Civic Interplay / Sovereignties

AI Sovereignties

A live map of the infrastructure behind AI in Australia, and the public debate forming around it. Data centres, mines, refineries, the energy and water they draw, and the lines of ownership running through them. Each point is a real site on the ground.

Mapping AI Sovereignties

Open the map in a new tab

Not loading? Your browser may be blocking the embedded site. Use the open in a new tab link to launch it directly.

What you're looking at

This is a living atlas of an increasingly contested historical terrain. The primary focus is on critical infrastructure required for the national AI sovereignty agenda to be realised on Australian soil. This focus includes data centres, mines, refineries, and related energy or water infrastructure. The legend does lots of work. Different ownership and governance models are called out: not just who owns but also who approves, and who contests.

Whose voices are heard

The Australian debates are following closely behind widespread public disapproval of data centre construction, particularly in areas facing water stress or growing energy prices. The tracker is listening to who is contesting the infrastructure roll-out, and those who advocate for it. The opposition to data centre approvals is taking place at a time when many government agencies are onboarding AI enablement officers, and the AI capabilities are becoming essential in the workforce. Australians have historically been fast adopters of new technology, and this wave of agentic disruption is no different. Many Australians are enjoying the MCP frisson, even if they begrudge the data centre expansion.

This is intentionally being crafted out in the open. A focus is on approval rates, the practices of resistance being deployed, and the roles played by different government agencies, whether local government, State Government agencies, or Federal Government. The different patterns for AI stack building are of great interest. A hyperscale data centre in a territory facing water shortage demonstrates a high degree of allegiance to hyperscale-subject relations, at the expense of other forms of civic belonging. Investment in a mix of local compute, AI innovation ecosystems and critical-curious minds, combined with reflexive self-awareness of the dangers of deepening dependencies, underpinned by supportive agencies and alternate infrastructures and 'AI enablement' is the preferred pathway. All eyes and ears are open for this pattern in the landscape.

See the contestation record, the trackers behind the map, in Notion.

The rise and rise of a term

The term "AI sovereignty" went from near-absent to everywhere-all-at-once in a short time frame. The chart below captures the change. Very low in 2022, a spike in early 2025 (Nvidia, Davos) and a peak continuing into 2026.

Line chart of global news coverage mentioning 'sovereign AI' or 'AI sovereignty', rising from near-zero in 2022 to a peak in 2026.

Source: GDELT news-coverage volume (relative), data and method. More in the project explainer.

How fast, how quietly

Most of Victoria's large data centres are approved by the Minister for Planning under the Development Facilitation Program, not by the local council. The public-notice step is exempted, so there is no window to object and no appeal to VCAT. Each step on the chart is a facility waved through without exhibition. The one application that met organised objection was one that went to public comment: a proposal in Perth, withdrawn before a decision.

Cumulative step chart of Victorian data-centre permits approved under the Development Facilitation Program between April 2025 and May 2026, each granted without public notice, with a marker noting the GreenSquareDC proposal in Perth withdrawn after 1,836 objections.

Source: Victorian Ministerial Permits Register (Planning Victoria). Confirmed approvals only; per-permit public-notice status is being verified. Drawn from the Critical Infrastructure Tracker.

Where the data comes from

The map sources data from a Critical Infrastructure Tracker, a database built in Notion updated every two weeks through a mix of agent-based research and human-in-the-loop weaving. The data is sourced agentically, supported through Claude Code, with compute costs also tracked recursively. The contestation record is built from planning submissions, local and national press, local governments and other inquiry material. Corrections are welcome; the effort feels (for now) gestural rather than exhaustive.

Spot a missing site, an error, or contestation we have not captured? Submit it here.

Methods, citation and licence

Formally, this is a living atlas of contesting and curating AI sovereignties (Australian view). How each entry is sourced and classified is set out in the project explainer. Entries are coded by agent and flagged for human review (agent, human, or human-verified), so the basis of each classification stays visible.

The project's methodology, classification and written analysis are openly licensed under CC BY 4.0. The underlying records are compiled from public sources, cited per entry; the facts themselves carry no such claim. Cite as: Sarah Barns, A living atlas of contesting and curating AI sovereignties (Australian view), Civic Interplay, 2026. doi.org/10.5281/zenodo.21026430.