An Urgent Need to Confront AI Data Centre Energy Use

Data centre electricity consumption could triple within the next decade due to AI adoption, according to research from Deloitte, the professional services firm.
The firm's Powering AI report reveals data centres currently consume 380TWh of electricity globally, representing 1.4% of global electricity use. This figure could reach 1,000 TWh by 2030 and 2,000 TWh by 2050.

Such a stark rise in energy consumption stems from increased use of specialised processors such as Graphics Processing Units (GPUs), the chips optimised for parallel processing and Tensor Processing Units (TPUs), custom-designed AI accelerator chips. These processors inevitably require more power than traditional computing hardware to train and run AI models.

As a result of their increased use within data centres, environmental concerns are mounting. Across the sector, carbon footprint levels are varying significantly depending on the energy sources powering the data centres - and particularly when AI is involved.

Continued rising power consumption

Deloitte’s report outlines an emerging concept called 'green AI', which encompasses technologies and practices designed to optimise resource usage while maintaining computational performance. This includes implementing efficient cooling systems, heat recovery mechanisms, and sustainable computing practices.

"The potential of AI to reduce waste and optimise supply chains is great but so is its energy consumption,” comments Geoff Tuff, Principal of Deloitte Consulting LLP and Sustainability and Climate Leader for Energy, Resources & Industrials. “I haven't yet seen, at scale, the rational conversation about balancing AI model sophistication based on dominant use with energy intensity and trade-offs."

Data centre operators are also exploring methods to reduce energy consumption through infrastructure design improvements and clean energy adoption. These efforts involve upgrading cooling systems, implementing heat recovery solutions and transitioning to renewable power sources.

The challenge of measuring and reducing energy consumption is complicated by the rapid evolution of AI technology. New models and applications emerge frequently, each with different computational requirements and energy profiles.

Pushing to develop global sustainability standards

Industry collaboration within the data centre sector will be essential to establish standardised reporting on energy usage and develop more energy-efficient hardware. Deloitte’s report suggests that, without regulatory frameworks and industry-wide cooperation, AI development could impede progress toward global sustainability targets.

Whilst global average Power Usage Effectiveness (PUE) has improved from 2.5 in 2007 to about 1.6 today, further improvements are needed to offset increasing demand. In regions still reliant on fossil fuels, the computational demands of AI will continue to pose particular environmental challenges. 

With this in mind, the data centre sector is starting to roll out a broader range of sustainability-led solutions in order to mitigate environmental impact. For instance, advanced cooling technologies, including liquid cooling and free air cooling, could offer potential solutions for reducing energy burden.

Likewise, Deloitte emphasises the importance of location selection for new data centres in its report. Specifically, regions with access to renewable energy sources and natural cooling opportunities can stand to significantly reduce both operational costs and environmental impact.

"We need immediate action from stakeholders across the AI and data centre industries to address these challenges," says Tuff. "The conversation about energy consumption in AI needs to happen now, not after the infrastructure is built."

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