Anthropic Anchors A$32 Billion Queensland AI Data Centre

Key Takeaways

• Anthropic has agreed to become the anchor tenant at Zerra DC’s proposed Western Downs Digital Park near Dalby.
• The campus is designed for up to 2.16 GW of power and would primarily support Claude inference workloads.
• Local approval, electricity generation, water protection, and community consultation remain central issues before construction advances.

Anthropic has secured space for its first Australian data centre deployment, signing an agreement covering the initial stage of Zerra DC’s proposed Western Downs Digital Park in regional Queensland.

The Queensland premier announced the agreement in parliament on Wednesday, September 16. The premier described Anthropic’s commitment as a show of confidence in Queensland and linked the investment to regional employment and additional electricity generation.

The proposed A$32 billion campus would be built near Dalby, north-west of Toowoomba. According to ABC, planning documents indicate that the development would become Australia’s largest data centre if completed as proposed.

Singapore-based Zerra DC is developing the project, while Anthropic is expected to serve as its anchor tenant. The first stage is planned to open in 2027. That schedule remains conditional, however, because the regional council has not yet approved the development application.

The Western Downs mayor stated the council would undertake a thorough and comprehensive assessment process. Its timing will depend on referral requirements, requests for additional information, and community participation during public notification.

Scale is the defining feature of the project. The Anthropic portion would occupy about 725 hectares, including land currently used partly as a feedlot. At full buildout, the campus is designed for power demand of up to 2.16 GW, comparable to the electricity consumption of approximately 1.5 million Australian households.

Integrating an AI campus with metropolitan-scale electricity requirements presents distinct infrastructure challenges for the regional grid.

The premier noted that data centre developments should add generation to Queensland’s grid, place downward pressure on electricity prices, protect local water supplies, and respect host communities. The Western Downs campus would connect to the Braemar power station, and Queensland has retained the option of using coal generation to supply new AI facilities after rejecting a Commonwealth push in July for renewable-energy requirements.

Inference workloads are just as consequential for infrastructure planning as model training. The Queensland facility is intended to run high-volume inference for Claude, meaning the computing work required when users submit prompts and receive answers. It is not slated to train Anthropic’s underlying AI models.

As Claude adoption grows, inference demand expands with every additional user, software integration, and enterprise workflow. This creates a need for persistent, geographically distributed capacity rather than occasional bursts of training compute.

The project reflects a much wider shift in global infrastructure. A Gartner forecast reported by Tom’s Hardware projects that global data centre electricity consumption will reach 565 TWh in 2026, up 26% from 2025. AI-optimized servers already account for about 31% of that electricity use, and Gartner expects total demand to double by 2030, driven largely by AI workloads (source).

Microsoft, Google, and Amazon Web Services are pursuing similar investments in hyperscale facilities, dedicated AI clusters, and energy supply. The resulting competition encompasses grid connections, planning approvals, cooling systems, land, and construction capacity, all of which now dictate where AI services can expand.

Water constraints also influence facility engineering. Zerra DC’s proposed design emphasizes air cooling to reduce pressure on local water supplies. That approach supports operations in resource-constrained regions, although operators must still demonstrate performance under high temperatures and dense computing loads. Standards including ISO/IEC 30134 and the ASHRAE Thermal Guidelines for Data Processing Environments provide established measures for energy efficiency, cooling, and operating conditions.

A recent Report AI infrastructure assessment highlights how access to electricity and compute capacity has become a limiting factor in AI deployment. The Western Downs project exemplifies this dynamic, pairing substantial commercial opportunity with rigorous physical requirements.

Regulatory and policy considerations factor heavily into the rollout. Anthropic signed a Memorandum of Understanding with the Commonwealth government in April, while federal lawmakers are considering whether AI developers should receive default access to Australian creative works. An independent senator has argued that such an arrangement would disadvantage creatives in pursuit of data centre funding.

Meanwhile, the Anthropic chief executive recently repeated a call for slower AI development amid growing safety concerns. The organization is urging caution around AI’s trajectory while committing to infrastructure capable of serving Claude at a much greater scale. For Western Downs, the next decisive step is whether Zerra DC can secure the necessary local approval, power connections, and community confidence for a campus unlike anything Australia has built before.