AI, Plumbing and the Illawarra: Smarter Systems, Thirsty Data Centres and a Sustainability Reality Check

If you spend your days around pumps, pipes and plant rooms in the Illawarra, you’ve probably noticed that artificial intelligence is no longer an abstract tech headline. It’s creeping into the toolbox, the service van, the SCADA screen and the tender pack. At the same time, AI’s appetite for compute is fuelling rapid growth in data centres across NSW, and that brings a big, practical question for our region: where does the cooling water come from, and can AI’s boom be sustainable for Wollongong, Shellharbour, Kiama and surrounds?

What AI looks like on the tools, today

On a typical industrial or commercial job, AI already shows up in small, helpful ways. Acoustic leak detection models sift through background noise from plant to identify sub-surface leaks that human ears miss. Pressure and flow telemetry from booster sets, separators and ring mains feed anomaly-detection algorithms that flag bearing wear or cavitation days before a catastrophic pump failure. Thermal cameras, paired with vision models, spot hot joints and wet insulation on large roofs faster than a manual walkdown.

Even office-side, AI is shaving hours off admin. Large language models can compare a spec to the NCC and AS/NZS 3500 series and highlight conflicts to resolve before construction. They turn field notes and photos into defect lists, draft method statements and SWMS, and produce take-offs from marked-up PDFs with less rework. Route-optimising dispatch tools cut kilometres between Port Kembla, Unanderra and Albion Park, reducing idle time and fuel burn. The result isn’t a robot plumber; it’s a crew that gets to the real problem sooner, with better information.

In plant operations, machine learning blends with familiar SCADA. For example, a chilled-water loop feeding an MRI suite or process line can be run with predictive setpoints that anticipate heat load, flattening peaks and reducing short cycling. In high-rise commercial buildings in Wollongong’s CBD, AI-based pressure control keeps fixtures within safe limits while avoiding the “yo-yo” effect that fatigues pipework. For strata, smart water meters use simple models to separate legitimate overnight use from silent leaks in risers or irrigation.

The data centre wave, and why water matters

Generative AI’s rise is driving demand for data centres, and most will cluster where there’s power, fiber, and planning certainty—Western Sydney will carry the bulk, but edge facilities and specialist hubs are likely to spill into regional nodes like the Illawarra over time, especially with local renewables and port logistics in play. For plumbing and hydraulics, the headline is cooling. Large data centres typically rely on chilled-water plants with cooling towers, or increasingly, high-efficiency liquid cooling. Towers evaporate water to reject heat; that’s where consumption skyrockets.

How much water are we talking about? Industry measures water use effectiveness, WUE, in litres per kilowatt-hour, written as $L / kWh$ . Typical evaporative-cooled sites report WUE in the range $0.2$ – $1.0 L / kWh$ , depending on climate and system design. Do the math on a mid-size $20 M W$ facility running at a $50%$ average load: that’s $10 M W \times 8, 760 h = 87, 600, 000 kWh$ per year. At a conservative $0.4 L / kWh$ , annual water use is roughly $35 mi ll i o n$ litres. At $1.0 L / kWh$ , it’s about $87.6 mi ll i o n$ litres. Those are city-scale numbers.

Big tech disclosures back the scale. Microsoft reported using about $1.7$ billion gallons of water in $2022$ (roughly $6.4$ billion litres), with year-on-year increases attributed in part to AI expansion, while Google disclosed around $5.6$ billion gallons in the same year. Academic work in 2023 estimated training a frontier model could consume hundreds of thousands of litres of freshwater for cooling, and—more controversially—that a typical ChatGPT exchange might indirectly “consume” about $0.5$ litres of water per $20$ – $50$ prompts, depending on datacentre location and cooling approach. The exact figures vary by site and season, but the direction is clear: AI workloads move a lot of heat, and water is often the cheapest heat sink.

Is AI sustainable for the Illawarra?

For our region, the answer depends on how, and where, facilities are built and operated. The Illawarra’s water is largely sourced from the Upper Nepean system and managed by WaterNSW and Sydney Water. That system has proved resilient, but every large, continuous, non-potable load must be weighed against drought risk and growth in residential, commercial and industrial demand.

There is a strong, positive case. Data centres can be designed to minimise potable draw by using alternative cooling strategies. Liquid-immersion or direct-to-chip cooling raises chilled-water temperatures and can enable dry coolers for a significant portion of the year, slashing evaporation. Hybrid towers with adiabatic assist use far less water on mild days. Seawater or harbour-water cooling is technically feasible near Port Kembla, with proper heat-exchanger design and discharge controls to protect marine ecosystems. Recycled water is a pragmatic option if supply and quality can be guaranteed; side-stream filtration, biocide management and robust blowdown handling become part of the plumbing scope, not an afterthought.

On the other hand, if new facilities rely on traditional towers fed by potable mains with minimal reuse, the cumulative impact during a dry spell could be felt across the network. Transporting large volumes of non-potable water by truck is neither sustainable nor neighbourly. And grid stress matters, too: if training runs concentrate on hot afternoons, they elevate both electricity and water demand when the system is under pressure.

Day-to-day, what would responsible AI infrastructure look like in the Illawarra?

It would start with water hierarchy baked into planning approvals: rainwater capture sized to meaningful fractions of cooling demand; recycled water as the primary source where practicable; mains potable as a last-resort backup. It would include $N + 1$ metering—separate meters for tower makeup, blowdown, humidification, domestic uses and landscape—linked to WUE dashboards. It would specify high-solids tolerance for side-stream filters, automated bleed setpoints linked to real-time conductivity, and corrosion control that minimises hazardous trade waste. It would require backflow containment at the property boundary sized for worst-case flow, and cross-connection control programs audited annually.

At a precinct level, there’s a bigger opportunity. Pair data centres with industries that have complementary thermal profiles. Waste heat from servers can regenerate desiccant wheels for dehumidification, feed low-temperature process heating, or support district hot-water loops to nearby commercial buildings. On very hot days, thermal storage (ice or phase-change) can shift peak cooling load into the night when the grid is greener and air temperatures allow more dry cooling. Those are plumbing and plant integration problems as much as they are electrical ones.

Where AI helps plumbing’s own sustainability

The irony is that the same AI driving data-centre demand is making water infrastructure more efficient. In the Illawarra, AI will keep helping us find leaks faster in buried mains and fire loops, cut non-revenue water for campuses and business parks, and run pumps at their sweet spots. Models that predict water hammer risk from valve operations can prevent pressure spikes and extend pipe life. Smarter irrigation controllers, trained on local weather and soil data from the escarpment to the coast, reduce outdoor demand without browning out landscapes. Inside buildings, AI-based fixture analytics can distinguish a faulty flush valve from a legitimate after-hours cleaning cycle, and prompt a targeted work order instead of a generic “high usage” alarm.

Known 2025 consumption, with caveats

By mid-2025, public numbers are still a patchwork. Vendors publish annual water totals, researchers publish model-based estimates, and operators announce new cooling designs, but there isn’t a single, real-time meter for “AI’s water”. What we can say with confidence is that:

Water use scales with energy use, and AI is energy intensive. A single large model training run can draw megawatt-months of power; inference at population scale adds steady baseload.
WUE is the key lever. Move from $1.0 L / kWh$ to $0.2 L / kWh$ , and you cut water by $80%$ . Move to dry or hybrid systems for most of the year, and evaporation drops close to zero outside heatwaves.
Location matters. Cooler, drier air supports more hours of dry cooling. Proximity to recycled water infrastructure or seawater intake/outfalls opens up non-potable options. The Illawarra’s coastal climate is favourable for hybrid strategies if designed well.

Practical next steps for our region

For councils and planners: require WUE targets, recycled water feasibility assessments and heat-reuse studies at DA stage for large digital infrastructure. For water authorities: co-design precinct-scale recycled water schemes that are economically viable and hydraulically robust. For developers: choose cooling technologies that de-risk both power and water, and engage hydraulics early so the plant room, roof space and pipe routes are reserved, not improvised.

For industrial, commercial and strata clients who won’t build data centres but will feel the ripple effects: lean into AI where it reduces your own water and energy use. Install submetering and telemetry on process lines and potable services, add leak analytics and pressure optimisation, and use AI-assisted maintenance to stretch asset life. Every kilolitre you save locally makes the broader system more resilient.

Our view from the plant room

We’re bullish on AI as a force multiplier in day-to-day plumbing, and cautiously optimistic about its infrastructure footprint—if the Illawarra sets the bar high now. The smartest datacentre in 2025 isn’t just the one with the latest chips; it’s the one with a plumbing design that respects our dams, uses recycled or seawater where it can, recovers heat cleverly, and proves its performance with transparent metering. That’s the kind of engineering challenge our region is built for.

If you’re planning a facility upgrade, a new industrial line, or you just want your building to run leaner, we can help you put AI to work where it counts: diagnosing problems earlier, reducing waste, and making compliance smoother. And if you’re exploring digital infrastructure, talk to us before you lock in a cooling choice. The right hydraulics strategy can turn AI from a thirsty neighbour into a good one.