Australia’s AI Copyright Fight Is Turning Into a Test Case for the Data-Center Boom

Australia’s next big artificial-intelligence decision is not only about chatbots, authors, musicians or whether a machine can summarize a novel without paying the person who wrote it. It is about whether a government trying to attract a wave of AI infrastructure investment can hold a bright line on copyright while the companies building frontier models argue that the rules for training those systems are too tight.

That fight moved back to the center of Australia’s tech agenda this week as Prime Minister Anthony Albanese prepared to deliver a major speech on AI policy, with his government facing pressure from two directions at once: technology companies and data-center investors seeking more certainty for AI development, and creators warning that any weakening of copyright rules would amount to licensing their work after the fact — or not at all.

The immediate trigger is a renewed debate over a possible text-and-data-mining exception. In plain English, that kind of rule could allow AI developers to copy or analyze copyrighted material for model training under specified conditions, instead of negotiating permission work by work. Supporters describe it as the missing legal plumbing for an AI economy. Opponents describe it as a shortcut around the people whose books, songs, journalism, images and websites give AI models much of their value.

For Shadowfetch readers, the stakes are bigger than one country’s copyright statute. Australia is becoming a clean-room test for a question every AI market is wrestling with: can governments get the promised investment from AI companies without quietly socializing the costs onto artists, publishers, software communities and open-web creators?

The policy split behind the speech

Guardian Australia reported that Albanese is expected to use a Wednesday speech to outline the government’s broader AI direction, while a concrete copyright announcement is not expected in that address. That timing matters because the government has spent months trying to square two goals that do not naturally fit together: making Australia attractive for AI infrastructure and keeping its promise that copyright protections will not be weakened for model training.

The tension is visible inside Labor. According to Guardian Australia’s reporting, Industry Minister Tim Ayres and Assistant Minister for the Digital Economy Andrew Charlton are among the figures most eager to attract AI investment. Attorney General Michelle Rowland, who is responsible for copyright, and Arts Minister Tony Burke are described as more focused on protecting creators’ rights. The government has publicly rejected claims that it plans to grant a broad text-and-data-mining carve-out, but the fact that the idea keeps resurfacing tells its own story.

AI companies want predictable rules for where they can train and host models. Data-center developers want grid access, land, permits and a stable policy environment. Creative industries want licensing, transparency and enforcement. Those demands collide most sharply around the open web, where the material needed for model training is abundant, fragmented and often hard to license one owner at a time.

This is not a theoretical problem. The Productivity Commission’s final report on data and digital technology, released in December and tabled in Parliament in January, said Australian copyright law generally requires AI developers to obtain a license before using copyrighted material to train models. The commission also noted that much AI training happens overseas, where Australian copyright law does not apply, and that some Australian content is likely being used without permission or compensation in other jurisdictions while litigation continues.

That creates a policy trap. If Australia keeps its rules strict, AI developers may train elsewhere. If it loosens the rules too quickly, creators may see the government as rewarding the same scraping behavior they say already harmed them.

Why data centers are now part of the copyright fight

The new pressure point is the data-center boom. AI models need enormous computing capacity, and countries are competing to host the physical infrastructure behind that capacity: server farms, power connections, cooling systems and fiber routes. Australia has several assets that appeal to that market: political stability, available land, proximity to Asian markets and a growing renewable-energy buildout.

But data centers also bring hard questions. Who pays for new power generation and transmission? How much water and land will be used? Will the jobs and tax base justify the public costs? And, in this case, will AI companies try to attach copyright concessions to investment promises?

Guardian Australia reported that independent Sen. David Pocock’s office received a tip in late June about an industry push for a copyright carve-out tied to at least $50 billion in data-center investment and contributions to a fund for creatives reportedly worth $350 million a year. Pocock called the idea a “dirty deal” and urged the government to rule it out. The government rejected his account as inaccurate and repeated that it had no plans to weaken copyright laws.

Even if no such package is adopted, the episode shows how fast infrastructure policy and creator rights are being fused together. In older internet fights, copyright was often framed as a battle between entertainment companies and users. In the AI era, it is also a competition policy issue, a labor issue, an energy issue and a national industrial strategy issue.

The most important thing to understand is leverage. If frontier AI firms believe Australia is valuable as a data-center and training hub, Australia can set terms. If ministers believe investment will vanish unless the country moves closer to U.S.-style fair-use flexibility or European-style mining exceptions, the pressure shifts toward compromise. The whole debate turns on which side thinks it can walk away.

What the Productivity Commission actually recommended

The Productivity Commission did not recommend an immediate rewrite of copyright law. That is a key fact in a debate where both sides are tempted to speak as if the only choices are “open the gates” or “ban AI training.”

The commission said AI could contribute to productivity growth and that poorly designed regulation could slow adoption. It also said licensing markets are developing for high-value works, including news and books, but open-web material remains difficult to license because transaction costs are so high. In its final recommendation on copyright and AI, the commission urged the government to monitor the issue over three years, including licensing markets for open-web material, the impact of AI on creative incomes from copyright royalties and how overseas courts define the boundaries of AI-related copyright exceptions.

If those issues remain unresolved after three years, the commission said the government could establish an independent review of copyright settings and AI. That review could consider whether copyright remains a barrier to using open material in AI training and whether copyright is still the right mechanism to incentivize new works.

That is a cautious recommendation, not a green light for a carve-out. It reflects three uncertainties that are still unresolved globally.

First, courts have not settled how far AI training can go under existing exceptions in major jurisdictions. U.S. fair-use litigation is still shaping the legal border. Europe’s text-and-data-mining framework includes opt-out mechanisms and remains contested in practice.

Second, nobody has a clean answer for how AI training changes creative incomes over time. Licensing deals may help large publishers, labels and agencies. They may do less for independent writers, photographers, visual artists, coders and small outlets whose work is swept into training sets but who lack bargaining power.

Third, the market for licensing open-web material is still immature. If voluntary licensing becomes practical, governments may not need a broad exception. If it does not, pressure for statutory reform will keep building.

The creator argument: consent first, compensation second

The creative-sector objection is not only about money. It is also about consent and control.

Australian author Anna Funder, speaking at Parliament House this month, framed the use of her work by technology companies as a violation, not merely a missed royalty check. That language may sound blunt, but it captures the emotional center of the dispute. Creators are being asked to negotiate in a market where much of the copying may already have happened, where model developers often disclose little about training data and where outputs can compete with the same people whose work helped train the system.

A fund for creators, if one ever emerged, would not automatically solve that problem. Who qualifies? How is money distributed? Does payment cover past use, future use or both? Can a creator refuse? Would small creators receive meaningful compensation, or would the system mostly reward large rights aggregators?

Those questions matter because “the creative sector” is not one thing. A major media company can hire lawyers and negotiate licenses. A novelist, session musician, illustrator, YouTuber, local journalist or open-source maintainer may have very little practical power. If a policy treats all copyrighted work as a bulk input to national AI strategy, the benefits are likely to concentrate where bargaining power already exists.

That is why the text-and-data-mining label can be misleading. It sounds technical and neutral. In practice, it decides whose labor becomes infrastructure.

The tech argument: legal friction can move the industry offshore

The technology industry’s case is also not imaginary. Training large AI models requires vast datasets, and the open web is part of that supply chain. A country that makes training legally risky may see companies route the work elsewhere, even if it hosts data centers for inference, cloud services or regional operations.

Scott Farquhar, Atlassian co-founder and chair of the Tech Council of Australia, argued last year that Australia’s copyright settings were out of sync with places that provide fair-use or text-and-data-mining exceptions and that the gap could deter AI training and hosting. The Productivity Commission quoted that argument in its report, while also warning that copyright is only one factor. Electricity costs, skilled labor, infrastructure approvals and the broader investment climate also shape where AI firms build.

That caveat is important. A copyright carve-out by itself does not create an AI hub. It may make lawyers more comfortable, but it does not solve power constraints, grid queues, community opposition, water demand, hardware availability or the shortage of specialized engineers. If policymakers trade away creator protections for investment that would still depend on all those other variables, they may give up something concrete for something conditional.

The better tech-policy question is not “Should Australia be pro-AI or anti-AI?” It is: what terms should a democratic government set before giving frontier AI companies the infrastructure, legal certainty and social license they want?

Why this story travels beyond Australia

Australia’s fight lands at the same time that AI policy is fragmenting around the world. The European Union has moved toward formal AI regulation while still wrestling with copyright opt-outs and enforcement. The United States relies heavily on litigation and fair-use arguments. The United Kingdom has repeatedly considered text-and-data-mining reforms and faced pushback from publishers and artists. Smaller markets are watching to see whether strict rules protect local creators or simply push model training offshore.

That makes Australia’s decision unusually consequential. It is not the largest AI market, but it is large enough, wealthy enough and strategically placed enough to test whether a government can bargain with the AI sector instead of simply adapting to it.

If Australia holds the line on copyright and still attracts major AI infrastructure investment, creators elsewhere will point to it as proof that governments do not have to weaken rights to win data centers. If it softens the rules under investment pressure, tech companies will have a fresh example to cite in other capitals: adjust copyright, unlock capital, move faster.

Neither outcome will settle the global fight. But it will change the negotiating script.

The practical path forward

The cleanest near-term move for the Albanese government would be to separate three debates that lobbyists have every incentive to bundle together.

The first is AI infrastructure: data centers, energy, permitting, grid costs and local benefits. Those projects should be judged on power, water, jobs, resilience, security and whether developers pay their share of the systems they use.

The second is AI safety and accountability: where existing laws are enough, where regulators need more resources and where technology-specific rules are justified. The Productivity Commission favors using existing legal frameworks where possible and adding AI-specific rules only as a last resort.

The third is copyright: whether model training should require permission, licensing, collective compensation, a narrow exception, or some combination. That debate should not be decided as a side letter to a data-center pitch.

For readers outside Australia, that separation may sound procedural. It is not. Bundling is how public-interest questions get converted into urgency theater. If copyright becomes the price of “not missing the AI boom,” then creators start the negotiation already behind. If data centers are assessed without the copyright sweetener, governments can ask a simpler question: is this infrastructure good on its own terms?

The coming speech may not answer the copyright question. But it will signal how Albanese wants Australia to position itself in the AI economy: as a rule-setter, a host site, a licensing market, or some uneasy blend of all three.

The real test is whether “AI opportunity” becomes a reason to modernize policy carefully, or a phrase used to rush past the people whose work made the systems useful in the first place.

Sources

• Guardian Australia: “AI companies want to water down Australia’s copyright laws. Artists are outraged, Labor is split.”
• Productivity Commission: “Harnessing data and digital technology,” final inquiry report, released Dec. 19, 2025, tabled Jan. 29, 2026.
• Productivity Commission: Recommendation 2.1, “A review of Australian copyright settings and the impact of AI.”
• Public statements reported by Guardian Australia from Prime Minister Anthony Albanese, Sen. David Pocock, Scott Farquhar, Ed Husic, Andrew Charlton, Sarah Hanson-Young and Data Centres Australia.