SYDNEY — The artificial-intelligence company that trained Claude is going to need a lot more electricity, and it has decided part of it will come from Australia. Anthropic is planning to secure at least 1.4 gigawatts of data-center compute capacity in the country, according to people familiar with the matter, with construction costs estimated at about $15 billion, or roughly A$21.6 billion. The plan, first reported on July 5, is far larger than the market had expected and marks one of the biggest single commitments to Australian computing infrastructure ever made by a U.S. technology company.
Anthropic aims to have at least 1 gigawatt of that capacity operational by the end of 2027, the people said, and it has opened an office in Australia to manage the build-out of localized infrastructure. The company has been racing to secure computing capacity around the world as demand for its Claude models grows faster than the data centers that run them. Unlike rivals that rely mainly on rented cloud capacity, Anthropic has pushed to own and control more of its compute, and Australia is a central piece of that strategy.
Australia is an unusual choice, and that is partly the point. The country has abundant land, a stable grid, and a government that has made clear it wants to host AI infrastructure, but it has historically hosted little of the world’s cloud computing. Power is the constraint that now shapes where AI companies build, and Australia’s energy resources, including large reserves of natural gas and a fast-growing renewable build-out, make it attractive at a time when grid capacity in the United States has become the industry’s scarcest input.
The scale of the plan says something about Anthropic’s ambitions. A gigawatt of data-center capacity is enough to power a small city, and 1.4 gigawatts is more than most hyperscalers operate in any single country outside their home markets. Anthropic has said it expects computing demand to keep growing faster than supply for years, and its founders have argued that whoever controls the most compute will control the future of AI. The Australian plan is a down payment on that conviction.
The economics are less certain. Data centers in Australia cost more to build than in the United States, the local market for AI talent is thin, and the country’s grid, while adequate, is not designed for the kind of load concentration a single AI campus creates. Developers working on Australian projects say interconnection and power-supply agreements have become the gating items, and that securing 1.4 gigawatts will require multiple sites and long negotiations with state governments and grid operators.
For Australia, the investment would be transformative. The federal government has been courting AI infrastructure as a national priority, offering fast-track approvals and arguing that the country’s location, energy, and political stability make it a natural home for the industry’s next wave of data centers. A commitment of this size from a leading AI company would put Australia on the map for hyperscale construction and could pull in suppliers, power developers, and a share of the capital flowing into AI worldwide.
The plan also carries risk. Anthropic is committing billions to a country where it has no operating history, and the 2027 target for 1 gigawatt is aggressive by any measure. Data-center projects of this scale routinely slip by a year or more, and the power agreements that make them viable can take longer than the construction itself. If the timetable slips, the company will have committed capital early to a market that may not be ready.
Anthropic’s compute strategy has evolved in public. The company has agreements with Amazon and Google to run its models on their cloud infrastructure, and it has said those relationships will continue — but it has also argued that frontier AI companies need control over their own computing capacity to train models on schedule. The Australian plan is the most concrete version of that argument to date, and it follows a pattern of locking in capacity years ahead in markets where the company can build quickly.
Australia’s pitch to AI companies rests on geography and energy. The country sits close to Asia’s fastest-growing markets, its electricity grid has room in regions far from its cities, and it has abundant natural gas and renewable resources that can be dedicated to large industrial loads. State governments have competed to attract data centers with streamlined approvals, and the federal government has signaled it views AI infrastructure as a strategic industry. Whether the approvals and the grid can keep pace with the 2027 deadline is the open question.
For now, the direction is clear. Anthropic is joining the ranks of companies that treat data centers as a strategic asset rather than a utility bill, and it is doing so in a country that most of its rivals have ignored. Whether Australia can deliver the power and the approvals to match the ambition is the question that will determine whether this becomes a model for the industry — or a cautionary tale about the limits of the boom.
