Government sets AI rules before the window closes—but offers few details on what they'll actually do
Labor says AI data centres will pay their own energy and water costs — but the rules aren't written yet, and the investment decisions won't wait.
When a government announces a policy position without announcing a policy, it is worth asking what exactly has been decided. The Prime Minister's address at the University of Sydney this week was full of conviction about artificial intelligence and almost empty of specifics. That is not necessarily a failure of ambition. It may be a deliberate bet: move early to claim the regulatory space, then fill it in before the money arrives.
The data centre rules are sensible in principle — the question is whether they'll exist in practice
The concrete part of the announcement is relatively narrow. Large-scale AI data centres would have to underwrite new electricity generation, pay the full cost of their grid connection, and contribute at least as much energy to the grid as they draw from it. They would be required to minimise water use and fund any additional water infrastructure their operations need. The intent is clear: prevent the cost of AI's enormous energy appetite from being quietly socialised onto household electricity bills and municipal water systems.
These are sensible obligations in principle. The pattern elsewhere has been exactly what the government says it wants to avoid: data centres securing generous grid access and water allocations because they arrived before regulations existed to price those costs properly. Locking in the principle now, before billions in infrastructure are committed, is the logical moment to do it. The Prime Minister acknowledged as much directly, noting that negotiating conditions after a facility is built is a structurally weaker position.
Announcing the principle is not the same as setting the standard. The former costs nothing; the latter determines whether the social licence the Prime Minister spoke about is a real condition on development or a rhetorical frame around deals that were always going to happen on the industry's terms.
The harder question is what the standards will actually say. South Australia is the only state with a dedicated data centre framework. Queensland last month flagged that it would not adopt what it described as "underdeveloped national proposals" around renewable energy requirements. Getting eight jurisdictions to agree on a consistent mandatory framework at next month's national cabinet meeting, then legislating it early next year, is an optimistic timeline for something that does not yet have the technical detail written in. The harder the specifics, the harder the consensus.
The copyright promise is politically resonant but legally underdeveloped
On copyright, the announcement was even lighter on mechanism. The Prime Minister offered artists a clear assurance, telling them they would maintain control over their work and be able to name their price in any deal with AI companies. "Anything less is theft," he said. That framing is politically resonant but legally underdeveloped. Australia's copyright act dates from 1968, and the question of whether training a large language model on copyrighted material constitutes infringement has not been settled here or in most comparable jurisdictions. The assurance was not backed by draft legislation, a model for licensing structures, or a timeline for either.
The government still holds leverage — but only while the investment decisions remain unmade
What the government does have is leverage it has not yet spent. Major AI developers are seeking the kind of policy certainty and regulatory goodwill that translates into data centre approvals, grid access, and a stable operating environment. The government's position, implicitly, is that those things come at a price, and that the price includes acceptable terms on copyright and energy costs. The Prime Minister said directly he had met with the major companies and made the position clear.
Whether that leverage holds depends on how badly those companies want Australian infrastructure specifically, and how credibly the government can threaten to withhold it. The answer to the first question is probably "quite a bit," given Australia's geographic position, political stability, and renewable energy potential. The answer to the second depends on whether the regulatory framework gets written before the investment decisions do.
That is the real race underneath the announcement. The policy substance the government will consult on over coming months needs to arrive before the window it described closes. Announcing the principle is not the same as setting the standard. The former costs nothing; the latter determines whether the social licence the Prime Minister spoke about is a real condition on development or a rhetorical frame around deals that were always going to happen on the industry's terms.
The government has claimed a position. Now it has to defend it with something specific.
Sources
ABC News — Labor aims to shape AI revolution with new data centre and AI copyright rules
Frequently Asked Questions
What are Australia's new AI data centre rules actually requiring?
Large-scale AI data centres would have to fund their own grid connection, contribute at least as much energy to the grid as they draw, minimise water use, and cover any additional water infrastructure costs. These are announced as principles — the measurable standards that give them legal force have not yet been written.
Why does it matter if the rules come before or after the investment decisions?
Once billions in infrastructure are committed, governments lose most of their negotiating power — companies can credibly argue that changing conditions mid-build is sovereign risk. The entire logic of announcing standards now is to set the price before developers lock in their Australian plans, not after.
Do Australian copyright laws protect artists from AI companies using their work?
Not clearly. Australia's Copyright Act dates from 1968 and has not been updated to address AI training data. Whether scraping copyrighted material to train a large language model constitutes infringement is unsettled in Australia and in most comparable jurisdictions.
What happens if states don't agree to the national AI data centre framework?
The framework depends on all jurisdictions adopting consistent mandatory standards — Queensland has already signalled it will not accept what it calls underdeveloped national proposals on renewable energy requirements. A patchwork result would allow data centre developers to choose the most permissive jurisdiction, undermining the national standard entirely.
How much leverage does Australia actually have over major AI companies?
Probably meaningful leverage, given Australia's geographic position for Asia-Pacific operations, political stability, and renewable energy potential — all of which major developers have flagged as attractive. But that leverage diminishes the longer it takes to produce a regulatory framework, because companies will make infrastructure commitments with or without Australian policy certainty.