What AI and copyright means for Australian data centre operators and AI infrastructure investment

This pairing is deliberate: data centres provide the physical infrastructure on which AI models are trained and deployed, while copyright governs the content that flows through that infrastructure.

The regulatory settings on both directly affect the commercial viability of AI operations in Australia, and for data centre operators, investors, and co-location providers, the copyright framework the government adopts is an important factor alongside energy pricing, land availability, and network connectivity.

On 15 July, the federal government announced what it describes as the world's first unified national AI framework, establishing a new national ‘office of AI’ within the Department of the Prime Minister and Cabinet and introducing mandatory Australian standards for AI.

The announcement signalled the government will “ensure the strongest possible protection for Australian artists and media”, ruling out any text and data mining (TDM) exception to the Copyright Act 1968 (Cth).

“The future of AI development in Australia must respect creator rights”, prime minister Anthony Albanese declared, reaffirming the government’s position that Australia has now held firmly for almost a year: that no company may use Australian books, music, art or journalism to train AI without the rights holder's explicit consent and control over the price and value of their work.

Why copyright matters for AI

Large language models (LLMs) and generative AI systems depend on vast quantities of human-created content as their primary input. The quality and cultural specificity of training data directly determines the quality of the AI system.

Where AI training physically occurs affects which copyright regime applies. This shapes demand for Australian data centre facilities relative to competing jurisdictions.

The process of ingesting copyrighted works to train AI models involves the reproduction and adaptation of those works, which are acts that, absent a statutory exception or licence, are capable of constituting copyright infringement.

Unlike the US, Australia has no broad fair use doctrine. Its “fair dealing” defences are confined to prescribed purposes, such as research or study, criticism or review, parody or satire, and news reporting, and are generally regarded as too narrow to accommodate large-scale, commercial AI training.

Australia has also declined to introduce a text and data mining exception of the kind adopted in the EU and proposed in the UK. The absence of either a general fair use defence or a dedicated TDM exception means that AI developers training large models on Australian soil face material legal uncertainty as to whether their ingestion of copyrighted content is lawful without express rights holder consent.

There is an underlying tension between attracting AI infrastructure investment which is beneficial for the economy and ensuring there are adequate mechanisms for Australian copyright holders to be compensated for their works. The government maintains that its copyright framework and AI investment objectives are compatible.

Three potential licensing models

With a TDM exception off the table, the government is looking to the work of the Copyright and Artificial Intelligence Reference Group (CAIRG) and is exploring paid collective licensing and voluntary licensing frameworks with opt-in and opt-out avenues for rights holders.

There are three licensing models currently under consideration.

Statutory licensing

This approach could involve a compulsory licence under the Copyright Act where rights holders cannot opt out and remuneration is agreed or determined by the Copyright Tribunal. This provides cost certainty for AI tenants but contradicts the government's stated commitment to rights holder control.

Collective society licensing

Collecting societies, including copyright management organisations the Australasian Performing Right Association (APRA) and Australasian Mechanical Copyright Owners Society (AMCOS), could grant licences on behalf of members with an opt-out mechanism. An extended collective licensing variant would require legislative amendment. This would introduce some cost variability for data centre operators.

Voluntary direct licensing

This could involve bilateral direct negotiations between copyright owners and AI companies. These are already operating to some degree through content licencing agreements, such as those brokered between Nine and Microsoft and OpenAI and News Corp Australia. However, in our view the voluntary direct licensing approach creates the greatest cost uncertainty for AI training in Australia as it risks challenges in identifying rights holders and risks protracted negotiations.

Global position

Australia's position carries both policy and commercial significance. The licensing framework that the government ultimately adopts will be a material factor in whether Australia becomes a competitive destination for large-scale AI compute.

The government is still considering feedback received from CAIRG participants on these issues. Meanwhile, further details of the AI standards framework are due to be presented to the National Cabinet in August. Legislation is expected to be introduced to parliament in early 2027.

Australia’s position may also be influenced by other countries stance on these issues. In the UK, the Intellectual Property Office has opened a consultation on AI and copyright in December 2024 examining TDM options. The UK government decided earlier this year not to extend the TDM exception in copyright law to facilitate AI training.

The UK approach differs in some respects to that taken by the EU, where policymakers are undertaking a formal statutory review of Directive (EU) 2019/790 on Copyright in the Digital Single Market, including the European Parliament's proposed licensing fee; and the European Commission’s parallel regulatory framework for data centre infrastructure.

The next phase of Australia’s AI and copyright journey will involve translating the government's commitment to consent and payment into a workable licensing framework. How that framework is designed will directly influence future demand for Australian data centre capacity.