Anthropic spent $15 billion to "enclose land" in Australia for 1.4GW of computing power, as the AI infrastructure arms race escapes US domestic bottlenecks.

When data center construction in the U.S. is mired in power shortages and community resistance, AI giants are turning their gaze overseas in search of new computing oases. Anthropic's latest disclosed overseas infrastructure plan marks that the computing power arms race among large model companies has officially crossed national borders.

According to a report in the Australian Financial Review on Sunday, Anthropic plans to secure at least 1.4GW of data center computing resources in Australia, with an estimated total investment cost of up to $15 billion. According to a confidential tender document, the company aims to have at least 1GW of that computing power operational by the end of next year.

This massive capital expenditure, far exceeding market expectations, is not only intended to support the exponential growth of its large model training and inference demands, but also reflects the industry pain point of constrained data center expansion in the U.S. At a critical juncture of surging revenue and an impending IPO, Anthropic is reinforcing its computing moat through global physical "land grabbing."

$15 Billion "Thrown" at the Southern Hemisphere, 1GW Computing Power Delivered Next Year

Anthropic's computing power deployment in Australia demonstrates strong execution and a clear timeline. The 1.4GW scale is equivalent to the output of several large nuclear power plants, enough to support the full-load operation of hundreds of thousands of top-tier AI accelerator cards. The $15 billion construction cost also makes it one of the world's largest single AI infrastructure investments in recent years.

The tender documents show the project adopts an aggressive delivery pace, requiring at least 1GW of computing power to be operational by the end of next year. To support this strategy, Anthropic has recently been aggressively recruiting data center engineers, electrical engineers, and project procurement specialists in Australia and Japan, as its computational department's physical infrastructure team accelerates its expansion into the Asia-Pacific region.

Escaping the Local "NIMBY Effect," Australia Becomes New Computing Oasis

Behind Anthropic's land grab in the Southern Hemisphere is the increasingly severe bottleneck in U.S. domestic data center construction. Just recently, Blackstone's QTS announced the termination of the "igitalGateway" project in Virginia, once touted as the "world's largest" at 850 hectares, due to ongoing local opposition and legal lawsuits.

With the explosion in AI computing demand, traditional U.S. data center hubs are facing multiple pressures from surging electricity needs, water consumption, and the community "NIMBY effect." In contrast, Australia, with its relatively abundant land resources, stable energy supply, and suitable climate conditions, has become an ideal alternative location.

Additionally, in March of this year, Anthropic signed a Memorandum of Understanding (MOU) with the Australian government to cooperate on AI safety research and national AI initiatives, paving the policy path for its large-scale infrastructure deployment.

Infrastructure Arms Race Among Giants: From Cloud Leasing to Physical "Land Grabbing"

Aggressive infrastructure expansion is a necessary requirement for Anthropic's commercialization explosion and accelerated capital market moves. Currently, Anthropic's annualized revenue has soared from $9 billion early this year to over $44 billion, with inference infrastructure gross margins rising above 70%, marking its first profitable quarter and entering a phase of rapidly increasing profit margins.

Supported by strong financial performance, the company completed a $65 billion funding round last month, achieving a valuation of $965 billion, and has officially filed its IPO application documents. However, the iteration of large model capabilities and the proliferation of agent applications are consuming massive amounts of computing power. To avoid computing shortages stifling growth momentum, Anthropic must lock in long-term, low-cost underlying computing resources in advance.

In the computing infrastructure arms race, top AI companies are shifting their strategies from pure cloud leasing to deep involvement in physical infrastructure construction. Similar to competitors like OpenAI and Google, Anthropic is building a diversified computing supply chain.

In addition to physical data centers in Australia, Anthropic recently signed a nearly $45 billion cooperation agreement with SpaceX, reached an $1.8 billion deal with Akamai, and continues to purchase chips and cloud services from Google. This hybrid model of "self-built physical infrastructure + alliances with giants + leasing from cloud providers" is not only a defensive move to hedge against single supply chain risks, but also the foundational base for AI giants to maintain competitive advantage in the next technology cycle.

Risk Warning and Disclaimer

        The market carries risks, and investment requires caution. This article does not constitute personal investment advice, nor does it take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article fit their particular circumstances. Investments made based on this article are at one's own risk.