Australia regulates data centres through multiple systems, but cannot trace one facility across them

What the current records cannot show

The submission asks whether Australia’s existing regulatory systems can identify data-centre facilities consistently, show which rules apply to them, and produce comparable information about their energy and water use. It addresses term of reference (a) and the energy and water elements of term of reference (b); other community, industry and biodiversity impacts are outside its scope.

Key findings

1. No public crosswalk was found connecting the records for the same data-centre facility.

   NGER reporting, grid connection, planning, environmental approval, water and rating systems use different names, identifiers and facility boundaries. In the public sources reviewed, the public cannot reliably determine which records describe the same physical site.

2. Most data-centre records in the NABERS extract used a scope that excludes IT load.

   Ten of the 11 data-centre records returned by the 15 August extract used the Infrastructure rating scope. Because that scope excludes IT equipment, the rating cannot by itself show the total energy performance of the data centre. When we checked again on 14 September, six of the 11 ratings remained current, five of them on the same scope. Five of the 11 entries were no longer current one month later, although no facility had closed. This shows why the register needs a stable reporting boundary and a record of expired ratings.

3. The proposed 30 MW threshold sits in the National Electricity Rules, which do not operate in Western Australia or the Northern Territory.

   The AEMC’s Package 2 draft places the threshold in rules that apply only in NEM jurisdictions. The advice we reviewed did not explain how the threshold would apply to facilities outside the NEM, including off-grid and islanded sites.

4. Published water figures do not use a consistent measurement boundary.

   One company’s FY2025 reporting labels the same 773 ML figure as both “consumption” and “withdrawal”. Without a common definition, figures from different facilities or reporting systems cannot be compared reliably.

5. One EPBC referral classified a data-centre proposal with about 1,038 MW of gas generation as renewable energy.

   Searches based only on industry category can therefore miss relevant proposals.

   Updated 14 September 2026 from the referral document; the lodged submission used an earlier industry-media range of 900–1,100 MW. The proposal was determined a controlled action on 4 September 2026.

Recommendations

1. Use one facility ID

   Assign each facility a persistent identifier, then publish a crosswalk connecting its non-sensitive energy reporting, connection, planning, environmental, water and rating records.

2. Publish the rules that apply to each facility

   Publish which energy, environmental and water frameworks apply to each facility, including NEM, non-NEM and off-grid sites.

3. Use common energy and water definitions

   Before facilities are compared, define what each figure covers: whether an energy figure includes IT equipment, and whether a water figure means withdrawal or consumption.

Open questions

The submission lists seven matters we could not verify. They include how NABERS star ratings correspond to PUE values, whether NABERS has published a timetable for a data-centre water framework, the final connection arrangement for one of the two Northern Territory proposals, and the water and planning instruments of all eight jurisdictions. Schedule 1 of the Safeguard Rule contains no production variable for a data centre, so the consequence for an off-grid facility is recorded as an uncertainty. We did not obtain legal advice.

Sources

The lodged submission is titled Regulatory data gaps that limit policy oversight and independent verification; this webpage uses a shorter summary title. The committee has authorised publication as Submission 89. Authorisation to publish is an administrative decision and does not indicate that any finding has been accepted.

Corrections

Two corrections have been issued since lodgement. Each is recorded with the date of the change, the wording as lodged, the corrected version, and the source relied on. The lodged PDF is not altered, as a lodged document should not be; this page carries the corrections.

Related research