Carbon Accounting for Data Centres in Australia: Where Your Emissions Actually Hide

Canberra Data Centres spent 18 months arguing that its greenhouse gas emissions were a trade secret. The Clean Energy Regulator disagreed. In March 2023, the CER rejected CDC's application to have its NGER data withheld from public release, and the numbers landed in the public register: 187,568 tonnes CO2-e for FY2021-22. Almost all of it — 187,397 tonnes — was Scope 2. Electricity. Just keeping servers on and rooms cold.

That single operator produced more emissions than most entire corporate groups ever will. And it wasn't an outlier. NEXTDC reported 350,049 tonnes of Scope 2 emissions in the same year, up 16% on the prior period. These are the kinds of numbers that blow through NGER corporate thresholds (50,000 tonnes) before anyone in the compliance team has finished their morning coffee.

Carbon accounting for data centres in Australia is about to get harder. Not because the methodology is complicated — it's actually more concentrated than mining or construction — but because the numbers are growing at a rate that makes last year's baseline feel like a rounding error. And if you're operating, building, or investing in data centre infrastructure right now, the emissions question isn't academic anymore. It's a regulatory obligation that's expanding in step with your power draw.

The Scale of What's Coming

Australian data centres consumed roughly 3.9 TWh of electricity in FY25. That's about 2% of the National Electricity Market. In FY18, it was 1.4 TWh. So consumption has nearly tripled in seven years.

Here's the trajectory. Between 2.2 GW and 3.2 GW of data centre capacity is expected to be operational by 2035, up from about 0.3 GW in 2024-25. AEMO projects data centres could account for 8-11% of Australia's projected electricity consumption across the NEM and WEM by 2035. Some analysts put the high-demand scenario at 15%.

The investment pipeline tells the story. AWS committed AU$20 billion between 2025 and 2029, announced by its CEO alongside Prime Minister Albanese in June 2025. Microsoft pledged AU$5 billion to expand from 20 to 29 centres across Sydney, Melbourne, and Canberra — a 250% capacity increase. OpenAI and NEXTDC announced a $7 billion hyperscale AI campus in western Sydney. Between 2023 and 2025, announced data centre investment in Australia exceeded $100 billion in total.

This isn't just tech industry hype. It's physical infrastructure being built right now, consuming electricity at industrial scale, in a country where grid decarbonisation is still a work in progress. Every one of those facilities will need to account for its emissions. Many will trigger NGER reporting thresholds. And the largest will fall under the Safeguard Mechanism.

Scope 2 Dominates Everything

If you've worked in carbon accounting for mining, you're used to Scope 1 being the main event — diesel, methane, process emissions. Data centres flip that completely. Scope 2 electricity emissions typically account for 90-95% of a facility's operational carbon footprint. The CDC numbers tell you everything: 187,397 out of 187,568 tonnes was Scope 2. That's 99.9%.

This makes state grid factors arguably the most important variable in the entire emissions profile.

Running the exact same data centre workload — same IT load, same cooling overhead, same PUE — in different states produces wildly different emission numbers under location-based accounting. Using the NGA Factors 2025:

State	Grid Factor (kg CO2-e/kWh)	Emissions from 50 GWh/yr
Victoria	0.78	39,000 t CO2-e
Queensland	0.67	33,500 t CO2-e
NSW & ACT	0.64	32,000 t CO2-e
WA (SWIS)	0.50	25,000 t CO2-e
South Australia	0.22	11,000 t CO2-e
Tasmania	0.20	10,000 t CO2-e

A 50 GWh facility — modest by hyperscale standards — produces 39,000 tonnes in Melbourne versus 10,000 tonnes in Hobart. Same servers. Same workload. A 3.9x emissions difference driven entirely by what's feeding the grid.

This is the point where we need to be honest about something. Most of Australia's data centres are concentrated in Sydney, with Melbourne as a secondary hub. Sydney's added 1,820 MW of rated capacity over the last decade. Neither city has particularly low grid factors. The majority of Australia's data centre emissions are coming from states where the grid still leans on coal and gas generation.

The implication for anyone doing carbon accounting: if you operate across multiple states, you can't use a single national average factor and call it accurate. Each facility needs the correct state-based factor from the NGA Factors workbook. And if you're doing site-selection analysis for new builds, the grid factor should be as prominent in your modelling as land cost and fibre connectivity. A facility in Adelaide that uses twice as much electricity as one in Melbourne still produces fewer emissions.

PUE: The Metric That Multiplies Your Problem

Power Usage Effectiveness measures the ratio of total facility energy to IT equipment energy. A PUE of 1.0 means every watt goes to compute. A PUE of 1.5 means you're burning 50% extra on cooling, lighting, power distribution, and everything else that keeps the IT running.

The Australian average PUE sits around 1.3-1.44 depending on which survey you reference. The government target for federal data centre panel providers is 1.4 or lower. Hyperscale operators claim PUEs as good as 1.15-1.19. Older enterprise facilities? We've seen PUEs pushing 1.8 or worse.

For carbon accounting, PUE isn't just an efficiency metric. It's a direct multiplier on your Scope 2 emissions.

If your IT load draws 10 MW and your PUE is 1.5, your total facility draw is 15 MW. That extra 5 MW isn't doing compute — it's running chillers, fans, UPS losses, and power distribution. In Victoria, that overhead adds about 34,000 tonnes of CO2-e per year to your emissions profile. Dropping your PUE from 1.5 to 1.3 at the same IT load saves 15,600 tonnes per year in Melbourne. That's not a marginal improvement. That's the difference between tripping the NGER corporate threshold and staying below it.

If you don't know your actual PUE, your carbon accounting is guesswork. Full stop. We've talked to facilities that estimate their PUE based on design specifications from five years ago, not metered data. A 0.1 difference in PUE might look small on paper, but for a 20 MW facility in NSW, it's roughly 11,200 tonnes CO2-e per year. You can't afford to approximate this number.

Scope 1 Is Small but Not Zero

Data centre Scope 1 emissions are usually less than 1% of the operational footprint. But "small" in absolute terms can still mean hundreds of tonnes, and it's the category people forget about until an auditor asks.

Diesel backup generators are the obvious one. Every tier-rated data centre has them. They run during grid outages and get tested monthly. The testing alone — typically 1-4 hours per month per generator, depending on your maintenance schedule and tier requirements — consumes diesel. A facility with ten 2 MW generators running monthly tests of two hours each burns roughly 10,000-12,000 litres of diesel per year just on testing. At 2.7 kg CO2-e per litre, that's about 27-32 tonnes. During an actual outage, a full facility running on diesel for 24 hours will burn through 30,000-40,000 litres depending on load.

Under NGER, diesel consumption for stationary energy gets reported as Scope 1 emissions using NGA Factors. You need litres consumed, not litres purchased — which means you need fuel reconciliation between deliveries and tank dip readings. We've seen operators report diesel purchases as consumption, which overstates emissions if tank levels haven't drawn down equivalently.

Refrigerant leaks are the sneaky one. Data centre cooling systems — whether DX, chilled water, or hybrid — use refrigerants. R-410A has a global warming potential of 2,088 under AR5. That means a single kilogram leaked is equivalent to roughly 2.09 tonnes of CO2-e. A large facility with multiple CRAC/CRAH units and chillers might carry several hundred kilograms of refrigerant charge across the cooling plant. Industry-standard leak rates for comfort cooling systems run around 5-10% per year.

If your cooling system carries 500 kg of R-410A and leaks at 7% annually, that's 35 kg lost — roughly 73 tonnes CO2-e. Not massive, but not nothing. And it's a number that only appears in your emissions inventory if someone is actually tracking refrigerant top-ups and reconciling against total charge.

We're not sure how many Australian data centre operators are accurately tracking refrigerant losses against their cooling plant inventory. Based on what we've seen across other building types, the answer is "not enough." The records tend to live in HVAC maintenance logs, not in the emissions accounting system.

NGER: The Thresholds Most Data Centres Will Hit

A data centre consuming 100 TJ of energy per year triggers the individual facility NGER threshold. How much is 100 TJ? It's about 27.8 GWh — roughly a 4 MW continuous load at the meter. Any facility drawing more than 4 MW from the grid is in NGER territory. For context, most colocation facilities being built today are in the 10-100 MW range. Even a small 5 MW facility is above the threshold.

At the corporate group level, NGER reporting kicks in at 50,000 tonnes CO2-e or 200 TJ of energy consumed. A corporate group running data centres across multiple states will almost certainly exceed this.

The penalties for getting NGER wrong aren't theoretical. The ANAO found that 72% of NGER reports contained errors, 17% of them significant. The Clean Energy Regulator's penalty unit value is $330, and penalties for late or incorrect reporting can stack quickly. Beach Energy accepted an enforceable undertaking in July 2025 after NGER reporting issues — and they're a company with significant compliance resources.

For data centre operators specifically, there's a wrinkle worth flagging. If your corporate group triggers NGER thresholds, you're automatically pulled into ASRS Group 2 mandatory climate reporting from 1 July 2026. That means full sustainability disclosures under AASB S1 and S2, including governance, strategy, risk management, and metrics. It's not just an emissions number anymore. NGER is the on-ramp to a much broader mandatory disclosure regime.

There's also the Safeguard Mechanism. Any single facility exceeding 100,000 tonnes CO2-e Scope 1 is covered. Data centres are unlikely to hit this on Scope 1 alone. But remember: Scope 2 electricity emissions are Scope 1 emissions of the electricity generator. If the policy ever shifts to include Scope 2 at the facility level — and there are international precedents for this — large data centres would be firmly in scope. Worth watching.

The PPA and GreenPower Question

Most large data centre operators have either signed renewable PPAs or are in active procurement. Microsoft signed a 315 MW green PPA in Australia. Equinix secured 115 MW to cover its 17 Australian centres. The industry has broadly committed to 100% renewable energy by 2030.

Here's where it gets technically interesting for carbon accounting.

A renewable PPA or GreenPower contract only reduces your market-based Scope 2 figure. Your location-based Scope 2 — the number AASB S2 requires you to disclose — stays the same regardless of what certificates you hold.

Under AASB S2 paragraph 29(a)(v), location-based Scope 2 is mandatory. Market-based is supplementary. You can report both. You should report both if you hold contractual instruments. But you can't swap one for the other.

For NGER, the default is location-based. Since 2023-24, voluntary market-based Scope 2 reporting has been available under section 7.4 of the NGER Measurement Determination. But since the 2025-26 reporting year, if you opt into market-based reporting, you must do it for all facilities under your corporate structure. No cherry-picking the sites with PPAs and going quiet on the rest.

And then there's the residual mix factor. Any electricity not covered by a contractual instrument (LGC, PPA, GreenPower) gets assigned the national residual mix factor of 0.81 kg CO2-e/kWh under NGA Factors 2025. That's higher than any individual state's location-based factor. So if you're only 60% covered by a PPA, your market-based number for the remaining 40% will actually be worse than the location-based number for that portion. We see people get tripped up by this regularly.

Scope 3: The Embodied Carbon of Everything in the Room

This is where data centre carbon accounting diverges sharply from the conversation about Scope 2 electricity. Schneider Electric's research found that Scope 3 emissions represent 38-69% of a data centre's total lifetime carbon footprint. And for colocation providers, the question of who owns which Scope 3 emissions gets complicated fast.

IT equipment — servers, storage, networking gear — is the biggest driver. Servers account for 58-85% of total embodied carbon across a data centre's lifetime. They contain rare earth metals, semiconductors, plastics, and complex supply chains originating primarily in Asia. The embodied emissions are front-loaded: most of the carbon cost is incurred at manufacture, not during operation.

For operators, this means the server refresh cycle is an emissions event. Replacing 1,000 servers every four years locks in embodied carbon upfront. Schneider Electric's modelling suggests extending server lifespan by one year cuts cumulative embodied carbon by about 16%. But there's a trade-off: newer servers are often more energy-efficient, and the operating emissions savings of upgrading to more efficient hardware can outweigh the embodied cost of replacement. We don't think there's a universal right answer here — it depends on the specific efficiency gain per generation and the grid factor where the facility operates.

The building itself contributes too. Concrete, steel, and aluminium in the facility shell, raised floors, and infrastructure carry significant embodied carbon. Using our construction emissions reference points — concrete at ~0.2 kg CO2-e/kg, steel at ~2.2 kg, aluminium at ~20 kg — a large data centre build uses thousands of tonnes of each.

Under ASRS, Scope 3 reporting is deferred by one year for all groups. But it's coming. And for data centre operators, the boundary question is genuinely hard. If you're a colocation provider, the servers in your facility are owned by your customers. Are their embodied emissions your Scope 3 (purchased goods and services), or your customer's Scope 1/2/3? The GHG Protocol provides guidance, but the practical application in multi-tenant facilities is — to put it plainly — still messy.

Water: The Reporting Obligation Nobody Prepared For

This isn't strictly carbon accounting, but it's heading toward the same regulatory bucket and data centre operators need to be across it.

Large data centres using evaporative cooling can consume staggering volumes. The Climate Council reported that some major facilities use up to 40 million litres per day — equivalent to the water needs of 80,000 households. A 1 MW data centre can use up to 25.5 million litres annually for cooling alone.

Water Usage Effectiveness (WUE), measured in litres per kWh of IT energy, ranges from 0.01 for air-cooled facilities to 2.5 for fully evaporative systems. The Australian average sits around 1.8-1.9 L/kWh depending on cooling technology.

Under AASB S2, climate-related risks and opportunities include physical risks from water scarcity. An Australian data centre in a water-stressed region — and parts of western Sydney certainly qualify — needs to disclose how water availability affects its operations. TNFD-aligned reporting, if it becomes mandatory, will sharpen this requirement further. The data infrastructure for tracking water consumption is often the same metering and billing pipeline as electricity. If you're already processing utility bills for emissions, adding water volume tracking is operationally straightforward.

The 5-Star NABERS Problem

From 1 July 2025, all data centres hosting Australian Government federal workloads must meet a minimum 5-star NABERS Energy rating. The Digital Transformation Agency monitors compliance, and non-compliant providers can be suspended or removed from the federal procurement panel.

A 5-star rating roughly corresponds to a PUE of 1.4 or lower. If your facility is running at 1.6, you're not just inefficient — you're potentially locked out of the fastest-growing customer segment in the market. Government contracts for classified data centre infrastructure alone total billions (AWS's AU$2 billion defence contract being the high-profile example).

The NABERS rating requires metered energy data broken down by IT load and total facility load. This is the same data you need for accurate carbon accounting. The same data your NGER return depends on. And the same data an ASRS auditor will ask for when verifying your Scope 2 numbers under ASSA 5010 assurance requirements. It's one dataset serving three compliance obligations.

How to Get This Right

We've now worked with enough energy-intensive facilities to know what the practical bottlenecks look like. For data centres specifically, three things matter more than anything else.

Get metered data, not estimates. PUE calculated from design specs is fiction. You need real power metering on the IT load and total facility load, at least monthly, ideally continuously. Same for diesel consumption — tank dip readings, not delivery invoices. Same for refrigerant — weigh your charge and track every top-up.

Apply the right state factor to the right facility. If you operate across NSW and Victoria, don't average. Each site gets its own factor. The difference between NSW's 0.64 and Victoria's 0.78 is 22% — enough to flip whether a facility is above or below a reporting threshold. We wrote a full guide to NGA Factors and how to apply them correctly.

Automate the bill processing. A data centre with 50-100 electricity meters, multiple diesel deliveries, water bills, and refrigerant maintenance records generates a significant volume of source documents per quarter. Manually keying these into spreadsheets introduces errors at a rate of 1-4% per entry. At the scale data centres operate, that error rate compounds into material misstatements. This is exactly the problem Carbonly.ai was built to solve — AI-powered extraction of consumption data from utility bills, matched to the correct emission factor, with a full audit trail for when the assurance team comes knocking.

The data centre sector's emissions profile is structurally simpler than mining or construction. It's overwhelmingly Scope 2. The calculation methodology isn't hard. But the volume of data, the speed of growth, and the regulatory obligations stacking up — NGER, ASRS, NABERS, Safeguard Mechanism — mean you can't wing it with quarterly spreadsheet updates anymore. Not when a single facility is producing more CO2-e than most mid-cap companies' entire operations.

If you're operating or building data centre infrastructure in Australia, the compliance clock is already running. The question isn't whether you'll need to report. It's whether you'll have the data to do it accurately when the deadline arrives.

If your data centre portfolio is growing faster than your ability to track its emissions, talk to us. We process electricity bills, diesel records, and refrigerant logs at the speed your operations actually move.

Related Reading: