At a glance
Prime Minister Anthony Albanese has announced an Office of AI inside his department to coordinate national AI standards, ending the government’s case-by-case approach.
For data centres, Albanese promised "greater clarity and speed for approvals" and a legal obligation on large-scale operators to underwrite new power supply.
Planning approval is rarely the binding delay. Grid connection is, with AEMO’s 5.4GW data centre queue facing multi-year waits.
On copyright, Albanese said the law would spell out that creators retain ownership and control of their work, with the mechanism left to the Attorney-General's consultation.
Large data centres would have to be net generators, putting at least as much energy into the grid as they take out, with legislation aimed at Parliament in early 2027.
The pitch to operators
The speech had been anticipated for a week as the moment Labor would set out how it intends to govern AI. On 15 July, in a keynote at the University of Sydney, Albanese used it to declare an end to the government’s case-by-case approach and to put himself at the centre of what comes next. The centrepiece is a new Office of AI, to sit inside the Department of the Prime Minister and Cabinet, that will coordinate new Australian Standards across energy, copyright, productivity, education and labour rights. He cast AI alongside civil aviation in the 1920s and genetics in the 1990s, technologies the government eventually wrote coordinated rules for. The office extends the National AI Plan the government launched in December 2025, its framework for backing AI infrastructure while managing its harms.
For operators, the relevant line was about approvals. Albanese said getting the settings right would “enhance our appeal to international investors, by delivering greater clarity and speed for approvals, and a streamlined process for verifying compliance.” Assistant Minister for Science, Technology and the Digital Economy Andrew Charlton said the government wanted an “Australian approach” to large data centres that avoided the overseas mistakes on energy, water and community impact. On copyright certainty, which Anthropic has tied its investment to, the plan changed nothing on the day and pointed to the Attorney-General’s existing consultation.
Large data centres would have to be net generators
The speech carried a second obligation the approvals line does not. Albanese said the government would "create a legal obligation for the next generation of large-scale data centres to underwrite new power supply. To pay their full share of grid connection, so no costs are passed on to homes or businesses." He set the quantum in the same passage, saying large data centres should "put at least as much energy into our grid as they take out of it. To be net-generators, not net-users."
White & Case reads the obligation as requiring new capacity, with a standard power purchase agreement over existing generation unlikely to suffice. Albanese said operators would "build new renewable generation, and firming, to strengthen our national energy resilience", and that the rules would require data centres to minimise water use, maximise energy efficiency and pay for any additional water infrastructure required. The national expectations published in March would be folded in. "Australia will be the first country in the world to bring these issues into a single, national framework," he said, describing the result as "clear, consistent and mandatory."
The timetable runs through the states. The government will seek agreement from premiers and chief ministers at National Cabinet in August, and aims to bring legislation to Parliament in early 2027. No capacity threshold has been announced for which data centres fall in scope.
The bottleneck is the grid
Faster approvals are worth having. They are not the only constraint that decides when a gigawatt-scale build energises. The other constraint is power, and specifically the queue to connect to the grid. AEMO has disclosed a 5.4GW pipeline of data centre load seeking connection, with lead times now measured in years. A planning approval that arrives faster does not move a connection date that is set by transmission capacity and the connection process.
The same holds for the workload that fills the building. Anthropic’s 1.4GW Australian tender, a build of up to US$15 billion, asks bidders to bring firm, contracted power, favouring on-site generation or deals with independent power producers, because grid supply is the hard part. A fast-track that speeds planning while leaving the connection queue untouched addresses the shorter half of the timeline. The plan does not yet describe a connection pathway or a priority queue for compliant projects, which is where the delay actually sits.
Speed comes with conditions
The second half of Albanese’s approvals line, a “streamlined process for verifying compliance,” points at the national expectations for data centres published in March, which set benchmarks on national interest, energy, water, skills and local capability. Freedom of information documents reported on 13 July showed the government folding those expectations into the foreign investment review. Read together, the trade on offer is faster approval in return for demonstrable alignment with the expectations.
That lands into public wariness. A 2026 Lowy Institute poll found 64 per cent of Australians believe AI’s risks outweigh its benefits, up 12 points since 2024, and opposition to specific data centre sites has spread across five states, which we track in our work on data centres and social licence. For operators, streamlined verification rewards the ones who can already evidence their energy and water position, and raises the bar for those who cannot. The government is separately weighing whether host communities should be paid directly.
Big speech, few new levers
An Office of AI and a doctrine speech is a coordinating step. Australia has been here before. Its Investor Front Door, launched to fast-track major projects in the national interest, ran a pilot of four projects worth up to A$20 billion, and data centres were not among them. The instrument was facilitation and a dedicated case manager, short of a statutory fast-track.
The United Kingdom paired planning with connections and price. Australia has moved on the energy obligation and left the connection queue alone.
Approach | Planning | Grid and power | Capital committed |
UK AI Growth Zones | Designated zones, consent target 18 to 12 months | Connections accelerator, price relief from April 2027 | Up to £100 billion targeted |
Australia Investor Front Door | Case-managed facilitation, no data centres in pilot | No dedicated energy pathway | Up to A$20 billion, four pilot projects |
Albanese Office of AI, 15 July | Promised “clarity and speed”, detail to come | Net-generator obligation, legislation aimed early 2027; no connection reform | None announced |
Source: UK Government AI Growth Zones; Department of Industry, Science and Resources; Prime Minister’s 15 July speech, 2026.
The approvals half of the plan is a modest mover, signalling lower policy risk without changing the connection queue. The power half is not modest: a net-generator obligation changes the economics of every large build. It becomes material when the Office of AI turns coordination into binding standards, when a defined data centre pathway and connection reform follow, and when the copyright question is settled. Until then the 1.4GW Anthropic is chasing still depends on power and copyright, not on planning speed, a dependency set out in its National AI Plan memorandum with the government. The reaction split immediately. Former industry minister Ed Husic called giving big tech social licence “doomed to failure”, independent MP Allegra Spender said the government should have acted sooner, and UNSW AI Institute’s Toby Walsh welcomed the government “waking up” to a more interventionist stance. The divide runs inside the government too, between Industry Minister Tim Ayres and Charlton, the keenest on investment, and Attorney-General Michelle Rowland, holding the line for creators whose bodies have rejected the tech sector’s offer of a fund in exchange for a copyright exemption.
The likely path from here
None of this moves capital until the coordination becomes a mechanism, and on balance that will take longer than the speech implied. Expect the Office of AI’s first Australian Standards to land as guidance before they carry force, so the near-term signal to operators is direction, with any obligation to follow later. The harder test is whether a defined data centre approval pathway and a grid-connection priority follow the speech, and the “clarity and speed” line likely stays rhetorical until connection reform is on the table, which is the item to watch most closely. On copyright, a licensing mechanism looks more likely than an exemption, though the Attorney-General’s consultation will move slowly, and the NSW data centre inquiry reporting by 30 September is likelier to add conditions than lift them.
If the standoff holds, the cost is more than a delayed announcement. Anthropic has said it will take any Australian capacity deliverable by mid-2027 and has cast the country as a second home for training, so that workload can shift offshore if power and copyright are not resolved in time, and data centre investment has been a notable recent contributor to Australian growth. The narrower risk is that folding the expectations into the foreign investment review adds friction without a matching pathway, raising the bar to invest while the power constraint stays unaddressed.
This is a standoff, and both sides believe they hold the cards. Anthropic has tied its US$15 billion investment to copyright certainty, while the government is betting that market access on its own terms, with faster approvals as the sweetener, is leverage enough. It moved on power, and on copyright it declared a position without a mechanism, saying the law would spell out that creators keep ownership and control of their work while the Attorney-General's consultation runs on. The certainty Anthropic tied its investment to is still outstanding.