transportation and energy – is transformational.”