Senate Inquiry into AI and Data Centres

Background

On 13 May 2026, the Senate referred an inquiry into artificial intelligence and data centres to the Environment and Communications References Committee. The terms of reference cover the effectiveness of existing regulatory frameworks in managing data centre growth, including deals between the Government and global AI companies, and the potential impacts of AI and data centres on Australian communities, industries, the environment, water and energy. The committee is due to report by 16 November 2026.

Our submission

The submission, Hosting AI Infrastructure Offers Australia More Than Economic Benefits, argues that hosting AI data centres, done right, could offer Australia more than economic investment. Hosting inference compute would reduce our dependence on foreign infrastructure and could help our regional partners do the same. Hosting training compute would give Australia a say over a technology set to reshape our economy and security.

Its starting point is that AI is not normal technology. AI is approaching or surpassing human performance in domains including scientific reasoning, coding, cybersecurity and persuasion, and the International AI Safety Report places its risks in three categories: deliberate misuse, malfunction, and broader harms to society. AI developed and operated offshore offers Australia no protection from these risks. It leaves us more exposed.

On inference compute, our best estimate is that Australia's supply meets less than half of domestic demand, and Australia hosts about 0.4% of global compute supply. Even counting announced data centres, we assess Australia will meet less than a quarter of its domestic inference demand by the end of 2027. Falling short creates a new foreign dependency: countries that control our inference compute could drive up costs, remove access, or exploit that access to create vulnerabilities. Hosting enough compute to be a net exporter would keep sensitive workloads under Australian jurisdiction and position Australia to serve regional partners who cannot build sovereign infrastructure of their own.

On training compute, few countries will host frontier AI training, and doing so would strengthen Australia's position to negotiate terms in the national interest. Australia's Position in the Age of AI, our recent report, assessed hosting frontier AI training as the highest value action Australia could take to establish its place in an AI future. Conditions Australia could seek include ongoing access to frontier AI models (public, defender, and government access), compute for Australian researchers on favourable terms, and oversight of AI development — pre-deployment testing by the Australian AI Safety Institute, disclosure of model capabilities and risks, and reporting of critical safety incidents.

On energy and water, the submission acknowledges real risks and downsides, and argues policy should be based on accurate data. Some widely cited figures, particularly on water use, are overstated or have since been corrected. Per-prompt water estimates have been revised from a bottle of water to millilitres, and a bestselling book's data centre water figure was overstated by a factor of approximately 1,000. Legislated data centre standards should still impose real conditions: additional renewable supply, coverage of grid-upgrade costs, efficient cooling, and appropriate water sourcing. The Government should not rely on voluntary compliance.

Realising the benefits of hosting AI training requires a copyright solution. Current Australian copyright law does not allow frontier AI training to occur here, and neither AI companies nor rights holders have the incentives to strike the licensing deals that would change that. The submission recommends the Government progress an Australian approach that supports rights holders while allowing training to occur in Australia. Our consultation paper on an Australian AI Training Scheme sets out one potential model, and is included as an attachment along with a summary of consultation feedback so far. The submission also recommends the Government clarify whether AI inference falls within section 43B of the Copyright Act 1968, rather than leaving the question to litigation.

The submission makes six recommendations:

1. Ensure Australia can meet its own inference demand and serve the region, including by setting a target for Australia to become a net exporter of inference compute.
2. Attract AI training to Australia and apply national interest conditions.
3. Establish a compute verification workstream in the Australian AI Safety Institute.
4. Ensure the new data centre standards impose sensible, evidence-based requirements on energy and water use.
5. Develop copyright pathways that enable AI training and inference to occur in Australia while supporting creators.
6. Clarify the position of inference under Australian law and ensure copyright law is fit for purpose.