Anthropic Secures Huge Deal For Australia AI Data Centre, Aims To Build One Of The Largest Computing Facilities In The World

Anthropic, an artificial intelligence company, has secured a significant data centre lease agreement in Australia. This deal grants Anthropic access to a substantial computing facility, potentially among the world's largest, leveraging Australia's favorable political climate and abundant renewable energy resources. The lease covers a data centre campus with an anticipated capacity of 2.16 gigawatts, comparable to some of the largest facilities globally.

The proposed site is situated on farmland approximately 250 kilometers from Brisbane. Construction is expected to commence in 2027. This agreement marks Anthropic's latest presence in Australia, following a similar off-take deal OpenAI reached for a Sydney data centre last year. The country has become a magnet for AI companies, who are vying for power and land in a nation known for its renewable energy and lower costs compared to congested markets in the United States and Asia.

Anthropic CEO Dario Amodei has cautioned against the rapid development of AI, urging companies to slow the pace of model advancements due to growing concerns about potential misuse of artificial intelligence. This sentiment was echoed by Elon Musk of xAI and OpenAI CEO Sam Altman. However, US President Donald Trump dismissed such concerns as exaggerated, labeling critics as negative voices.

Australia's government has faced community opposition over the resource demands of data centres, including land, water, and electricity usage. Authorities have announced restrictions on data centre resource use starting in 2027, while also encouraging investment in the sector. Economists estimate that the AI industry could contribute A$150 billion to Australia's economy by 2030.

The Anthropic lease is tied to a data centre planned by Singapore-based developer Zerra DC, which aims to power the facility through renewable energy agreements and cover grid connection costs independently. The facility will focus on inference workloads rather than training and will employ a closed-loop, air-cooled system to minimize water usage for server cooling.