The global race for artificial intelligence infrastructure has found a massive new foothold in regional Australia.
Anthropic has committed to a staggering $32 billion data centre project in Queensland’s Western Downs, specifically designed to power its flagship
Claude AI assistant.
The development application, filed last month by Singapore-based operator Zerra DC, outlines a sprawling 725-hectare campus near Dalby that will focus exclusively on running inference tasks, meaning it will handle live user queries rather than train new foundational models.
The Scale of the Compute Ambition
Construction is projected to span four to six years, with Anthropic aiming to bring the facility online by 2027. At its four-phase peak, the campus will demand a staggering 2.16 gigawatts of power. To put that into perspective, this single facility will draw roughly 47 gigawatt-hours daily, equivalent to the energy consumption of 1.5 million average Australian households.
Local officials have framed the project as a significant economic catalyst. Western Downs Mayor Andrew Smith indicated that early discussions point to the creation of approximately 1,000 local jobs. Queensland’s leadership has also highlighted the potential for broader grid investment, though the sheer scale of the power draw requires careful navigation of the state’s existing
energy roadmap [[12]].
| Metric Category | Traditional Enterprise Data Centre | Proposed AI Inference Mega-Campus | Primary Grid Impact |
| Peak Power Draw | 10 to 50 Megawatts | 2,160 Megawatts (2.16 Gigawatts) | Requires dedicated high-voltage substation upgrades |
| Daily Energy Use | ~0.5 Gigawatt-hours | ~47 Gigawatt-hours | Strains regional baseload generation capacity |
| Primary Workload | General cloud hosting and storage | Continuous large language model inference | Sustained, non-fluctuating 24/7 power demand |
| Cooling Requirements | Standard air or basic liquid cooling | Advanced direct-to-chip liquid cooling systems | Massive localized water and thermal management needs |
The Energy and Environmental Friction
Despite the economic promises, the project faces intense scrutiny over its environmental footprint. The proposed site sits near three state-owned coal-fired power stations that share the local substation, meaning the facility will rely heavily on fossil fuels in the near term. Independent analysis estimates this setup could generate up to 6.6 million tonnes of carbon emissions annually.
This reality has sparked fierce local resistance. Community groups like Save Our Darling Downs have voiced deep concerns over the proposed energy consumption, backed by a local petition gathering over 1,300 signatures. The
Climate Council has also warned that unchecked data centre growth risks driving up power prices and increasing pollution if developers do not bring new, additional clean energy to the grid.
The deal remains contingent on Foreign Investment Review Board approval and local council planning clearance. As hyperscalers continue to hunt for land and power, this Queensland project serves as a stark microcosm of a global dilemma. The world wants smarter AI, but the physical infrastructure required to run it is rapidly colliding with the limits of legacy power grids and community tolerance.