{"id":12465,"date":"2026-08-19T17:14:42","date_gmt":"2026-08-19T06:14:42","guid":{"rendered":"https:\/\/insight.thomsonreuters.com.au\/legal\/?p=12465"},"modified":"2026-08-19T17:14:44","modified_gmt":"2026-08-19T06:14:44","slug":"on-a-collision-course-data-centres-and-copyright-in-australia","status":"publish","type":"post","link":"https:\/\/insight.thomsonreuters.com.au\/legal\/posts\/on-a-collision-course-data-centres-and-copyright-in-australia","title":{"rendered":"On a collision course? Data centres and copyright in Australia."},"content":{"rendered":"\n

Data is not stored in \u2018the cloud\u2019. It’s stored in massive physical data centres around the world \u2013 more than 150 are already operational in Australia, consuming limited physical resources such as energy and water, somewhat controversially.

What’s less widely understood is that some of these data centres are also used for training generative AI models \u2013 for example, training large language models on scraped content, potentially including the work of individual Australian creators.

The threat to Australian content trained by these models on our soil triggers a copyright issue. In the 2026 Jill McKeough Lecture, \u201cCan Copyright Survive Generative AI, (Westlaw subscribers can access here<\/a>), delivered earlier this year at the University of Technology Sydney, David Lindsay \u2013 Adjunct Fellow at UTS Law School and, until his recent retirement, general editor of the Australian Intellectual Property Journal since 2012 \u2013 highlights the great uncertainty around applying existing copyright law and policy to generative AI, whether in relation to data centres here or offshore.<\/p>\n\n\n\n

<\/p>\n\n\n\n

The law is territorial <\/h2>\n\n\n\n

“This is clearly one of the most important (and contentious) issues \u2013 how does copyright law adjust to the automated production of content using generative AI systems like ChatGPT? Generative AI is obviously already there in terms of uptake of use, but law and policy continue to lag.”

Copyright law is territorial, Lindsay explains. If Australian works are reproduced within Australia, such as in data centres based in Australia, Australian copyright law applies and infringement actions can be brought. But if training happens overseas \u2013 say, in US data centres \u2013 it’s very difficult for Australian creators to assert their rights, because for a claim to succeed, the infringement must occur within the relevant copyright territory.

If a model trained overseas is later imported into Australia, creators face a difficult task \u2013 and that’s exactly the scenario tested, unsuccessfully, in the UK’s Getty Images case. In Getty Images (US) Inc v Stability AI Ltd<\/em> [2025] EWHC 2863 (Ch), decided on 4 November 2025, Getty accepted there was no evidence its images had been used to train Stable Diffusion in the UK, so its primary infringement claim was dropped. That left only a claim of ‘indirect infringement’ \u2013 importing an infringing product into the jurisdiction.

That claim failed too. The Court found that Stable Diffusion’s “foundation model” \u2013 its parameters and weights \u2013 did not contain a copy of the original images, even though similar-looking images could be produced from it. So, at least on current UK authority, the scope for creators to sue for overseas training is doubtful.

Interestingly, last year a German court reached the opposite conclusion on similar facts. In GEMA v OpenAI<\/em>, the Munich Regional Court ruled in November 2025 that memorising copyright-protected content within a language model’s parameters does infringe reproduction rights \u2013 even though the content exists only as probability values distributed across the model. David believes the German court “stretched things”, but the divergence between London and Munich shows just how unsettled this area remains internationally.

In the US, where much AI training currently takes place, David says there are “special considerations”. The administration is, in his words, pro-big tech, in a country that has effectively “bet the farm” on AI investment, which is keeping markets afloat (at least for now). Federal regulators have largely taken “all the guardrails off”, and the US “fair use” defence gives US-based AI developers real scope to argue their training use of copyright works is lawful \u2013 though it will take some time for case law to settle the question.<\/p>\n\n\n\n

The Australian Prime Minister’s announcement – a game changer?<\/h2>\n\n\n\n

In his 15 July 2026 speech “AI in Australia’s interests” <\/a>at the University of Sydney, the Prime Minister outlined in broad terms how AI and data centres in Australia will be regulated. David stresses this actually bundles together three separate issues that shouldn’t be conflated:<\/p>\n\n\n\n

    \n
  1. Data centre location and resource use<\/strong> \u2013 governed by new mandatory “Australian Standards” replacing today’s voluntary guidelines, with legislation targeted for early 2027. It appears that data centres in Australia will be required to be “net-generators, not net-users” of energy.<\/li>\n\n\n\n
  2. Broader AI ‘guardrails’<\/strong> \u2013 including safety and consumer-protection regulation which, while the Australian government has announced some broad AI consumer safety priorities (including a proposed digital duty of care and further consultation on privacy law reforms), David says remains genuinely unclear from the speech.<\/li>\n\n\n\n
  3. Copyright and creators’ rights<\/strong> \u2013 the strand most relevant to IP lawyers, and one of the Prime Minister’s most important policy commitments: “Nowhere do artists or rights holders have sufficient control of their work, when it comes to AI training\u2026 No company should use Australian books, music, art or news to build or train AI without the artist’s control.”<\/li>\n<\/ol>\n\n\n\n

    The commitment to protect creators and copyright owners is heartening and has generally been well-received \u2013 but David cautions the speech was “an outline of the policy”, not a policy. “We don’t have a clear idea as to the protections. Much of that work still has to be done.”

    As far as future regulation is concerned, a new Office of AI within the Department of the Prime Minister and Cabinet is intended to coordinate work on AI policy across the Attorney-General (copyright), the Treasurer (productivity), the Employment Minister, the Education Minister and others. While a     joint ministerial media statement<\/a> was released on 20 July 2026 outlining a series of high-level AI consumer safety priorities, this does not affect the copyright issues, which fall within the Attorney-General\u2019s portfolio.

    According to David, two big questions remain: first, as mentioned previously, whether protecting Australian creators when training in offshore data centres requires legislative amendment; second, what other reforms might be needed to protect creators and copyright owners.<\/p>\n\n\n\n

    Why the market alone won’t fix this. <\/h2>\n\n\n\n

    The government’s stated preference is for creators and AI developers to reach voluntary licensing deals. Large media companies can often negotiate such deals, and many have been struck, but David warns the imbalance of bargaining power between big AI developers and, say, individual authors makes this unreliable more broadly. “It is unlikely that the market will be able to work for all copyright owners and for all authors and creators unaided, without any government intervention,” he says.

    If the market fails to find a solution, David identifies three intervention options, which have been canvassed in the academic literature:<\/p>\n\n\n\n