How Google’s AI Chips Are Tripling Their Carbon Efficiency
Google recently published a comprehensive study on the lifecycle emissions of its Tensor Processing Unit (TPU) chips which revealed strong progress in sustainability.
The company’s research reveals that its latest generation of TPUs, branded as Trillium, achieves three times the carbon efficiency of earlier versions.
TPUs are specialised accelerators that are designed to optimise AI workloads. As part of its study, Google examined five models of its TPU hardware to estimate their full lifecycle emissions to understand how hardware design can impact carbon efficiency.
Kate Brandt, Google’s Chief Sustainability Officer, explains: “The study found that innovation in our chip hardware design led to a 3x improvement in the carbon-efficiency of AI workloads over two generations and that decarbonising our electricity-related emissions will drive the biggest carbon reductions for our AI footprint.”
Measuring compute carbon intensity
Central to Google’s study is a new metric called Compute Carbon Intensity (CCI). This measures an AI accelerator chip’s carbon emissions per unit of computation, expressed in grams of CO₂ per Exa-FLOP (a measure of computing performance).
A lower CCI score indicates a more carbon-efficient hardware platform for AI workloads. This strategy can be used to help track enterprise progress in increasing the carbon efficiency of TPUs.
On this, Robert Little, Sustainability Strategy Lead for gTech at Google, shares: “[TPU] efficiency impacts AI's environmental sustainability. This progress is due to more efficient hardware design, which means fewer carbon emissions for the same AI workload.”
The study also analysed five generations of TPU hardware to assess their lifecycle emissions. It found that operational electricity emissions account for more than 70% of a TPU’s lifetime carbon footprint.
This highlights the need for energy-efficient chips and low-carbon electricity sources to power them.
Manufacturing and operational emissions
The findings also showcase Google’s dual focus on leveraging AI for innovation while addressing its environmental implications.
Kate notes, “We also know it’s equally important to manage its environmental impacts, and we’re working to do that through efficient infrastructure, model optimisation and emissions reductions.
“This study is an important step in those efforts, unlocking critical insights for Google and others looking to reduce emissions across the full lifetime of AI hardware.”
Adam Elman, Google’s Director of Sustainability for Europe, the Middle East and Africa (EMEA) adds: “This is just the beginning with huge opportunities to continue optimising hardware and software for carbon efficiency.”
While operational emissions dominate a TPU’s lifecycle footprint, manufacturing emissions also play a role. As operational emissions decrease through improved energy efficiency and cleaner electricity sources, manufacturing will constitute a larger share of total emissions.
In response, Google’s lifecycle assessment seeks to identify high-impact areas for decarbonisation across its supply chain. The tech giant is now working with suppliers to reduce these emissions, particularly when it comes to chip manufacturing.
Robert Little reflects on these findings: “These findings highlight the importance of optimising both hardware and software for a sustainable AI future,” Robert says.
“It's important to remember where AI has important implications for reducing emissions and fostering sustainability.”
The implications for data centre operators
Google’s advancements come as the data centre industry faces increasing pressure to meet growing demand from AI workloads while addressing sustainability challenges.
Its study is timely, given the current push for operators to adopt renewable energy sources and invest in more energy efficient infrastructure to support power-intensive technologies like AI.
For Google, its AI-led data centres have been creating challenges for some time, with the company’s investment in the technology leading to a sharp spike in its emissions. To confront this, the company has highlighted its eagerness to invest in alternative energy sources and nuclear power to fuel its data centres.
The company even partnered with Kairos Power in October 2024 to accelerate the clean energy transition with small modular reactors (SMRs).
Kate Brandt concludes by emphasising the broader significance of this work: “This study is an important step in those efforts, unlocking critical insights for Google and others looking to reduce emissions across the full lifetime of AI hardware.”
Explore the latest edition of Data Centre Magazine and be part of the conversation at our global conference series, Tech & AI LIVE and Data Centre LIVE.
Discover all our upcoming events and secure your tickets today.
Data Centre Magazine is a BizClik brand
