An AI Lab Wants to Double Australia's Data Centre Capacity

Anthropic — the US-based AI company behind the Claude model — is in advanced discussions to secure 1.4 gigawatts of data centre capacity across Australia. To put that in context: 1.4 gigawatts is roughly equal to the entire existing Australian data centre industry's capacity at the end of 2025. One company. Matching the entire industry.

The investment is estimated at up to $21.6 billion, according to reporting based on government briefing documents obtained by the media. CEO Dario Amodei met with Treasurer Jim Chalmers in April to discuss Anthropic's Australian market entry, which includes making this country its secondary AI model training hub — the first significant training facility outside the United States. Anthropic wants at least one gigawatt operational by mid-2027, with CDC Data Centres reportedly in line for the largest share at around 500 megawatts and the remainder split across four or five providers.

There is a catch. Government briefing notes, cited across multiple outlets, state that Anthropic's investment in 'AI model development capability and associated infrastructure, like data centres, is contingent on clarity of copyright settings.'

Anthropic wants to train its AI models on copyrighted content — books, articles, images, music. In the US, the company argues this falls under 'fair use.' Australia does not have a broad fair use exception. Australian law requires AI companies to obtain permission from copyright holders through voluntary licences, and the government told Anthropic it would not introduce a text and data mining exception to change that.

On 15 July, PM Albanese was direct: 'No company should use Australian books, music, art or news to build or train AI without the artist's control. Anything less is theft.' We covered the full scope of Australia's new AI standards commitment that same week. Anthropic cited the 'long tail' of smaller rights holders — independent photographers, musicians, writers — as a practical barrier to negotiating licences at scale. The government's position: that is a cost of doing business here, not a reason to rewrite the law.

Data centres need electricians, HVAC technicians, and plumbers — in large numbers, for years. A 1.4-gigawatt buildout is not a single site. It is a portfolio of facilities requiring continuous construction, fit-out, and maintenance across multiple states.

We wrote in April about how the existing $52 billion data centre pipeline will squeeze every tradie. That analysis covered 15 NSW projects endorsed by the Investment Delivery Authority and a national electrician shortage of 53,000 to 84,000 by 2030. Anthropic's bid sits on top of that pipeline. If it proceeds, the squeeze intensifies — more work chasing the same pool of technicians with a 54 per cent vacancy fill rate.

The federal government's National Expectations for Data Centres require operators to invest in apprenticeships and trades training as a condition of fast-tracked approvals. Anthropic's facilities would fall under these rules. For trades business owners, this is an operations throughput signal: the demand is building faster than the workforce can match it.

But the 'if' matters. If copyright negotiations stall and Anthropic redirects the investment to Singapore, Indonesia, or the Middle East — all competing for hyperscale AI compute capacity — the Australian pipeline loses its single largest prospect.

For accounting and law firms, the relevant question is about the tools themselves. If Anthropic builds training infrastructure in Australia, the models it produces gain deeper access to Australian content — legislation, case law, accounting standards, regulatory frameworks. An AI trained with genuine exposure to Australian professional content is more useful than one calibrated primarily for US and UK markets.

Australia's stance also sends a signal about regulatory seriousness. The PM's commitment to legislated AI standards, combined with the Privacy Act disclosure deadline in December and the copyright protection framework, means AI tools and companies operating here will face higher compliance standards than in most other markets. For a professional services firm handling client data, that regulatory floor is not a burden. It is a quality filter — the tools that meet it are the ones worth trusting.

If you run a trades business, the data centre construction pipeline is real and growing regardless of whether Anthropic's specific bid proceeds. Microsoft, Google, and multiple domestic operators are building. Position your capacity — whether that means upskilling team members in data centre-grade electrical or HVAC work, or ensuring your scheduling and throughput can absorb larger commercial contracts.

If you run a professional services firm, pay attention to which AI tools are being trained on Australian content and which are not. As AI becomes embedded in practice management, document review, and client advisory, the quality of the model's Australian knowledge base becomes a competitive factor. Tools built on Australian data, under Australian rules, will outperform generic global models for Australian-specific work.

Key takeaways

Related notes

Wondering how the AI infrastructure buildout affects your business?

The data centre pipeline is creating demand, and the policy landscape is shaping what AI tools will look like in Australia. A 30-minute call can map where your business sits in both — and what's worth preparing for. Book a call.