Australia just became the surprise frontier in the AI infrastructure land grab—and it's happening 1,500 kilometers from Sydney in cattle country.
The Summary
- Anthropic will build a $31bn datacentre in western Queensland, marking the Claude maker's first Australian facility and one of the largest AI infrastructure investments outside the US
- Queensland Premier David Crisafulli framed it as a "major win" for jobs and grid energy capacity
- The location choice—remote western Queensland instead of coastal tech hubs—signals a new phase in global AI infrastructure strategy where power availability matters more than proximity to talent
The Signal
Anthropic isn't building in Sydney or Melbourne. They're going inland, to a region better known for mining and agriculture than machine learning. The $31bn datacentre deal in western Queensland represents the largest single AI infrastructure commitment in Australia's history, and the geography tells you everything about where this industry is heading.
The pattern is clear: AI companies are following power, not people. Western Queensland has abundant solar potential, lower land costs, and critically, grid capacity that coastal cities can't match. When you're training models that consume megawatts per rack, you build where the electrons are cheap and plentiful. Talent can work remote. Transformers cannot.
"When you're training models that consume megawatts per rack, you build where the electrons are cheap and plentiful."
This is Anthropic's first datacentre outside North America. That they chose Australia—and specifically rural Australia—over Europe or East Asia says something about regulatory calculus and energy economics. Australia has stable governance, English common law, and crucially, no data localization requirements that would complicate a global training infrastructure. The EU's AI Act looks straightforward until you try to physically comply with it across 27 jurisdictions.
The $31bn price tag deserves scrutiny. For context, that's roughly what Meta spent on all datacentre construction globally in 2024. Either Anthropic is building something truly massive—speculation suggests 500+ megawatts of compute capacity—or they're bundling long-term energy purchase agreements into the headline number. Queensland's premier mentioned the deal would "put more energy into the grid," which suggests Anthropic might be co-investing in generation capacity, not just consuming it.
Key implications for the AI infrastructure race:
- Geographic arbitrage is real: Remote locations with power and cooling beat cities with universities
- Australia is positioning as the "AI Switzerland"—stable, neutral, energy-rich
- Model training is becoming a heavy industry, with site selection closer to aluminum smelting than software engineering
The jobs angle is interesting but probably overstated. Modern datacentres are remarkably lights-out operations. Construction will employ thousands temporarily. Long-term operations might be a few hundred highly specialized roles. The real economic impact is indirect: grid investment, renewable energy buildout, and whatever second-order effects come from having frontier AI infrastructure on Australian soil.
The Implication
Watch where other AI labs build next. If Anthropic is right about western Queensland, you'll see OpenAI, Google, and others scouting similar regions—places with sun, wind, land, and governments hungry enough for investment to move fast on permits. The competition for AI infrastructure sites is about to look a lot like the competition for semiconductor fabs: strategic, geopolitical, and billions deep.
For anyone in energy, construction, or regional development: the AI buildout is the biggest infrastructure story of the next decade. These aren't just datacentres. They're the foundries where intelligence gets manufactured at scale.