Anthropic signed its first Australian data-center agreement on September 16, committing to a project Reuters described as part of roughly $22 billion of data-center construction in the country. On the same day, Fortune reported that the AI lab would open its first Southeast Asian office in Singapore in October, with nine roles already posted.
The two moves land on opposite ends of the same strategy. Singapore is where Anthropic will hire the people who sell and support its models. Australia is where it will secure the physical capacity to run them. A company that spent its early years renting servers from cloud providers is now signing directly for infrastructure on a new continent.
The Australian project is the more consequential of the two. Reserving data-center capacity ties Anthropic to a site and a power contract for years, a commitment model developers have historically avoided by renting from Amazon, Google and Microsoft. Anthropic has trained its frontier models on those clouds, and it keeps a deep commercial relationship with Amazon, which has invested billions in the company.
The Singapore office is a quieter signal, but a telling one. Southeast Asia has become one of the fastest-growing markets for AI services, and rival labs, OpenAI among them, have spent the past two years building out local teams. The nine posted roles, according to Fortune, point to a small start focused on go-to-market staff rather than a research hub.
The expansion was not the only Anthropic news of the day. Pharmaceutical maker Novo Nordisk announced a partnership to use Claude to speed drug discovery, another step in the lab’s effort to court industries with deep research budgets and long development timelines.
The commercial momentum stands in contrast to the scrutiny gathering around the company’s leadership. Moneycontrol reported on September 16 that scientists have questioned the AI safety plan championed by Chief Executive Dario Amodei, along with the independence of the safety oversight body he has proposed.
That tension sits at the center of Anthropic’s public identity. The company has built its brand on the idea that frontier models should be developed slowly and carefully, a position that distinguishes it from rivals. Amodei has spent the week arguing that leading labs should slow down, a stance that put him at odds with Nvidia’s Jensen Huang and Meta’s Mark Zuckerberg.
The safety argument carries business implications as well as scientific ones. Anthropic holds less infrastructure and fewer compute contracts than OpenAI, and a slowdown in frontier development would, in theory, narrow the gap between the two. Critics inside the scientific community, according to Moneycontrol, see the proposal as a way for a leading lab to freeze the field before rivals catch up.
Anthropic has not publicly answered the criticism. Its expansion plans, meanwhile, proceed on schedule, a sign that whatever its leaders say about slowing frontier research, the commercial team is still racing.
Analysts said the Australia and Singapore moves reflect a broader reality for AI labs: compute and distribution are now the two constraints that matter most. Model quality, once the industry’s obsession, is being joined by questions of who controls the data centers and who reaches the customers first.
The Australian commitment also signals something about where the next round of capacity is being built. Data-center developers have turned to Australia for its land, its renewable energy and its access to Asian markets, and the country has become a destination for the kind of long-term power contracts that AI training requires.
For Anthropic, the week framed both halves of its story. It signed for compute in Australia and opened a sales front in Singapore, while its founder defended a safety agenda that a growing number of voices are challenging. The company is expanding across two continents and arguing about how fast it should go, in the same breath.
Anthropic’s expansion mirrors a wider scramble among AI labs to lock up capacity and customers before the next generation of models arrives. OpenAI has struck its own data-center agreements, and the cloud providers have announced hundreds of billions of dollars in new capacity. The labs that want to train and serve frontier models are no longer satisfied renting from a rival’s cloud.
Anthropic has leaned on a narrow set of backers. Amazon and Google have both invested billions in the company, an arrangement that gives the lab privileged access to their clouds even as it pursues its own infrastructure deals. The Australian agreement suggests the company wants a third option, capacity it controls outright.
The drug-discovery work with Novo Nordisk points to where the company expects its enterprise revenue to grow. Claude has found traction among programmers and corporate customers, and partnerships with companies that hold large, specialized data sets, like a pharmaceutical firm’s library of compounds, are the next frontier.
The next test, analysts said, will be whether the enterprise deals and infrastructure commitments translate into the revenue growth investors have come to expect from the leading labs. Novo Nordisk and the Australian data-center project are forward-looking commitments, and their payoff will arrive over quarters, not days.
