Australian businesses spent 2026 racing to put AI into everyday work. Now the two countries that build the frontier models everyone is racing to use are separately moving to keep the best of them closer to home.
China's commerce ministry is reportedly weighing new export controls on its most advanced AI systems, following a similar move out of Washington weeks earlier. For Australia, which develops little frontier AI of its own, the timing is awkward: two of the world’s largest model providers are independently deciding how much of their technology can leave their home countries.
According to a Financial Times report, China's Ministry of Commerce has been consulting with leading AI and semiconductor companies, including Alibaba, ByteDance, and Zhipu, on rules that could prevent foreign users from downloading the model weights behind China's most advanced systems.
Reuters separately reported that Chinese officials had already met with AI companies this month to discuss limiting foreign access to advanced models, including some not yet released.
The proposals extend to hardware. Regulators are also asking whether overseas manufacturers such as Qualcomm and TSMC should be barred from producing advanced chips designed by local manufacturers, and whether foreign acquisitions of Chinese firms working on agentic AI should be subject to tighter scrutiny. Nothing has been finalised, and Beijing is still gathering industry feedback.
Two capitals, the same instinct
China's deliberations echo something that already happened in the US. In June, the US government ordered Anthropic to cut off foreign nationals from its two most capable models, Claude Fable 5 and Mythos 5, citing national security concerns tied to their cybersecurity capabilities. Access was suspended for over two weeks before restoration.
Neither government is banning AI exports outright. Both are drawing a line around their most advanced systems and deciding, model by model, who outside their borders gets to use them, and on what terms. That is a meaningful shift from the past few years, when frontier labs in both countries competed mainly by giving models away as widely as possible to build market share.
What it means for Australian AI adoption
That line falls squarely across markets like Australia's, which has no frontier lab of its own and depends almost entirely on foreign model providers.
Australian SME AI adoption hit 44% in February 2026, the strongest reading in months, according to the National AI Centre's SME AI Pulse tracker. Deloitte's 2026 State of AI in the Enterprise report similarly found Australian organisations investing in AI at a pace comparable to global peers.
Rapid adoption built on borrowed infrastructure carries a specific kind of exposure.
Every enterprise, hospital network, or government agency that has wired workflows around a foreign-frontier model is, in effect, one export-control decision away from losing access to it, as Anthropic's customers briefly discovered in June.
As Washington and Beijing increasingly treat their leading models as strategic assets rather than ordinary commercial products, Australian buyers may need to start factoring licensing terms, geographic carve-outs and "trusted user" status into procurement decisions that used to come down to price and performance alone.
Does Australia have what it takes to build its own?
The instinctive fix is that sovereign AI capability that a foreign export-control order can't reach has moved from a niche policy paper to a live national debate. The Australian Computer Society has put the cost of a competitive national AI infrastructure build at roughly AU$2 billion to AU$4 billion, a fraction of what training a genuine frontier model from scratch would require.
Analysis published in The Conversation breaks the sovereignty question into four parts: where data physically sits, who controls the compute, whether the models themselves are built locally, and whether Australia sets its own AI rules rather than inheriting another country's export controls by default.
Australia has only partial answers to any of the four. Providers such as Macquarie Government, Vault Cloud and AUCyber offer sovereign data and compute, but there is no Australian-built frontier model, and most model inference for Australian users still happens offshore.
Currently, what seems realistic is a hybrid approach: Using global frontier models where there's genuinely no local substitute. Then, investing in the compute, workforce and purpose-built local systems needed for sensitive workloads in health, biosecurity and public services, where dependence on a foreign model carries the highest cost if access is ever cut.
Where does this leave Australian decision makers?
For Australian IT and procurement leaders, waiting to see how China's proposals land isn't a strategy. Mapping which internal workflows depend on a single foreign model, diversifying vendors where credible alternatives already exist, and pushing suppliers to disclose where a model actually runs and under whose jurisdiction are all things that can start now.
Australia doesn't need to immediately build the next frontier model to cut its exposure. But after watching Washington switch off access to a frontier system overnight, and Beijing consider doing the same, it can no longer assume that access to today's models is guaranteed tomorrow.