Sydney’s scrapped ‘Project Mars’ reveals the real choke point for AI: siting the megawatts

Goodman Group’s withdrawal of its proposed A$1.2bn, 90-megawatt “Project Mars” data centre in Sydney is the kind of story that punctures the floaty rhetoric around “AI growth” and forces the debate back onto physics, pipes, and permission slips. The BBC reports the facility would have been built about 20 metres from the nearest home in Lane Cove, and that the consultation drew 374 submissions with only nine in support BBC. That’s not a communications problem. That’s a feasibility problem.

Ninety megawatts is industrial infrastructure: an always-on load that must be fed, cooled, and backed up. Whether you call it “cloud” or “AI,” the substrate is the same: grid interconnection capacity, redundancy, heat rejection, construction disruption, and a local risk profile residents can see and feel. When a project is sited metres from houses, “social licence” becomes a first-order constraint, not a late-stage stakeholder-management exercise.

Goodman attributed its decision to changes in the “policy and regulatory environment” since it began the planning process in March 2025. That phrasing matters because the BBC also notes New South Wales has moved toward a fast-track assessment pathway—75 days—if projects meet standards and fund the extra water and energy supply they require. In other words, the state is signaling: you can go faster, but you may have to internalise more of the infrastructure bill.

This is where AI meets reality. The winning projects won’t be the ones with the biggest press releases; they’ll be the ones with bankable power and water arrangements, acceptable locations, and a credible plan for who pays for upgrades. If Australia wants to be a “global destination” for data centres, as the BBC describes, the bottleneck won’t be developer ambition—it’ll be where the megawatts go, and how much the host community has to absorb to make that possible.