Science1 distinct publisher3 min readPublished
The Perth site is now a wetland planting project. The argument attached to it: Australia regulates AI models closely but leaves Traditional Owners outside the siting decisions that settle land, water and power.
The Scientist · Science desk
Compiled by The ScientistSomething wrong?How this is made
Four Australian projects appear in this account. They are worth counting before a trend is drawn from them [20]. Two were contested: the Perth site, and a proposed AI Hub in Victoria where objections centred on cultural heritage and environmental impacts [10]. Two are being negotiated: Western Australia's largest data centre, pursued through an Indigenous Land Use Agreement that makes the Karajarri People co-owners of the project [12], and Energy North's proposed Northern Territory facility, where the company says it is engaging Traditional Owners as part of the assessment process [13]. There is no sampling frame, and no count of how many Australian proposals sit on land carrying heritage values. The case that consultation is the binding constraint therefore rests on the mechanism, not on a measured rate.
The mechanism shows up in the timing. The proposal was cancelled in May 2026, before Western Australia's Development Assessment Panel considered it [4]. A project that ends there leaves nothing behind for the next applicant to read: no conditions, no reasons, no assessed record of what the panel would have accepted. The thing this does not tell you is why the developer withdrew. The account reports the cancellation and reports that campaigners praised it [5], and offers no company statement of reasons [19]. Opposition preceding a withdrawal is not the same as opposition causing one, and on this evidence the two sit in sequence rather than in a demonstrated causal chain.
Size is the other number to handle carefully. A 120-megawatt facility [3] run flat out for a year would draw 120 x 8,760 hours, or about 1.05 terawatt-hours [18]. That arithmetic is an upper bound and not a forecast: the figure as published does not say whether 120 MW describes computing load or grid connection capacity, what utilisation the operator expected, or how much cooling water the design would need. What the account does say is that data centres consume land, water and energy in quantity [16], and that this river feeds Perth's drinking water as well as local ecosystems in the warmer months [8].
The standard the opponents invoked was free, prior and informed consent [9], which is a consent test rather than a notification exercise, and which Indigenous challenges to data centres in North America have leaned on through the UN declaration [17]. Australian policy, according to this account, has not translated that into siting rules: few policies require Aboriginal leadership in decisions about AI infrastructure [14], even as regulatory attention concentrates on the technologies themselves [11]. Kaley Nicholson, chief executive of the Federation of Victorian Traditional Owner Corporations, calls the government's current expectations for data centres "an investment failure" because they make no reference to collaboration with Traditional Owner groups [15].
The Karajarri arrangement is the interesting datapoint, because it solves the problem by contract rather than by statute [12]. Co-ownership converts a party who can only object into a party with something to lose from delay. That is a repeatable structure, and it is available now, without waiting for a policy that requires it. My reading, conditional on this being four cases rather than a survey: for anyone siting compute in Australia, the expensive uncertainty sits upstream of any model rulebook, in whether the people whose Country carries the load have a seat before the drawings are lodged.
Ranked by verification strength, evidence, and original report placement.
In July 2026, more than 200 community members gathered at Mandoon Bilya (Helena River) in Perth's eastern suburbs and planted more than 3,000 native plants as part of a wetland restoration project.
Until recently Mandoon Bilya was the proposed site of a hyperscale AI data centre scheduled to be one of Australia's biggest.
The proposed facility was a 120-megawatt AI data centre.
In May 2026 the proposal was cancelled before being considered by Western Australia's Development Assessment Panel.
Campaigners widely praised the cancellation and have since shifted their focus toward broader ecological restoration plans.
The campaign was led by the Bibbul Ngarma Aboriginal Association and supported by environmental and community groups.
Distinct publishers with included, body-backed reporting in this cluster.
phys.org
1 article · August 28, 2026
Follow any of these and your For You feed starts watching them — no settings page required.
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
One telling, proponent absent
The 120 megawatts, the May cancellation, the 3,000 plants — all of it comes from a single researcher-authored piece carried by phys.org. There is no panel document, no application reference, no named company, and no second outlet in our coverage that has been over the same ground. What is checkable is the quotation from Kaley Nicholson and the general resource argument; what is not is the specification and fate of the project the story is named for.
Partnership is the exception, by the story's own count
Asked to show Indigenous authority over AI infrastructure siting in Australia, this reporting can produce two instances: the Karajarri Indigenous Land Use Agreement with co-ownership in Western Australia, and Energy North's stated consultation in the Northern Territory. Against that, it says few policies require Aboriginal leadership at all. The shortness of the list is the finding, not a gap in the writing.
Restrained prose, thin paper trail
Nothing is oversold in the writing — the case made is a policy gap and a defeated proposal, not a movement sweeping the country. The stretch is documentary. 'One of Australia's biggest' and the withdrawal itself arrive with no developer, no reference number and no stated reason, and a national pattern is argued from four named projects. The derived annual draw of roughly 1.05 terawatt-hours is our own arithmetic on an unverified nameplate figure, and should be read as a ceiling, not a forecast.
Authors with a thesis, developers with no microphone
The writers' own fieldwork on Aboriginal communities' agreements with the resource industry supplies both the framing and the conclusion, and the piece ends on a prescription: Aboriginal people at the table before concept design. That position is held openly, which is the honest version of this. It is also the only position in the room — the proponent's reasoning, the Development Assessment Panel's view, the Western Australian government's, and any account from the Victorian AI Hub's backers are all missing, and Energy North appears only through its own statement.
Plausible spine, unchecked in every direction
Low for structural reasons rather than suspicious ones. The factual backbone is unremarkable and probably accurate; nothing in it strains belief. But with one account and an absent counterparty we cannot say why a 120-megawatt project died in May 2026, and that is the fact the headline, the argument and the restoration scene all lean on.