Anthropic’s Queensland Lease Puts AI Power Demand on Display

Anthropic has signed its first Australian data-center lease, securing capacity at a planned Queensland campus that could reach 2.16 gigawatts, according to two people familiar with the deal cited by Reuters. The proposed site is about 250 kilometers from Brisbane, with operations expected to begin coming online in 2027. (finance.yahoo.com)

The agreement gives Anthropic a large foothold in a country increasingly marketed as an alternative to crowded U.S. and Asian markets. Australia offers substantial renewable-energy potential, available land and a government eager to attract artificial-intelligence infrastructure, but the project still requires approval from the Foreign Investment Review Board. (finance.yahoo.com)

What changed

The deal converts Anthropic’s broader Australian infrastructure ambitions into a named commercial commitment. The company had previously signaled interest in expanding local computing and energy capacity, including through an April memorandum of understanding with the Australian government. That agreement said Anthropic was exploring data-center and energy investments aligned with Australia’s national AI and infrastructure expectations. (minister.industry.gov.au)

Queensland officials and project reporting describe the development as part of a proposed data-center hub on the Western Downs. ABC reported that the wider project could cost about A$32 billion and consume power comparable to roughly 1.5 million Australian households. Those figures describe the proposed campus, not necessarily Anthropic’s own financial commitment or eventual electricity use. (abc.net.au)

Neither the lease’s financial terms nor its duration has been disclosed. The 2.16-gigawatt figure is also a planned campus capacity, not proof that all of that power will be available to Anthropic immediately. Construction schedules, grid connections, financing and regulatory approvals remain unresolved.

Why it matters

The announcement illustrates how AI competition is shifting from model releases toward long-term control of electricity, land and specialized computing capacity. A lease at this scale can help Anthropic reduce exposure to constrained U.S. power markets and support future inference demand, where models serve users continuously rather than only during training runs.

It also sharpens the contradiction at the center of the AI buildout. Anthropic chief executive Dario Amodei has called for a slower pace of capability development because of misuse and safety concerns, yet the company is simultaneously reserving infrastructure on a scale associated with aggressive expansion. The lease does not establish that Anthropic will use the full campus, but it signals that the company expects demand for advanced-model compute to remain structurally high.

For Australia, the project offers investment, construction activity and possible digital-sector growth. It also creates pressure to demonstrate that new demand will not raise household power costs, strain transmission networks or intensify competition for water and land. The approval process will therefore test whether Australia can attract frontier-AI infrastructure while enforcing conditions that protect local communities and the grid.

The immediate uncertainty is execution. Until foreign-investment approval, financing, grid arrangements and construction milestones are confirmed, the project remains a major planned commitment rather than an operating facility.