Australia wants AI data centres to pay their own way

The office is real. The rules are not.

Australia has picked a very physical place to start its next round of AI policy: the power lines, water systems and large buildings behind the models.

On 15 July, the federal government opened an Office of AI inside the Department of the Prime Minister and Cabinet. It also announced a plan for national standards covering large data centres and AI training infrastructure.

One part happened today. The other is still a proposal.

National Cabinet is expected to discuss the plan in August. The government says legislation should follow in early 2027. Until a bill appears, we do not know which facilities will be covered, who will enforce the rules or what happens when a company breaks them.

Who pays for the power?

The clearest part of the plan is about electricity.

Large data centres would have to underwrite their own new power supply and pay their share of grid-connection costs. Operators could also be asked to reduce demand when the electricity system needs help. Water use and the location of new facilities would be part of the rules too.

This sounds dry. It is not. A new data-centre campus can need a substation, new generation and a lot of cooling. The bill does not disappear. Someone pays it.

Australia's position is simple: that cost should sit with the project, not quietly move onto household energy bills or other businesses.

From expectations to law

The country already published a set of expectations for data centres in March. Developers were asked to add clean power or storage, cover network costs, use water responsibly, invest in local skills and support Australian research.

Those expectations had some weight. Projects that matched them could be prioritised during federal assessments. But they did not create new legal duties.

The July announcement is the next step. The government now wants parts of that approach turned into nationally consistent law. It calls the framework world-leading and says it would be the first legislated system of its kind. For now, those are government claims. There is no legal text to test them against.

A copyright promise, with details missing

The announcement also makes a promise to Australian writers, artists and journalists. The government says they should keep control of their work and that AI companies should not train on it without that control.

That is clear as a political message. It is much less clear as a rule.

There is no proposed consent system, licensing model, disclosure requirement or enforcement process in the release. It also says nothing about models trained overseas or datasets built from material already online.

So creators do not have a new right they can use today. They have a government commitment. The difficult part comes next.

Two offices, two jobs

Australia already has an AI Safety Institute in the Department of Industry, Science and Resources. Its work is technical. It analyses advanced models, looks for emerging harms and helps regulators understand what new systems can do.

The new Office of AI sits much closer to the centre of government. Its job appears to be coordination: turning policy into something that can work across energy, planning, copyright and consumer protection.

The exact split between the two bodies has not been explained. That matters. Testing a model is one job. Deciding what government does with the evidence is another.

What comes next

The interesting thing about this plan is where it draws the line. Most AI rules focus on what a model says or does. Australia is also looking at what the system consumes before anyone types a prompt.

Power connections, water use and infrastructure costs can be measured. They also affect local communities long before a new model launches.

The harder questions are still open. Australia needs to define a large data centre, publish the bill, name the regulator and explain how its copyright promise would work. The government says broader consumer-safety priorities are coming in the next few weeks.

For now, there is a new office and a clear message: if AI infrastructure creates a large bill, the public should not automatically pick it up.