Key Highlights

  • Canberra is pitching the country as a stable, resource-rich home for the world's AI factories.
  • Australia has spent a generation selling the world its iron ore, coal and gas.
  • The ambition is no longer confined to speeches.
  • The scale under discussion is large.

Canberra is pitching the country as a stable, resource-rich home for the world's AI factories. The catch is whether the electricity network can be built fast enough to carry the load.

Australia has spent a generation selling the world its iron ore, coal and gas. Now the pitch from Canberra is that the next great export may be something less tangible: artificial-intelligence compute, sold as processed digital capacity rather than raw commodities shipped offshore.

The ambition is no longer confined to speeches. In recent months Deputy Prime Minister Richard Marles and Assistant Minister for Science and Technology Andrew Charlton travelled to San Francisco to court Nvidia, OpenAI and Anthropic alongside Microsoft, Google and Amazon, according to reporting on the government's outreach. The proposition on offer is that Australia can host the sprawling data centres that train and run frontier AI models, drawing on political stability, land, and one of the world's deepest pipelines of renewable energy.

A pipeline measured in the hundreds of billions

The scale under discussion is large. Industry analysis of the proposals suggests Australia's data-centre Investment pipeline could exceed A$155 billion, with schemes potentially adding up to 10 gigawatts of new capacity, though these figures represent proposals and expressions of interest rather than approved construction.

Individual deals are beginning to give the ambition shape. Anthropic has reportedly been weighing 300 to 500 megawatts of Training capacity from Australian providers, with CDC Data Centres named as a front-runner, while OpenAI has signed a memorandum of understanding with ASX-listed NEXTDC contemplating a GPU cluster at its S7 campus in western Sydney valued in the billions. Domestic operators including NEXTDC, CDC, AirTrunk and Firmus Technologies would carry much of the physical build.

The logic of "exporting compute" is that these facilities serve global customers. A model trained or served in Sydney can be sold to users anywhere, capturing high-value activity onshore without a container ship in sight. Proponents argue it is a way for a small, energy-rich economy to attach itself to the most capital-intensive technology build-out in history.

The grid is the binding constraint

The obstacle is physical. The Australian Energy Market Operator's 2026 Electricity Statement of Opportunities projects data-centre electricity use rising roughly seven-fold, from about 5 terawatt-hours today toward 34 terawatt-hours over the next decade — moving from around 3 per cent of operational Demand toward roughly 13 per cent. Building generation and transmission at that pace, while the grid is simultaneously decarbonising, is a formidable task.

The connection queue tells its own story. AEMO has noted that more than 40 per cent of known data-centre projects have either dropped out or regressed in the connection process since 2025, a sign that securing firm power, not Capital or customer demand, is the real bottleneck.

Canberra's response has been to try to turn that constraint into a condition of entry. Under a framework the Albanese government has described as a world first, large AI data centres would be expected to function as "net generators, not net users" — putting at least as much energy into the grid as they draw out, Underwriting new renewable generation and firming, and paying full connection and water costs rather than passing them to households. A new Office of AI within the Department of the Prime Minister and Cabinet is to coordinate standards, with legislation targeted for early 2027.

From ambition to delivery

Whether the export dream materialises will hinge on execution. Definitions of which facilities count as "large," and how quickly renewable generation and transmission can actually be connected, remain unsettled. The commodities Australia has traditionally exported came out of the ground; AI compute must be built, powered and cooled first. Investors and hyperscalers alike will be watching whether the grid can keep pace with the pitch.