top of page

The rise of AI-specific data centres in Australia

Australia currently ranks 15th out of 62 countries across the world for AI adoption, according to consultants FDM citing Tortoise’s Global AI Index. And yet, in Australia while there has been a strong commitment from the Federal Government – including high spending – the CSIRO report on Australia’s AI ecosystem momentum reveals the tepid commitment shown by Australian businesses towards adopting AI. FDM said that the Tortoise report shows 44% are not expanding or upgrading AI adoption. This, coupled with a lack of AI talent, means it’s early days for AI in Australia, despite the initial promise. 

Data centre operators can’t afford to wait for businesses to wake up to AI. Australia is one of the most densely served economies in terms of data centres per capita and stands to gain not only from the rise of AI but also from the opportunity to adopt AI and automation. Both will create a seismic shift in data centre investment, design and site selection. 

Last year the average kilowatt density per rack, globally, was 12kW per rack. But AI and high-performance computing have considerable need for high-density infrastructure. According to JLL, hyperscalers now have an estimated average density of 36kW per rack. IDC estimates this will grow at a 7.8% CAGR in the coming years to approach 50kW by 2027. Many AI clusters are projected to hit requirements of 80-100kW/rack.

While Australia is seeing growth in edge infrastructure, the greater capacity demands of cloud and AI require very substantial power and cooling. These workloads can only be run efficiently in purpose-built, high-density data centres, and this is where operators see the biggest impact from AI. 

This impacts power and water – the latter to cool the former. According to the Royal Society of Victoria. AI consumes 1.8-12L of water for each kWh of energy usage across Microsoft’s global data centres. In the United States, it is estimated that 1 MWh of energy consumption by a data centre requires 7,100L of water. It warns that in Australia, which has been an importer of technology rather than an innovator:  “a lack of understanding on how the growth of AI can impact energy and water consumption has meant that there has been a lack of action.” 

“Training AI models and the likes of analytics workloads created through inference require incredible amount of capacity,” said Macquarie Data Centres group executive David Hirst. “The current data centre infrastructure in Australia wouldn’t be near enough to manage this, and many existing facilities aren’t actually suitable for these kinds of workloads, hence the emergence of AI-specific data centres.”

Macquarie has responded with a purpose build cloud and AI data centre, IC3 Super West on its Macquarie Park Data Centre campus in Sydney’s North Zone, which will bring the total campus IT load up to 63MW.

“[IC3] includes optimising both air and liquid cooling, increased floor loading, flexible design and advanced monitoring systems to support the new wave of GPU & CPU technology,” he said. 

AI may solve several of the issues it creates

While considerable amounts of energy is needed to power AI, the technology also has the power to bolster efficiency. Innovations such as liquid cooling and AI-driven energy monitoring systems used in data centres are making an impact. Ironically, AI adoption has the power to solve many of the issues it’s creating. For example, if you’re running 100 kilowatts a rack and can resolve the cooling, your datacenter could be 1/10 the size. DC operators could then shift to modular builds and containerised solutions, massively reducing lead times and build costs. 

Modern data centres already provide increasingly efficient environments to run AI, with PUE numbers that come in far lower than traditional computer rooms or edge facilities. 

“Macquarie Data Centres was among the first data centres in Australia to introduce liquid cooling through our work with ResetData and our Macquarie Park Data Centre Campus is optimised for both air and liquid cooling,” Hirst told W.Media.  “We’re using automation and real-time monitoring to improve sustainability, for example through variable speed fans in our computer room air conditioning (CRAC) units that automatically adjust to the current temperature, which reduces power and cooling waste,” he added, pointing out Macquarie was using AI as an operational tool to do this.

“Liquid Cooling is going to drive a seismic shift in terms of densities, and it already is largely driven through the requirements that AI is bringing to bear,” Equinix Australia managing director Guy Danskine told W.Media. “So we do see a large increase in densities, but there’s significant upside in from a sustainability perspective in terms of needing less physical footprint, less equipment to drive a higher outcome.”

Last month, Equinix partnered up with Nvidia to offer the chip firm’s AI computing systems to corporate clients meaning they could run closed systems and control their data. Danskine said Australia was “absolutely in the “early innings” with AI adoption. “There is definitely a significant tailwind coming from AI for our industry overall,” he said. We’re seeing the centralisation of where the large language models will reside, and there’s considerations around data sovereignty and how you access those.” Corporations such as and Tape Ark are among the Australian organisations utilising Equinix’s infrastructure to provide private AI services for their customers.

[Author: Simon Dux]

5 views0 comments


bottom of page