Arm: How Sustainable is AI's Data Centre Power Demand?

The energy requirements of AI systems are becoming a sustainability challenge.

Dr. Vanessa Just, Founder and Chief Executive Officer of JUS.TECH GmbH, a sustainability-focused technology consultancy, says: “Today's data centres already consume lots of power: Globally 460 terawatt-hours (TWh) of electricity are needed annually.

"That's equivalent to the entire country of Germany.”

This increase in power demand, driven by the computational needs of AI model training and inference, presents systemic challenges to global electricity grids and sustainability targets, as highlighted by research from Arm.

The study involves insights from 650 business leaders across various industries.

The data centre problem

Current trends indicate that the electricity consumption of data centres in the US will increase from 2.5% of the national total in 2022 to 7% by 2030, according to analysis from Boston Consulting Group.

This amounts to roughly 390 terawatt-hours, equivalent to the electricity usage of 40 million US households.

Dr. Nicole Höher, Project Manager for Sustainability & Digitalisation at JUS.TECH GmbH, says: “Without significant infrastructure investment, the risk of grid instability and supply constraints grows.”

Grid operators in several regions face capacity constraints.

In areas like Virginia in the US and Ireland, high-voltage transmission networks are congested, delaying the establishment of new data centres.

Some utility companies have responded with connection restrictions or rationing measures.

Arm suggests that tackling these challenges requires coordination between infrastructure providers, technology companies and energy suppliers to ensure AI's energy demands do not undermine climate objectives.

Hardware innovations

Arm highlights the importance of hardware design in reducing AI's environmental footprint, stressing that hardware optimisation is essential for sustainability advancement even with improvements in software efficiency.

For example, AWS Graviton processors, which utilise Arm technology, have the potential to reduce workload carbon intensity by up to 67% compared to traditional x86 processors for cloud-based AI tasks.

In mobile and edge computing environments, Arm-based accelerators can cut energy consumption by 50 to 80% compared with general-purpose GPUs.

In data centre settings, Arm Neoverse processors offer energy efficiency enhancements that can cut server rack energy consumption by up to 40%, making large-scale AI deployments potentially more sustainable.

Maureen McDonagh, Head of Sustainability at Arm, warns: “Without proactive measures, AI-driven energy consumption could push the world further off track from climate targets, with projections indicating an increase of over 2°C in global temperatures; breaching the recommendations to limit the rise to a much safer 1.5°C.”

Arm further notes electronic waste concerns, citing research from The Register that suggests Gen AI could increase electronic waste by up to 2.5 million tonnes annually by 2030 without implementation of waste reduction strategies.

AI workforce readiness challenges

Arm says that although 75% of companies have adopted AI technology, only one-third of employees have received AI-related training in the past year, creating a skills gap that jeopardises technology investments.

According to the Arm AI Readiness Index referenced in the report, 34% of organisations report being under-resourced with AI talent.

An additional 39% lack dedicated programmes for developing AI skills among existing employees.

The employee preparation statistics reveal a disconnect as only 15% of US workers report that their organisation has communicated a clear AI strategy, while just 11% feel “very prepared” to work with AI in their role.

Mark Hinkle, Chief Executive Officer and Founder of Peripety Labs, a publisher focusing on AI in enterprise environments, says: “Corporate leaders risk underutilising their AI investments when the workforce isn't on board.

“When employees aren't empowered to use AI, those expensive platforms and algorithms can turn into costly shelfware, delivering only a fraction of the promised productivity gains.

“AI Adoption isn't just a Digital Transformation – it's a People Transformation,” he adds.

Training programmes for AI success

Arm recommends that organisations invest in comprehensive training and upskilling programmes.

The company also notes that 57% of workers desire AI training from their employers.

Business leaders in the survey mention various approaches to training.

One respondent says: “Our organisation offers a comprehensive AI training program that includes workshops, online courses and practical projects.”

Another explains: “Selected personnel from each department undergo intensive Python development education twice a week for one month, with certification exams conducted upon completion.”

Khaled Benkrid, Senior Director of Education and Research at Arm, says: “In a world where humans and machines are working together more than ever, the ability to build and use AI tools effectively is becoming a fundamental skill.”

To address these challenges, Arm has established partnerships with academic institutions, including funding for the University of Cambridge's CASCADE (Computer Architecture and Semiconductor Design) Centre. This initiative will support 15 PhD students over five years researching processor designs for AI applications.

“Companies must invest in training programmes that help employees understand AI's capabilities and applications,” says Khaled.

“Furthermore, the future workforce will need to combine human ingenuity with new and emerging AI technologies; going beyond just the technical skills.”

Explore the latest edition of Sustainability Magazine and be part of the conversation at our global conference series, Sustainability LIVE. 

Discover all our upcoming events and secure your tickets today.

Sustainability Magazine is a BizClik brand