Andrew Hastie called on Australia to dramatically scale up investment in artificial intelligence on June 15, 2026, comparing the global race to develop and control AI to the Cold War nuclear arms competition — a framing that positions AI capability not as an economic ambition but as a defence imperative.

Speaking at a forum focused on national resilience, Hastie argued that Australia needed to reduce its dependence on the United States for AI innovation. The nuclear analogy is deliberate and pointed: it implies that relying on an ally's technology is a form of strategic exposure, not just an efficiency gap. In the Cold War frame, deterrence required indigenous capability — borrowed arsenals did not count. Hastie appears to be making the same structural argument about AI infrastructure.

The call lands at a moment when Australia's policy community is actively debating what "sovereign AI" actually requires. The Australian Academy of Technology and Engineering laid out an investment blueprint arguing that government investment is essential to establishing sovereign AI capability and capturing a higher share of potential AI value. The academy's framing is less martial than Hastie's but reaches the same conclusion: passive market participation leaves national AI capacity dependent on foreign capital, foreign compute infrastructure, and foreign data governance decisions.

The strategic logic here is straightforward. A country that does not own meaningful AI infrastructure — the data centres, the trained foundation models, the chip supply lines — cannot guarantee that infrastructure will remain available, unmodified, or uncompromised when relations with supplier nations shift. The US-Australia alliance has been extraordinarily durable, but strategic planning does not anchor to current political temperatures. Hastie, as a former defence minister with deep exposure to intelligence community thinking, is reading off a familiar risk register.

What makes the nuclear comparison useful — and also imprecise — is that nuclear weapons had a relatively clear threshold: you either had a deterrent or you didn't. AI capability is distributed across a spectrum, with interdependencies at every layer from semiconductor fabrication to inference costs. Australia cannot replicate the full stack domestically at any realistic budget. The practical question Hastie's statement opens up is where on that stack sovereign control matters most. Compute sovereignty, data sovereignty, and model sovereignty each carry different costs and different strategic payoffs.

Australia's existing AI posture sits well below most comparable economies in state-directed investment. The ATSE blueprint makes that gap explicit. Hastie's forum remarks, framed around national resilience rather than industrial policy, are a bid to move this debate from an economic register to a security one — where higher funding thresholds are politically easier to justify and where the relevant ministerial authority sits closer to cabinet's centre of gravity.

The Cold War framing also carries a warning about pace. Nuclear programmes that fell behind rarely caught up. If Hastie's analogy holds, the cost of delay is not a missed productivity dividend but a permanent structural dependency. That is the argument the defence establishment tends to find most persuasive, and it is clearly the audience Hastie is pitching.

Whether the government moves from forum language to funded programmes is the next test. Strategic rhetoric about AI sovereignty has circulated in Canberra for several years without producing the kind of capital commitment that would close the gap the ATSE blueprint identifies. Hastie's intervention raises the political salience of the issue, but salience and appropriations are different things.