Australia’s rapidly expanding artificial intelligence (AI) industry is placing increasing pressure on the nation’s electricity infrastructure. As demand for data centres grows, significant investment in electricity generation and transmission networks will be needed. However, building enough infrastructure to meet every data centre’s maximum potential electricity demand could prove unnecessarily expensive.
In his Capital Brief article, Victor Feoktistov, co-founder of FlexSysAI and former energy trader at AGL, argues that Australia should reconsider how it plans for the energy requirements of AI data centres.
The central issue is the assumption that data centres require a constant electricity supply at maximum capacity. While some computing operations must run continuously, others, including AI model training and batch processing, can be delayed, slowed down or moved to different locations without disrupting essential services.
This flexibility creates an opportunity to manage electricity demand more efficiently. Rather than constructing new energy infrastructure to accommodate every data centre operating at full capacity simultaneously, electricity networks could account for the ability of certain computing activities to reduce their energy consumption during periods of high demand.
Such an approach could help Australia accommodate the growing AI industry while reducing the amount of additional infrastructure required.
The broader message is that Australia's energy planning should recognise the differences between essential, continuous computing and AI workloads that can be scheduled more flexibly. By incorporating this flexibility into electricity network planning, Australia could support data centre growth while avoiding unnecessary infrastructure investment and associated costs.
