Case study 03 · Research series

AI data centres: the public-interest test

Are Australian AI data-centre impacts significant, do benefits outweigh costs, and can approval systems cope?

What is being tested: Australia's AI data-centre build-out earns public-interest support only if nationally manageable impacts, genuinely additional clean energy, local benefit and enforceable governance can be demonstrated.

Evidence available to 24 June 2026
Three reports
Australia

Cumulative series result

CONDITIONAL / NOT YET

The build-out can pay off for Australia, but only under binding governance that Australia has already described and not yet made enforceable.

What this does not mean: This is neither blanket approval nor blanket opposition: national manageability does not override evidenced local breaches, and announced connection capacity is not built load.

Four linked tests

Each link remains visible

The cumulative result depends on the significance, net-benefit and adequacy findings staying intact.

R1 — SignificanceREJECTED AS A WHOLEDo the expected national and local impacts remain below the report's pre-committed significance thresholds?

R2 — Net benefitCONDITIONALLY SUPPORTEDDo benefits outweigh costs, and what would have to be true for Australia to retain a net benefit?

R3 — Process adequacyREJECTED / NOT YETCan Australia's approval and planning machinery govern the build-out adequately?

Series — Public-interest caseCONDITIONAL / NOT YETDoes the combined evidence justify public-interest support now?

Report extracts

Read the tests in sequence

Report 1 — When Does It Count? Testing the Significance of Australia's AI Data-Centre Build-Out to 2035

The build-out is nationally manageable on most axes, but fails the emissions test without additional renewables and fails locally where persistent breaches are evidenced. The report does not establish a breach at every named location.

Read Report 1 extract

Report 2 — Do the Benefits Outweigh the Costs? Australia's AI Data-Centre Build-Out and What Would Have to Be True

Benefits can outweigh costs, but only with effective governance. The large productivity gains often attributed to AI use are not the same thing as the smaller and more leakage-prone gains from locating data-centre infrastructure in Australia.

Read Report 2 extract

Report 3 — Can the Machinery Cope? Whether Australia's Approval and Planning Processes Are Adequate to Govern the AI Data-Centre Build-Out

Australia has relevant laws, approval pathways and emerging policy expectations, but the machinery is not yet adequate to make the series' public-interest conditions enforceable, coordinated and transparent.

Read Report 3 extract

Evidence discipline

Reading controls

Retrospective synthesis: the series spine was created after the reports and is not the original Report 1 hypothesis.
Connection requests are not built load; the AWS-commissioned phantom-demand estimate is useful but its confidence is discounted.
The local breaches carrying Report 1's rejection are qualitative judgements against pre-committed tests, not numeric cut-offs.
The 40–45%, 70–75%, 25–30% and 26% figures are author-assessed estimates, not forecasts or observed probabilities.
The NSW and federal inquiries are open evidence; their outcomes are not pre-empted.
Additionality means genuinely new clean generation and storage, not annual matching against existing clean supply.

Retest rule: New evidence changes the published result only through documented evidence review and Product Owner acceptance.