Rapid AI-driven data centre growth is forcing Australia’s water sector and digital infrastructure industry to confront challenges neither side believes it is fully prepared for, according to speakers at the Melbourne Cloud & Datacenter Convention. While panellists agreed that recycled water networks, early engagement with utilities, and flexible cooling strategies will be critical to supporting future capacity, they also acknowledged significant uncertainty around how AI workloads will reshape demand for power, cooling, and water.
The discussion, titled AI, Data Centres & The Water Crisis: Can Australia power the AI revolution sustainably without compromising its most precious resource?, brought together Danfoss data centre and commercial buildings lead Jim Rose, NEXTDC head of sustainability Nishi Vissamraju, VicWater CEO Jo Lim, and TBH director and data centre lead Rob Hammond.
AI changes the cooling equation
Rose argued that the industry’s challenge should be viewed primarily as a heat management issue rather than a water issue. “At its core, a data centre is a heat problem – almost all power coming in becomes heat that we need to manage,” he said.
According to Rose, AI is changing not only the scale of demand but also the density of computing infrastructure, creating more heat within a smaller footprint and exposing gaps between traditional facility designs and operational reality. “AI is driving a step change in intensity, not just overall demand – more heat, less footprint – and that’s where the gap starts to show between real operation and design,” he said.
He warned that operators can no longer consider power, cooling, and water as separate infrastructure challenges. “Power, water, cooling must be considered all together, not separately.” Rose also rejected the idea that a single cooling technology would emerge as the industry’s answer to AI. “There’s no perfect technology; it’s about thermodynamics. You still have to reject the heat out of the data centre. It’s a system decision, not a technology issue.”
Vissamraju described how NEXTDC approaches cooling as inherently site-specific. “There is no right cooling solution for a data centre.” In Victoria, the climate makes a compelling case for water-based approaches at facility level combined with liquid-to-chip cooling – “you can get PUEs as low as 1.1 to 1.5 in our facility, which is fantastic from a peak energy grid pressure perspective.” In Queensland, she said, humidity makes that approach unworkable. “It’s really about designing to the facility and local climate conditions.”
She added that water-based infrastructure carries advantages beyond efficiency that are often overlooked. Compared to air-based cooling, it is significantly quieter – relevant for metro facilities neighbouring residential areas – and has an embodied carbon advantage, with infrastructure lifecycles of 20 to 30 years compared to five to eight for air-based equivalents. “If you look at scope three emissions, it’s really important to have a zoomed-out systems view.”
The hybrid approach is gaining ground for exactly these reasons, Rose said. “It adapts to the load and requirement on site” – though he was careful to add that getting the balance right in practice, particularly with the variable load profiles of AI workloads, requires precise engineering. AI workloads “ramp and shift, they don’t follow smooth patterns,” creating variation across systems “with a thinner margin and very less room for error.”
The infrastructure readiness gap
Hammond, whose firm conducts due diligence for data centre investors and developers, described what he called the four Ps constraining the sector: people, procurement, power, and permitting. On power, he echoed the earlier panel’s conclusions – generation is not the constraint, transmission is. “We’ve got enormous amounts of generation that we’re building – solar farms and wind farms in regional parts of Victoria and up the East Coast – which we can’t actually connect up.”
On the broader readiness question, he was candid: “We’re not ready for what’s coming in terms of the demand.” He argued that infrastructure should be leading development rather than following it, and that government has a role in directing where data centres are built. “The infrastructure is being built, and there’s an encouragement for data centres to be built in the right locations.”
On permitting, he noted a meaningful difference between states. Victoria is processing approvals in around two months compared to a year in New South Wales. “The Victorian government has said we’re serious about this, and driven in that direction – that’s a really positive sign.”
Vissamraju addressed the frequently raised question of why data centres can’t simply relocate to where resources are more available. Latency, redundancy, and uptime requirements make proximity to population centres a functional necessity for many workloads. “Every one of us in this room are using AI on our phones – we all want fast streaming, fast reduced latency.” She referenced a Mandala report, commissioned for Data Centres Australia by AirTrunk, AWS, CDC Data Centres and NEXTDC, which found that data centres currently support close to 40% of their energy consumption through renewable energy generation, largely via PPAs, and have invested $3.1 billion in grid infrastructure since 2020.
Phantom demand and the forecasting challenge
Lim described the difficulty water corporations face in distinguishing genuine demand from speculative applications. “We don’t necessarily know which applications are genuine or which are speculative – it doesn’t cost a lot to put an application in.” Water corporations conduct urban water strategies on five-year cycles; a connection application that falls within an existing demand forecast is straightforward to assess, but one that exceeds it triggers a full infrastructure assessment. “Can it happen? How will it be paid for? That is a challenge for us.”
She urged the data centre sector toward much earlier engagement with water corporations, both for site selection guidance and to identify where recycled water infrastructure already exists. “Recycled water infrastructure is located in certain areas, it’s not everywhere, so that early engagement is really key. And transparency – we would love greater transparency from the data centre sector to really understand what the activity is.”
Hammond noted the same dynamic from the investor side. “Phantom demand” inflates the apparent scale of infrastructure need, making it harder to plan and potentially leading to enabling infrastructure – water treatment plants, substations – being built in the wrong locations.
Recycled water as the primary pathway
Despite the complexity, there was strong consensus that recycled water represents the most viable long-term supply pathway, and that data centres are well positioned to act as the off-take customers that make the necessary infrastructure financially viable.
Lim noted that Victoria’s water security planning – including expanded desalination capacity, purified recycled water for drinking, and greater use of recycled water generally – is already underway independently of the data centre sector, but that the sector’s involvement could accelerate it. “We absolutely see an opportunity for the data centre sector to be part of that work.”
Vissamraju described active engagement with relevant water utility providers on this model – exploring whether recycled water could play a greater role in supporting NEXTDC facilities, and whether broader precinct-level demand could help underpin the investment case for recycled water infrastructure.
She was direct about the timing mismatch that makes this difficult. “Our builds are 18 months. We can’t wait for budgets to be released and then talk about it and then four or five years of actual building; that’s eight years already gone.”
The social licence dimension
All four panellists touched on social licence as an increasingly material consideration. Lim noted that water corporations carry their own public trust obligations. “It’s in our interest to be working collaboratively with the data centre sector to come up with the best solution for a particular area or a particular development.”
Hammond suggested that community pushback on irresponsible construction is “a healthy thing to challenge the pure economics of let’s do this as cheap as we can.” He closed with a longer view: “In the 1800s we built railways, in the 1900s we built highways – we’re now building digital infrastructure that’s going to service the next 100 years plus. There is amazing opportunity to take advantage of that with the resources we have to build renewable energy and water infrastructure, and provide benefits not only to the data centres but also to the community.”
Vissamraju echoed the need to bring communities along. “Taking them along the journey, not just being a dystopian building that they have no idea about – explaining what that’s actually contributing to the digital economy.”
Rose’s closing message returned to fundamentals. “We need to design for how the system operates in real life, not ideal conditions. In Australia, that means sustained heat and variability. When you get it right, you enable capacity and AI. When you get it wrong, it becomes the constraint.”
Lim’s final word was on cost. Early engagement, transparency, use of non-potable water, and water-efficient technology all featured in her list of practical requirements – alongside “a fair and equitable contribution towards the costs of the infrastructure that’s going to be needed.” On the data centre side, she added, cost may not be the primary constraint. “It certainly is on the water side.”
The panel was moderated by Tom McMahon, partner at Mandala.