Tag: National

The uninterruptible power supply spent fifty years as one of the least discussed assets on site: power electronics backed by a room of lead-acid batteries whose most visible job was covering the seconds between a grid failure and the generators taking load. Three changes have made it one of the most interesting parts of the building. Lithium-ion has become the default for new large installations, the grid now pays storage to help hold frequency and AI loads have broken the assumptions the whole architecture was built on. Assessment practice has not caught up with any of the three.

Every conventional data centre carries two layers of backup: generators for the long haul and a UPS for the gap, the seconds of ride-through between a grid disturbance and stable standby power, plus continuous conditioning of whatever the grid delivers. Generators got the public attention because they are visible and audible at the fence line. The UPS hid indoors, on valve-regulated lead-acid batteries sized for minutes, replaced every five or so years, chosen because the technology was proven, cheap and understood. That era is ending faster than most consent documents acknowledge. The UPS is becoming one of the defining elements of a facility’s backup philosophy rather than simply its bridging power.

The quiet takeover

Lithium-ion has moved rapidly from alternative chemistry toward the default choice for new large data centre UPS installations, particularly in hyperscale and colocation facilities. The drivers are prosaic: around 60 per cent less floor space for the same capacity, service life stretched from five years toward twelve, which removes one or two full battery replacements from the facility’s life. The switch is not automatic virtue. Lead-acid is one of the most recycled products on earth, with recovery rates near total; lithium-ion’s end-of-life pathways are still maturing and its supply chain carries its own footprint. A facility swapping chemistries trades a solved recycling story for a better operating one, which is exactly the kind of trade an assessment should see stated rather than assumed.

From cost centre to grid asset

The structural change is what the battery does while nothing is wrong. A lead-acid string waits; a lithium-ion UPS can work. Microsoft’s Dublin campus demonstrated the model, with UPS batteries certified to provide frequency response to the Irish grid; the direction has since hardened into policy: Ireland’s regulator decided in December 2025 that new data centre connections must bring generation or storage matching the maximum import capacity they request, alongside additional renewable supply for most of their annual demand. Google has argued the same logic for its battery-backed sites, that storage which strengthens the grid beats a generator that only ever waits. For Australia the implication is significant. Every megawatt of suitably designed storage inside a facility represents potential frequency response, demand flexibility or other grid services, subject to chemistry, reserve requirements, the connection agreement and market registration. That last qualification matters: a gigawatt of installed UPS is not a gigawatt available to the market. If large new loads are increasingly expected to demonstrate how they interact with a renewables-heavy grid, storage embedded inside the facility should not automatically be treated as a private resilience asset. Its potential contribution to system stability belongs in the assessment conversation.

AI broke the duty specification

The third force is the load itself. AI accelerator clusters can produce rapid power swings in milliseconds, at magnitudes conventional AC UPS architectures were never designed to accommodate; training clusters can move tens of megawatts between synchronised compute phases faster than any generator or conventional plant can follow. The industry’s answers are arriving in layers. Power-dense chemistries such as nickel-zinc are being packaged specifically for the swing duty, supercapacitors are appearing at rack level to absorb sub-second transients before they propagate upstream and the deeper fix is architectural: the shift toward 800 volt DC distribution, which strips out layers of AC conversion. That may not eliminate the UPS function so much as break it apart. Power conversion, ride-through and transient buffering that once lived together in a central AC machine can migrate to different layers of the facility, from rack-level supercapacitors handling millisecond excursions to DC-connected storage nearer the utility interface handling seconds to minutes. Storage remains fundamental; where it sits and how it meets the power train is what changes. We flagged that signal in our analysis of 800 VDC earlier this year; the practical point for assessors is that the AI facility arriving in an Australian application may carry a fundamentally different backup architecture from the one the template conditions assume.

The technologies compared

Technology	Energy density and duty	Emissions and end of life	Service life	Standing
VRLA lead-acid	Low density; minutes of autonomy in dedicated rooms	Mature chemistry; near-total recycling, the best end-of-life story in industry	~5 years, temperature-sensitive	The fading incumbent
Lithium-ion (LFP)	~60% less footprint for equal capacity; minutes to hours	Thermal-runaway risk requires engineered detection, separation and suppression; recycling infrastructure developing rapidly	10-12 years	Increasingly the default for new large installations
Nickel-zinc	Power-dense rather than energy-dense; built for high-rate swings	Aqueous, non-flammable electrolyte; avoids lithium-ion thermal-runaway characteristics	Long cycle life in bridging duty	Positioned for the high-rate swings of AI training loads
Flywheels	Seconds of ride-through, no chemistry at all	No chemicals, no thermal runaway; bearings and vacuum systems to maintain	20+ years	Niche but persistent, often paired with DRUPS
DRUPS (diesel rotary)	An alternative UPS architecture rather than a storage chemistry: ride-through and generation in one rotating machine, no batteries	Diesel emissions profile when running; no battery room at all	25+ years	Established in Europe and on large sites; couples the facility to diesel
Supercapacitors	Sub-second buffering; cycle life in the hundreds of thousands, so ageing is driven by time and temperature rather than use	No thermal-runaway pathway	15+ years	Complements storage rather than replacing it; appearing in rack-level designs for the fastest transients

The table’s honest reading mirrors the generator one: nothing wins on every axis. Lead-acid still owns end-of-life, lithium-ion owns the economics, nickel-zinc and supercapacitors own the new AI duty and DRUPS quietly asks whether the battery room should exist at all.

Four architectural tendencies

These are tendencies rather than rules. Topologies differ between operators, between generations of design and between sites.

Colocation tends to retain conventional centralised architectures, because tenant service level agreements specify autonomy and redundancy, which makes chemistry substitution far easier than topology change. Lead-acid rooms give way to lithium-ion on refresh cycles. Hyperscale cloud has greater freedom to distribute backup and redundancy through the IT and electrical architecture together, including rack-level battery backup in some designs, with fleet software carrying part of the redundancy. That freedom is what makes grid-interactive duty practical to contemplate. Edge can make battery-only resilience viable where workloads fail over elsewhere or shut down gracefully, although generator-backed designs remain common where local continuity matters. AI campuses are increasingly forced to separate very fast transient buffering from longer ride-through, which pushes storage to several different layers of the electrical architecture at once.

At the far end of the spectrum sits the same counterexample as the generator question. Cryptocurrency mining runs its mining load without a UPS at all. The flexibility runs both ways. It can shed in seconds when the grid is short, which is why these loads participate in the frequency control markets. It can also soak up generation nobody else wants when prices go negative, acting as a buyer of last resort for a renewable generator facing curtailment. A conventional facility buys storage so the compute is protected from the grid; mining leaves the compute unprotected so it can do both of those jobs for the grid instead.

What an assessor should ask

An assessor should be able to get clear answers to five questions.

1. Which chemistry and architecture? Stated explicitly, because fire design, ventilation and emergency response differ fundamentally between a lead-acid room, a lithium-ion installation certified under UL 9540A test regimes and a flywheel hall. Local fire services need to know which they are walking into.
2. What is the end-of-life commitment? Battery fleets at data centre scale are a recycling obligation that belongs in the application, not discovered at decommissioning.
3. Can the installation support the grid? Storage that can deliver frequency response is a public benefit an assessment should recognise and, where appropriate, encourage.
4. Does the autonomy claim match the workloads it protects? An edge site claiming battery-only backup and an AI campus claiming conventional UPS coverage are making very different engineering assertions. At least one of them deserves scrutiny.
5. What does the storage do when the grid is healthy? The traditional question was what happens when the grid fails. The more useful one is what this asset does for the other 99.99 per cent of its life.

The bottom line

The UPS is where the industry’s three transitions meet, in what used to be one room: the chemistry transition, the grid-services transition and the AI architecture transition. It is also the layer communities never see and assessments rarely question, which made sense when the batteries simply waited and makes none now. The generator question, as we argued in our companion piece, is what actually needs backup, for how long and why. The UPS question is sharper: whether the storage between the grid and the compute is only ever a private resilience asset, or a working public asset as well. Australian assessment practice should start asking these questions now, because the answer is designed into a facility years before anyone flips the first switch.

A topical commentary from the Digital Infrastructure Institute. Companion to our analysis of diesel backup and its successors. For assessment practice, see our State & Territory Frameworks.

Sources: Microsoft — Dublin datacenter batteries supporting the Irish grid; The Register — Microsoft’s Dublin datacenter to feed the grid; Arthur Cox — Ireland’s new connection policy for data centres; DCD — Microsoft to roll out battery-sharing worldwide; Global Market Insights — data centre battery market; UL 9540A test standard

Diesel generators are the part of a data centre that neighbours actually experience: the monthly test runs, the fuel deliveries and the stacks along the fence line. They are also the part of the industry changing fastest and most unevenly. The hyperscalers are pursuing different pathways beyond conventional diesel; much of the rest of the market has yet to articulate one. For assessors, the first question is no longer “how many generators?” but “what actually needs backup, for how long and why?”

Google has estimated that more than 20 gigawatts of diesel generator capacity stands behind the world’s data centres, most of it idle for all but a few hours a year. Every megawatt of new data centre capacity has traditionally brought a little more than a megawatt of diesel with it, sized for the day the grid fails and tested weekly or monthly whether it fails or not. In Australian development applications the generator yard is routinely the largest source of the three impacts communities raise first: noise, diesel exhaust and fuel storage risk.

Diesel has survived every attempt to retire it for three unglamorous reasons. Its energy density is exceptional: a litre of diesel holds roughly 36 megajoules, which is why multi-day autonomy is physically possible at all. The volumes are still substantial. Generators carrying 100 megawatts burn roughly 25,000 to 30,000 litres an hour, depending on set efficiency and loading, so two days of autonomy is well over a million litres. A 300 megawatt campus at the same duration is a fuel farm rather than a tank compound. Nothing else on offer comes close, which is the reason it endures, though it also explains why fuel storage sits alongside noise and exhaust in the objections communities raise. Days of autonomy are a choice rather than a universal standard: Uptime Institute sets a minimum of twelve hours of fuel storage at N load for all Tiers, with anything beyond that requiring a risk-based assessment of the facility’s energy supply. The equipment is proven across half a century and every failure mode is understood. It is cheap per kilowatt of standby capacity, precisely because it does almost nothing for decades. Any replacement has to beat that combination or change the question.

Who is moving, to what, by when

The largest operators are pursuing different pathways away from conventional diesel, which tells you something important: the successor has not been settled.

Microsoft has committed to eliminating its dependence on diesel fuel for data centre backup by 2030. It demonstrated a 3 megawatt hydrogen fuel cell system with Plug in 2022 and in January 2024 completed a separate 1.5 megawatt demonstration with Caterpillar and Ballard at Cheyenne, Wyoming, running a simulated 48-hour outage at 1,855 metres in below-freezing conditions. It also uses renewable diesel in its Swedish region. Google is working to 24/7 carbon-free energy by 2030 and has replaced a diesel generator with a lithium-ion battery system backing 3 megawatts of live production load at St. Ghislain in Belgium, arguing that batteries earn their keep by supporting the grid while diesel only ever waits. AWS, under Amazon’s net-zero 2040 Climate Pledge, is converting its European generator fleet to hydrotreated vegetable oil (HVO), beginning with Ireland, Sweden and Oregon in 2023: a drop-in fuel change rather than a technology change. Meta, targeting net zero across its value chain by 2030, began piloting HVO at Clonee in Ireland in 2024.

Two things are worth noticing about those timelines. The near-term moves (HVO) change the fuel, not the machine; the structural moves (batteries, hydrogen) are still at single-site scale a decade after the commitments were made. Both facts deserve to be part of every assessment conversation.

Meanwhile the AI build-out has scrambled the picture from the other direction. In the United States, grid connection queues have pushed AI campuses into generating on site as primary power: xAI has deployed hundreds of megawatts of gas turbines, OpenAI’s Stargate site ordered 29 turbines of 34 megawatts each and solid oxide fuel cell suppliers signed billions of dollars in data centre contracts in early 2026 alone. When a facility brings its own power station, the backup question inverts: the grid becomes the backup and the generator yard becomes the front of house, with an entirely different emissions, noise and assessment profile.

Backup is not one question

The biggest analytical error in assessing backup power is assuming that megawatts determine resilience requirements. They do not. Workloads do. A facility carrying financial transactions or emergency communications has a fundamentally different continuity requirement from an AI training campus whose jobs can checkpoint and restart, whatever their relative size. Facility class is useful shorthand because it bundles a typical workload mix with a typical set of contractual constraints, so it is where an assessment can reasonably start. Four classes, four different answers.

Colocation carries the strictest and slowest-moving requirements: tenant contracts specify redundancy and autonomy, so the operator cannot unilaterally swap the philosophy. HVO is colo’s realistic near-term path because it changes nothing the contract cares about. Hyperscale cloud holds the opposite position: workloads and services can be distributed across availability zones and regions, so resilience can be designed partly at the software and network layer rather than entirely within the facility. That does not automatically mean lighter backup, since hyperscale facilities still carry substantial systems, but it creates architectural options that conventional colocation contracts do not permit. That is exactly why the battery and hydrogen experiments are happening at hyperscale first. Edge facilities are small enough that for some of them batteries plus network rerouting may be the whole answer; their backup is architectural, not mechanical. AI facilities split in two: training loads can checkpoint and restart, which argues for less backup, yet the US experience shows AI campuses installing more generation than anyone, as primary supply. That contradiction exposes the question sitting underneath the whole section. Should backup requirements follow the size of the facility or the criticality of the workloads inside it? A 300 megawatt training campus whose jobs can restart does not carry the same continuity requirement as a far smaller facility supporting payments, hospitals or emergency communications. From an assessment perspective, very large AI campuses increasingly resemble power stations with attached computing infrastructure, rather than conventional data centres.

At the far end of the spectrum sits the counterexample. Cryptocurrency mining carries no backup on the mining load itself, because that load is interruptible by design and earns demand-response revenue by switching off when the grid is stressed. In Texas, flexible loads of this kind now represent around a tenth of forecast consumption. That is proof that a very large computational load does not inherently require backup generation. What decides the answer is what the load is for.

The reversal is no longer theoretical. During grid emergencies through 2026 the US Department of Energy repeatedly authorised PJM and other grid operators to call on backup generation at data centres and other large-load industrial and commercial sites, as a last resort before firm load shedding. Equipment installed to protect the data centre was being called on to help protect the grid. Backup fleets are becoming grid assets, which is an argument communities and assessors should hear as a benefit when the technology is clean and a cost when it is not.

The technologies compared

Technology	Autonomy and storage challenge	Local emissions	Lifecycle and operational characteristics	Standing in backup duty
Diesel generator	Days of autonomy, though at campus scale that means bulk fuel storage on site; deliverable anywhere	NOx and particulates at the fence line; ~2.7 kg CO2 per litre	Engines last 20-30 years at standby duty	The incumbent default
HVO (renewable diesel)	Identical to diesel; same tanks, same engines	Combustion emissions including NOx remain; local air-quality benefit depends on engine and operating conditions	Up to about 90% lower lifecycle greenhouse gas, though feedstock, certification and land-use impacts decide whether that claim holds	The pragmatic near-term move (AWS, Meta, Microsoft Sweden); supply chains still thin
Natural gas engines and turbines	Piped supply, no onsite storage; autonomy depends on the pipeline staying up	Lower particulates than diesel; NOx remains	Roughly a quarter less CO2 per unit of energy; upstream methane risk; mature, 20-30 years	Growing fast, but mostly as primary power for AI campuses rather than standby
Hydrogen fuel cells (PEM)	Very low volumetric energy density against liquid fuels; substantial onsite storage or delivery infrastructure required	Water vapour at point of use	Lifecycle depends entirely on how the hydrogen is made; stack life suits standby duty; refuelling logistics unproven at fleet scale	Demonstrated at 3 MW (Microsoft/Plug) and 1.5 MW (Caterpillar/Ballard); the long-term bet
Solid oxide fuel cells (gas-fed)	Piped supply	Lower NOx, no particulates	CO2 remains; high electrical efficiency; designed for continuous duty, stacks replaced within engine-lifetime timeframes	Primary power economics, not standby economics (Bloom’s 2026 contract surge)
Lithium-ion battery systems	Hours, not days; autonomy is the binding constraint	Zero onsite	Lifecycle follows the grid mix that charges them; 10-15 years with augmentation as cells fade	Proven at single-facility scale (Google St. Ghislain)
Long-duration storage (flow batteries, emerging chemistries)	Designed for many hours; footprint and cost rise with duration	Zero onsite	Long cycle life, low degradation; limited operating history at data centre scale	Not a diesel replacement yet, but the reason “hours, not days” should not be read as permanent
Linear generators	Piped or stored fuel; fuel-flexible including hydrogen and ammonia	Ultra-low NOx	CO2 follows the fuel; new technology, early fleet	A watching brief

The table’s honest summary: nothing on it beats diesel on all three of density, cost and proof at once. Everything on it beats diesel on emissions somewhere. The choice is therefore a design decision that should be argued in an application, not a default inherited from 1985.

What an assessor should ask

An assessor should be able to get clear answers to six questions.

1. What loads are genuinely critical? Which workloads must continue through an outage? Which can stop, shift or restart?
2. What autonomy is required and can it actually be delivered? How many hours or days, on what evidence, given the facility’s redundancy architecture. Raw tank volume is not the same as usable autonomy: a fuel system is only as good as its least redundant pump, day tank or transfer path, which is why the Uptime minimum is twelve hours while still meeting the facility’s topology objective. Where the figure assumes tanker resupply, that assumption should be tested against the events most likely to cause the outage in the first place, since bushfire and flood close roads at exactly the moment the fuel is needed.
3. Why this technology? What alternatives were considered, on what grounds was this fuel and backup architecture chosen over them?
4. What will neighbours experience? What are the noise, NOx, particulate, fuel-storage and testing impacts at the nearest receivers, given that test runs rather than outages are what neighbours actually experience?
5. Can the asset support the grid? Can storage, generation or flexible load provide demand response or other grid services rather than wait idle for an outage?
6. What is the transition pathway? If diesel or another transitional technology is installed at opening, what happens across the facility’s 20 to 30 year life?

The March national expectations put efficiency and community impact firmly into Commonwealth assessment policy and forthcoming standards are set to turn those expectations into obligations. Backup power intersects directly with both: the efficiency with which a facility uses and manages energy, together with what its operation imposes on the surrounding community. It is also one of the few parts of a data centre where practice is being rewritten in public, with named dates, by the largest operators on earth. Australian assessments can simply ask for it.

The bottom line

Diesel earned its place through density, cost and proof and it will not vanish by decree. But the era of backup-by-default is ending: the largest operators are pursuing different pathways beyond conventional diesel, the AI build-out has turned generators into power stations and interruptible designs have shown that some facilities need no backup at all. The question for an Australian assessment table is no longer how many generators a facility has. It is which loads genuinely need to survive an outage, what the neighbours will hear and breathe while the backup is tested and when diesel’s replacement arrives on site. Applicants should be prepared to justify all three.

A topical commentary from the Digital Infrastructure Institute. For assessment practice on backup power, noise and air quality, see our State & Territory Frameworks and community information series.

Sources: Data Center Knowledge — Google: batteries can replace generators; Baxtel — Google replaces diesel at St. Ghislain; DCD — Microsoft 3MW hydrogen fuel cell backup; Amazon — renewable diesel (HVO) for European generator fleets; Data Center Knowledge — replacing diesel in AI-scale data centres; US DOE §202(c) emergency orders, 2026

Every watt a data centre draws ultimately becomes heat somewhere in the system. Where that heat goes, into the local air, water, a district heating network or another productive use, is becoming an infrastructure question in its own right. For years the heat island question was dismissed as unmeasured speculation. In 2026 it became measurable from orbit and immediately contested. The mitigation toolkit exists, the reuse economics work in cold climates and the hot-climate half of the problem, the half Australia lives in, is close to wide open for innovation.

The argument changed this year. A study of more than 6,000 data centres worldwide, using two decades of NASA satellite land surface temperature data, reported that land surface temperatures around AI data centres rise by around 2 degrees Celsius on average after operations begin, with extreme cases near 9 degrees and measurable effects extending as far as 10 kilometres. A published critique responded that the signal mostly measures land-cover change rather than heat exhaust: roofs and car parks replacing paddocks read hot from orbit whether or not any waste heat reaches the neighbours. Importantly, satellite land surface temperature is not the same thing as the air temperature people experience at street level. Both sides are reading the same satellites and there is no agreed protocol for separating the two effects. That dispute is the finding. The heat island question is now measurable enough to argue about and assessment practice has nothing in place to settle it.

The physics is unforgiving

A data centre is a machine for converting electricity into computation and the computation into low-grade heat, nearly watt for watt. A campus drawing 100 megawatts continuously has roughly 100 megawatts of thermal energy to reject, recover or export, as warm air from dry coolers and condensers or as water vapour from evaporative systems, from warm-water temperatures up toward 60 or 70 degrees in newer high-temperature liquid cooling architectures: too cool to do traditional work, too warm to ignore. Three things compound the local effect. Clusters multiply it, which is why precinct-scale concentrations like Northern Virginia show the strongest signals; a single facility’s plume disperses, a corridor of them changes the neighbourhood’s thermal budget. Site design compounds it: dark roofs, hectares of car park and cleared vegetation are heat island generators before the first server switches on. Hot climates tighten the loop, because cooling plant working against a 40 degree afternoon runs harder and less efficiently, while poorly dispersed or recirculated rejected heat can worsen that condition locally, at precisely the hour the surrounding suburb is least able to absorb it.

At sufficient scale, a data centre is not only an electricity consumer. It is a continuous thermal plant. A 300 megawatt campus is simultaneously a 300 megawatt-class heat source, yet planning systems scrutinise the electrical connection in detail while barely asking where the equivalent thermal output goes. That asymmetry becomes harder to justify as facilities and clusters grow.

What mitigation looks like today

The toolkit has three tiers. The first is ordinary good site design, rarely demanded of data centres: high-albedo roofs and surfaces, retained and planted vegetation, water-sensitive landscaping and heat rejection plume design that considers height, velocity and dispersion rather than simply pointing fans at the sky. None of it is exotic; much of it is still absent from many applications because it is rarely required.

The second tier is raising the temperature of rejection. Liquid and immersion cooling capture heat at 40 to 60 degrees rather than diluting it into vast volumes of slightly warmed air. That single design choice converts waste from a disposal problem into a potential product, which is why heat reuse and liquid cooling are the same conversation.

The third tier is reuse and Europe has proven it at scale. Stockholm’s open district heating market buys waste heat from more than 30 connected data centres, a practice its utility has run for more than 25 years; the crucial innovation there is commercial, not thermal, because heat became a tradable product with a published price. In Finland, Fortum’s large heat pump plants at Espoo and Kirkkonummi began operating in May, with data centre waste heat being integrated progressively; fully implemented, that waste heat is expected to supply around 40 per cent of a 2 terawatt-hour district heating network serving roughly 250,000 people, using about three quarters of the heat the data centres produce each year. Germany has gone furthest: its Energy Efficiency Act requires new data centres to reuse a rising share of their waste heat, 10 per cent from July 2026, 15 per cent from July 2027 and 20 per cent from July 2028, measured through the Energy Reuse Factor, the same ISO metric we have argued belongs in Australia’s forthcoming obligations. Reuse is no longer experimental; it is legislated practice in the world’s fourth-largest economy.

The hot-climate problem

District heating solves the cold-climate half of the problem and Australia barely has district heating. Most of our data centre capacity sits in warm-temperate to subtropical cities with no winter heat demand worth the pipework, which is why European practice cannot simply be imported and why this is the genuinely open field. The candidate answers exist at various maturity. Absorption chillers can turn waste heat into cooling, the one product a hot climate wants year-round, but conventional machines need driving heat around 70 degrees or hotter and most data centres reject well below that. Research systems are attacking the gap from both ends, sorption chemistries that fire at lower temperatures and liquid cooling that rejects at higher ones; closing it, heat from computing driving cooling for computing, would be a globally exportable result. Controlled-environment agriculture works commercially where a greenhouse, vertical farm or aquaculture operation co-locates; Japan’s White Data Center farms eels on its cooling water. Water applications, desalination pre-heating and purification, suit coastal sites and are attracting research attention alongside more speculative pairings like carbon capture. Temperature matters as much as quantity: two campuses can reject identical megawatts with radically different reuse value, one exhausting air in the 30s, the other capturing liquid at 60. The honest constraint is thermodynamic: 35 degree heat carries little exergy and every reuse pathway fights that fact, which is exactly why raising rejection temperature through liquid cooling is the enabling move for all of them.

That changes the planning question. Heat reuse cannot be bolted on as an efficiency measure after a site is chosen. If heat is to become a resource, the potential users, their temperature requirements, the distances involved and the connecting infrastructure need to be part of the location decision, which is exactly how the Finnish plants came to sit beside district heating networks.

Is it an innovation field? Three gaps say yes

The first gap is hot-climate reuse at commercial scale, absorption and adsorption cooling above all. The challenge is turning low-grade data centre heat into an efficient, repeatable and commercially compelling system rather than a bespoke engineering demonstration. The second is market design: Stockholm’s tradable heat contract is a transferable invention waiting for an Australian precinct to host it, most plausibly where a planned data centre cluster neighbours industrial heat users, glasshouse agriculture or an aquatic centre. The third, once again, is the measurement layer. Satellite land surface temperature auditing is now demonstrably feasible at global scale, yet no Australian consent requires post-commissioning thermal monitoring, no application routinely models cumulative precinct heat and no efficiency metric in local use captures heat exported to the neighbourhood versus heat productively reused. The verification gap is the same shape we found in noise and in backup power; it is the recurring finding of this series.

What an assessor should ask

Five questions do the work. What is the facility’s thermal budget at the fence line, modelled as a plume and cumulatively with its neighbours, in the hottest week of the design year rather than the average one? What are the surface and landscape commitments, since albedo and vegetation are the cheapest heat mitigation on the site plan? At what temperature is heat rejected and what did the applicant assess for reuse, a question Germany now requires by statute and Australian assessments can simply borrow? What productive heat users exist within an economically viable radius of the site and was that assessed before the site was selected? What will be measured after commissioning, given that a heat island claim can now be tested from orbit and an operator confident in its design should welcome the verification?

The bottom line

Heat is the least regulated of the data centre’s three neighbourhood impacts, behind noise and water, largely because it was the hardest to measure. That justification is becoming increasingly difficult to sustain. The verification gap is now visible from orbit. The mitigation toolkit is real, the reuse economics are proven where climate allows and the hot-climate versions of both are an open field in which Australia has every reason to be the innovator rather than the importer. The March national expectations put efficiency, energy and community impact firmly into Commonwealth assessment policy and state frameworks are beginning to translate expectations into measurable requirements. The Energy Reuse Factor is how heat joins that conversation and assessment tables that start asking the five questions above will be ahead of both the regulation and the complaint letters.

A topical commentary from the Digital Infrastructure Institute. Companion to our analyses of backup power and data centre noise. For assessment practice, see our State & Territory Frameworks.

Sources: Marinoni et al. — The data heat island effect (arXiv preprint); Masley — critique: land-cover change, not heat exhaust; Eurelectric — Stockholm Exergi Data Parks tradable waste heat; AFRY — Fortum district heating from data centre waste heat, Finland; Germany’s Energy Efficiency Act (EnEfG) heat reuse obligations

New South Wales has put numbers around data centre efficiency and attached something valuable to meeting them: speed. The result is Australia’s most substantial state-level data centre framework yet. The bargain raises a second question: once a project earns its faster assessment on design performance, who verifies what the facility actually delivers? This analysis walks through what the framework does, the trade at its centre and the accountability loop still to be closed.

What the framework actually is

The framework has three working parts. The first is a set of guidelines built on six principles: world-class environmental and efficiency standards, no net cost to consumers and communities, funding for additional water and energy supply, enhanced local community infrastructure, investment in future industries across the supply chain and a demonstrated commitment to training and skills. The second is a planning bargain: qualifying projects are offered a streamlined pathway that includes Secretary’s Environmental Assessment Requirements within two months and a development application assessment taking no longer than 75 days in state government hands, supported by a dedicated concierge function inside the planning department. The third is cost allocation reform: regulatory changes intended to ensure data centres pay for the electricity network upgrades they require rather than shifting those costs to households, with the Independent Pricing and Regulatory Tribunal reviewing whether water pricing reflects the full cost of servicing these facilities.

Around the edges sit the machinery pieces: an Industry Advisory Forum, annual review of the guidelines and a new Office of AI within the Cabinet Office. Nineteen projects valued at $50.3 billion are currently in the State Significant Development pipeline, which is the scale this framework will immediately govern.

The numbers are real numbers

The most significant feature of the guidelines is easy to miss: they contain actual figures. Projects can qualify through one of two efficiency pathways. The first requires a design PUE of 1.25 or lower together with a design WUE of 1.0 or lower for potable water (1.6 for non-potable). The second allows a design PUE of 1.3 where design WUE is held to 0.44 or lower, recognising that power and water efficiency pull against each other in cooling design: the trade we walked through in our water analysis, where zero-water designs buy their savings with electricity. Both pathways are defined at mature utilisation, assuming 100 per cent IT load under average annualised climate conditions, with PUE and WUE measured under the ISO/IEC 30134 series. Water-intensive cooling must use recycled water for cooling operations or have a clear agreement with the relevant utility to transition to rainfall-independent supply and the Guidelines recognise that potable water may be used during that transition.

This matters because Australian assessment practice has mostly run on adjectives: world-class, best-practice, industry-leading. A numeric threshold can be checked. A claim of 1.25 either holds or it does not. For the first time in Australia, a government has written efficiency numbers into an approval gate and that alone moves the public conversation from assertion toward evidence. The Guidelines are candid about the starting point: existing planning requirements specify no water efficiency measure at all and the government’s own analysis found nine of twelve proposed projects reporting an average WUE of 1.0.

The trade at the centre

The bargain is speed for standards. For operators, assessment time is money at a scale most public debate underestimates: a committed hyperscale tenant, financed land and a connection agreement can make a year of planning delay worth more than any single compliance cost in the guidelines. A 75-day pathway is therefore a powerful incentive and attaching it to measurable commitments rather than to lobbying is good design. The cost allocation reforms complete the trade: the network upgrades a facility requires are to be paid for by the party creating the need, which addresses the most common and most legitimate community objection to data centre growth, that everyone else’s bill quietly carries it.

Taken together, the six principles read remarkably like the questions communities, councils and this Institute have been asking all year. Efficiency standards, full-cost water, network costs carried by the operator, local benefit, skills. The framework is in large part a governmental restatement of the assessment table.

What the published materials leave open

The thresholds in the guidelines are design values: design PUE, design WUE. A design value is a promise about how a facility will perform, made at approval time, before the facility exists. Operating performance is a different number and every practitioner knows the two diverge: real facilities run partial loads, hot summers, commissioning years and equipment ageing that no design calculation fully anticipates.

There is a further complication in the methodology itself. The design metrics assume mature utilisation at 100 per cent IT load. Real facilities may take years to approach that level and some never reach it. At lower utilisation the supporting infrastructure does not scale down in proportion to the IT load: UPS systems, transformers, pumps, fans and cooling plant carry fixed and part-load losses, while redundancy architecture keeps multiple systems running below their most efficient point. A facility designed for a PUE of 1.25 at full load may legitimately operate at a materially higher figure at partial load. Verification is therefore more than asking whether 1.25 became 1.25: it means understanding performance across the load profile the facility actually carries.

The published materials are detailed about how promises will be assessed and much quieter about how performance will be verified. Three questions follow directly. Who measures the operating PUE and WUE of an approved facility and how often? To whom are the results reported and are they public? What happens when a facility that qualified for the 75-day pathway at 1.25 runs at 1.4?

This is not a hypothetical concern. The government’s own demand analysis estimates that six of every seven megawatts of connection requests may represent phantom demand. That is a different issue from environmental performance, but it carries the lesson that matters for a fast-expanding market: commitments made before construction are not the same thing as outcomes delivered afterwards. A framework that rewards strong commitments with speed needs a mechanism that measures delivery with the same care.

The honest reading is that some of this machinery exists in outline. The Guidelines say applicants will need to demonstrate compliance, commitments made in an environmental impact statement can become conditions of consent and compliance is monitored through existing processes, including annual reviews and independent audits. That is more than nothing and less than a complete answer. Existing compliance processes can enforce conditions of consent, including commitments carried through from an environmental impact statement. What the Guidelines do not clearly establish is whether the design PUE and WUE values that qualify a project for streamlined assessment will translate into ongoing operational measurement, comparison and public reporting. Our noise and water analyses reached the same structural finding at facility scale and the NSW framework now poses it at market scale: the verification gap is the distance between what is promised at approval and what is measured in operation.

The distinction matters because verification cannot be reduced to checking a single number. A facility approved at 1.25 at full IT load should not be expected to report 1.25 at half that load; what should be tested is whether actual performance is consistent with the approved design at the utilisation and conditions actually experienced. That requires operating data, context and a repeatable method for comparing design intent with delivered performance. It is why the Institute’s assessment work treats verification as a distinct stage of the infrastructure lifecycle rather than an extension of development approval and why we have built tools to help communities, councils and decision-makers test whether commitments made at assessment can later be measured against operating outcomes. The NSW framework makes that capability more relevant, not less: measurable approval thresholds create the basis for measurable accountability.

What it means beyond New South Wales

Every other Australian jurisdiction now has a reference point. Councils assessing proposals outside NSW can reasonably ask why a project that would need to commit to 1.25 in Sydney should be assessed against adjectives elsewhere. States drafting their own approaches will be compared with this one. The federal Senate inquiry into the data centre build-out, due to report in November, now has a live state example of what codified standards look like. Tasmania, where major data centre proposals are currently being assessed without an equivalent state framework, illustrates the contrast particularly clearly. That is an observation about frameworks, not about any project.

The bottom line

The NSW framework is the most substantial piece of data centre policy any Australian state government has produced: real numbers, a real incentive and cost allocation that answers the fairest community objection. What its published materials leave open is the operating half of the bargain. NSW has written the performance promise and attached real value to making it. The next step is closing the accountability loop: measure what gets built, understand how it performs at the load it actually carries and verify that the operating outcome matches the approved design intent. The numbers now exist. The question is who verifies them.

A topical commentary from the Digital Infrastructure Institute. For assessment practice across all jurisdictions, see our State & Territory Frameworks.

Sources

• NSW Government, Nation leading framework to harness NSW data centre investment, ministerial release, August 2026
• Infrastructure NSW, Data Centres and the NSW Data Centre Guidelines, August 2026
• w.media, NSW puts data centres on fast track, but developers face higher costs, August 2026

Water is the data centre impact communities ask about first and understand least, partly because the industry often answers with a single number when the real story is a journey: in as drinking water, around the loop a few times, partly into the sky, the concentrated remainder to the sewer under a trade waste agreement, carrying the chemistry that kept the system safe. Follow the litre and the assessment questions write themselves. The engineering endpoint, meanwhile, has already been named: the first zero-water designs are already running.

A large evaporatively cooled facility can draw as much water as a small suburb and most public debate stops at that comparison. What the litre actually does between the meter and the sewer is better understood than almost any other industrial water use, because cooling water chemistry is seventy years old and thoroughly documented. The opacity is commercial, not technical: few operators publish what any water treatment engineer could describe from memory.

The journey

The litre arrives as makeup water, usually potable, sometimes bore water, increasingly recycled supply. It joins the condenser loop and heads for the cooling towers, where the system’s entire purpose is partial sacrifice: a fraction of the circulating water evaporates and that phase change is what carries the heat away. Evaporation is the principal consumptive loss: that water leaves the local liquid system as vapour. Most of the remainder eventually leaves as blowdown, with a much smaller share lost as drift.

What does not evaporate becomes more concentrated with every pass, because evaporation leaves the dissolved minerals behind. Operators track this as cycles of concentration: at four cycles the circulating water carries four times the dissolved solids of the incoming supply. Systems commonly run between two and four cycles and well-managed systems with suitable source water and treatment can reach six or more; push higher without that support and scale formation and biological risk climb steeply. To hold the balance, the system continuously bleeds off a share of the concentrated water as blowdown, typically carrying dissolved solids several times more concentrated than the incoming supply, with the actual level set by the source water chemistry and the cycles of concentration. Blowdown leaves through a trade waste agreement to the sewer, which is where the water story meets the regulatory system. A final small stream leaves as drift: fine airborne droplets carrying whatever the water carries, which is why drift eliminators and their maintenance matter more than their obscurity suggests.

Closed-loop chip cooling can change the journey dramatically, but a closed loop alone does not make a facility waterless. Direct-to-chip and immersion systems circulate a fixed charge of treated water or coolant indefinitely, yet the heat they collect still has to be rejected somewhere and that final stage may itself be evaporative. In genuinely zero-evaporation designs, the closed loop rejects heat through dry or mechanically assisted cooling without consuming water.

What is in the water

Four broad classes of additive keep an evaporative system safe and efficient. Scale inhibitors, led by phosphonates such as HEDP and PBTC with polymeric dispersants alongside, hold calcium carbonate in suspension instead of on heat exchange surfaces. Corrosion inhibitors protect the metals: azoles for copper, phosphates and molybdates for steel. Biocides are the environmentally consequential class, oxidising agents such as chlorine and bromine rotated with non-oxidising chemistries to prevent microbial resistance, with Legionella control the non-negotiable reason the dosing never stops. pH control and anti-foam agents complete the program. Modern sites increasingly meter all of this through automated dosing tied to real-time water chemistry rather than timers, which cuts both chemical consumption and risk.

Those treatment chemicals, their residuals and reaction products become part of the circulating water chemistry and much of that inventory leaves the system in the blowdown. Makeup volume tells us what the facility demands from the catchment or utility; blowdown quantity and chemistry tell us what it returns to the wastewater system. A credible assessment needs both and the blowdown half is the number applications least often state.

Can the litre be cleaned and used again?

Yes, and at a steadily falling cost. Side-stream filtration keeps circulating water cleaner and buys extra cycles of concentration, which is the cheapest recycling there is because it simply makes the first use last longer. Reverse osmosis can recover a substantial share of the blowdown for return to the towers, though recovery depends on the water chemistry and pretreatment and it leaves a concentrated brine that still has to be managed. Full zero-liquid-discharge, evaporating the brine to solids, is proven but energy-intensive and reserved for water-scarce sites. The more elegant move runs the journey backwards: source the makeup from recycled water in the first place, as Google has done with reclaimed municipal supply and Singapore does systematically with NEWater. Where recycled water infrastructure is available, recycled makeup plus higher cycles plus recovery on blowdown can cut potable demand dramatically using entirely conventional equipment; what is missing is usually the connection to a recycled water main, which is a planning question rather than an engineering one.

The innovations

The structural innovation is the same one reshaping heat and noise: move the cooling to the chip. One example is Microsoft, whose new designs adopted since August 2024 use a closed-loop, zero-evaporation architecture, with next-generation deployments at Phoenix in Arizona and Mount Pleasant in Wisconsin, where the first facility is now operational, and a claimed saving of more than 125 million litres per facility per year. Their fleet water use effectiveness had already fallen 39 per cent since 2021 to around 0.30 litres per kilowatt-hour. Around that headline sit quieter advances: smart dosing replacing scheduled chemical injection, non-chemical treatment using filtration, UV and electrochemical methods to shrink the biocide inventory, plus the water replenishment pledges of the major operators, which fund genuine catchment projects but should be read as offsetting alongside reduction, not instead of it.

The honest caveat belongs in every conversation about zero-water designs: closed loops and dry coolers trade water for electricity. Rejecting heat without evaporation generally takes more fan and compressor energy, though the size of the penalty depends on climate, coolant temperatures and the heat rejection design; Microsoft reports only a nominal energy increase for its zero-water architecture. Water use effectiveness and power use effectiveness still pull against each other at the margin and a facility quoting only its best number is telling half the story. In a dry inland catchment the trade is usually worth making; in a jurisdiction with abundant clean energy and stressed water it is worth mandating; either way the assessment needs both numbers.

What an assessor should ask

The new NSW Guidelines establish the performance outcome. The next question for an assessor is what evidence demonstrates it. Seven questions cover the journey. What is the full water balance, makeup, evaporation, blowdown and drift, at design load in the hottest week of the year, including peak daily demand? What cycles of concentration will the system run and what treatment supports that claim? What is the additive inventory and the blowdown quality entering trade waste, stated in the application rather than left to the sewer authority to discover? What is the Legionella management plan, since public health sits inside this system, not beside it? Was recycled makeup assessed and if refused, why, given the reference projects exist? What will be reported after commissioning and will actual operational WUE and PUE, potable and recycled volumes and peak demand be published against the design commitments made in the application? What is the source water hierarchy for normal operation and for drought and what happens to the facility’s demand when restrictions apply? None of these is onerous; all of them are answerable from the design documents of any competent applicant.

The bottom line

Water is where community concern and industry opacity meet head-on, yet the litre’s journey is the best understood story on the site. The chemistry is documented, the recycling technology is conventional, recycled makeup is proven and the zero-water endpoint is now a build program with dates rather than a research aspiration. What remains scarce is disclosure rather than technology and that is an assessment choice, not a technical constraint. The March 2026 national expectations put sustainable water use squarely inside the assessment conversation, New South Wales wrote water thresholds into its new framework this week and the Australian Government has announced that forthcoming standards will make water efficiency a legal obligation for large facilities. An assessment table that asks for the journey, not the summary number, will find that the industry already knows every answer; it has simply been waiting to be asked.

A topical commentary from the Digital Infrastructure Institute. Part of our series on what neighbours actually experience, alongside backup power, noise and heat. For plain-English community information on data centre water use, see our community information series.

Sources: Microsoft — next-generation datacenters consume zero water for cooling; DCD — Microsoft closed-loop, zero-water evaporation design; DCD — Google recycled water at Georgia data centre; DGTL Infra — data centre water usage guide

Noise is the data centre impact neighbours actually live with, yet most consent conditions measure it in units that cannot hear the problem. The fan makers have delivered real innovation, the unit makers offer quiet versions of everything and the hardest bands of the sound spectrum remain wide open. For assessors, the fix starts with what gets measured.

When Chandler, Arizona voted down a data centre proposal, noise was central to the objection. Across Northern Virginia, the densest data centre market on earth, residents describe a hum that follows them indoors, while compliance reports insist the facilities are within their limits. Both things are true at once and the reason is the unit of measurement. Most noise conditions are written in A-weighted decibels, a scale designed around the sensitivity of human hearing to mid and high frequencies. Cooling plant emits much of its energy below 200 hertz, where A-weighting discounts it steeply. A facility can meet an LAeq limit precisely while at the same time producing the low-frequency drone that generates every complaint. The difference between C-weighted and A-weighted sound levels (dBC–dBA) is widely used as a screening indicator for low-frequency noise. When the difference approaches or exceeds about 15 decibels, it suggests that low-frequency energy is becoming a significant part of the sound and that A-weighted measurements alone may not fully represent the low-frequency character experienced by nearby residents.

Three problems, not one

Data centre cooling noise is three distinct problems that are typically lumped together under one name. Broadband fan noise across the mid and high frequencies is largely solved using a combination of distance, orientation, acoustic barriers, louvres and shrouds delivering 5 to 15 decibels of attenuation at known cost. Tonal noise is harder: dozens of fans running at similar speeds produce energy concentrated at the blade-pass frequency and its harmonics; near-identical tones then beat against each other to create the wandering, pulsing character that people find disproportionately annoying at modest measured levels. The genuinely unsolved problem is the low-frequency band from 20 to 200 hertz, where wavelengths run from 1.7 to 17 metres. Sound at that scale diffracts over barriers, carries for kilometres, passes through standard housing construction and excites the room resonances that turn a distant plant item into a bedroom problem.

Below all of that sits infrasound, the energy under 20 hertz that is felt rather than heard: pressure sensations, rattling windows and the sense of a presence nobody can point to. Data centres genuinely produce it, mostly from large slow fans whose blade-pass frequencies can fall into the teens of hertz and from generator exhausts under test. The honest science: measured infrasound levels near well-designed facilities generally sit below human perception thresholds and the direct-harm evidence is weak, but the secondary effects are real, because building elements rattle at these frequencies and the rattle is audible even when the source is not. Infrasound also travels furthest of all and no barrier touches it, so the fixes live at the source: fan selections that keep blade-pass frequencies above the band, vibration isolation on the structure-borne paths that dominate down here and exhaust silencing on the generator side. Measurement has its own convention, G-weighting under ISO 7196 plus unweighted third-octave data below 20 hertz; modern monitors log it cheaply, which removes the last excuse for not looking. One development suits this band unusually well: active noise control, marginal outdoors at mid frequencies, works best against long coherent low-frequency waves in confined paths such as ducts and exhausts, which is where the emerging products aim.

What the manufacturers have actually done

The most substantial innovation sits a tier below the brands on the equipment schedule. The axial fans inside most branded cooling units come from specialist makers and the leaders have spent two decades on aeroacoustics. Ziehl-Abegg’s owlet series is the emblem: sickle-shaped blades with serrated trailing edges copied from the barn owl’s wing and a rippled leading edge in the latest generation, claiming up to 12 decibels of improvement over its predecessors, in many cases removing the need for separate silencers. ebm-papst’s equivalents follow the same logic. Because noise falls steeply as blade speed drops, the other quiet revolution is simply diameter: bigger, slower fans moving the same air.

The unit makers integrate this work into tiered offerings. Stulz builds its low-noise chillers around oversized 910 millimetre fans, encapsulated compressors and redesigned internal airflow. Vertiv’s Liebert chiller ranges ship in standard, low-noise and quiet versions with EC fans and acoustic insulation. Schneider’s Uniflair line offers the same laddered configurations. The operational lever matters as much as the hardware: EC fans obey the fan laws, so a facility willing to raise its chilled water temperature overnight can slow its fans and shed several decibels exactly when limits tighten and backgrounds fall. Free cooling reduces compressor hours. Liquid and immersion cooling remove the server-level fans but the heat still leaves the site through outdoor plant, so the fence-line question relocates rather than disappears.

The honest summary is that manufacturers compete hard on the A-weighted sound power of a single unit and that competition has worked. Nobody yet sells the thing communities need: a site that remains genuinely quiet at the property boundary, not simply a collection of individually quiet components.

The open ground

Three gaps stand out, all of them live territory for innovation.

The first is passive low-frequency control. Conventional splitter attenuators create back-pressure, which forces fans to run faster, which makes more noise: a genuinely vicious loop. Ventilated acoustic metamaterials break the loop in the laboratory, using arrays of Helmholtz resonators and tuned membranes to deliver around 10 decibels of insertion loss in the 100 to 200 hertz band while letting air pass. At the time of writing, they are not yet commercialised at data centre scale.

The second is tonal management across the fan array. The beating hum is an interaction effect between dozens of similar fans, which means it can be attacked in software: coordinated speed control that deliberately spreads blade-pass frequencies so tones neither align nor beat. The control hardware already exists in every EC fan on site. Specialist consultancies have also shown that small aerodynamic retrofits can remove specific tones at source for a fraction of the cost of enclosure. Both approaches remain boutique solutions.

The third is measurement itself. Continuous, community-visible noise monitoring with C-weighted and third-octave data published alongside the A-weighted compliance figure would move most disputes from assertion to evidence. The technology is ordinary; the practice is rare.

What an assessor should require

The assessment fixes follow directly. Write limits that can hear the problem: an A-weighted limit alone is not adequate for plant of this character, so add a C-weighted or third-octave low-frequency criterion and a tonality penalty of the kind long standard in the UK’s rating method, with G-weighted infrasound measurement where large fan or generator installations sit near homes. Require the night-mode question to be answered in the application: what do the fans do between 10pm and 7am and what water temperature supports it? Make continuous monitoring with published data a condition rather than a complaint response. Ask how the fan array is managed as an array, not as a catalogue of individually compliant units. Require the plant schedule to state the acoustic tier actually purchased, because every major manufacturer sells a quieter version of the same unit and the delta is a procurement decision made years before anyone complains.

The bottom line

Data centre noise is not an unsolvable problem; it is a mismeasured one with a thin market for the hardest fixes. The fan makers have done their part and the unit makers will sell quiet to anyone who specifies it. What remains open is the low-frequency band, the site-level soundscape and the verification gap; the first movers in each will find communities, councils and operators equally motivated. In the meantime, a regime that measures only in A-weighted decibels risks overlooking the part of the spectrum that communities most often identify as problematic.

A topical commentary from the Digital Infrastructure Institute. For plain-English community information on data centre noise, see our community information series; for assessment practice, our State & Territory Frameworks.

Sources: Chandler City Council vote (Dec 2025); EESI — communities raising noise pollution concerns; WUSA9 — the Northern Virginia hum; Ziehl-Abegg — biomimetic fan concepts; AskNature — owl-wing fan design; Noise Monitoring Services — data centre noise control and dBC–dBA screening; ISO 7196:1995 (infrasound G-weighting)