Australia’s home-grown artificial intelligence developers have long suspected that the gap between Canberra’s rhetoric and Canberra’s cheque book is a wide one. This week one of them said it out loud. Xaana.ai, a Canberra-based company building large language models on Australian soil, has accused the Albanese government of preaching sovereign capability while quietly steering its real spending toward overseas providers, according to reporting in The Australian Financial Review.
The charge lands at an awkward moment for the government. It has spent much of the past year talking up the idea of a domestic AI industry, standing up a new Office of AI and pitching sovereign capability as a pillar of national resilience. Yet when it comes to the specific and expensive business of buying foundation models, the developers argue, the money keeps flowing to the American labs whose systems already dominate the market. The complaint is not that overseas tools are bad. It is that a government which claims to want a local industry is not behaving like a customer of one.
The news
Xaana.ai’s argument, as reported by the AFR, is blunt. If Canberra genuinely wants Australia to develop its own foundation LLMs, the single most powerful lever it holds is procurement. Government is one of the largest buyers of technology in the country, and a handful of early contracts can be the difference between a local model company reaching commercial scale and quietly folding. When those contracts default to offshore incumbents, the developers say, the sovereign ambition becomes a slogan rather than a strategy.
The company frames the issue as one of consistency. Ministers have spoken warmly about Australian AI, about keeping data and capability onshore, and about not being wholly dependent on a small number of foreign suppliers for a technology now woven through defence, health, welfare and the courts. The developers’ point is that none of that language shows up where it counts, in the tender documents and the purchase orders. Talk of sovereignty, in their telling, has not survived contact with the buying decision.
Two ways of reading it
There is a sympathetic case for the government’s position, and it is worth stating plainly. Public money carries a duty to deliver value and to work reliably, and by most measures the frontier models built by the large overseas labs remain ahead of anything produced locally. A procurement officer told to choose between a proven American system and a smaller Australian challenger is not being unreasonable to pick the one with the longer track record. Buying local for its own sake can shade into paying more for less, which is a hard thing to defend when the bill lands with taxpayers.
The developers’ answer is that this reasoning, followed to its conclusion, guarantees the outcome it claims to lament. No local model will ever match the incumbents if it is never given the revenue, the data and the reference customers that would let it improve. Every established AI company on the planet grew on the back of early buyers willing to take a chance, and governments have historically been among the most important of those buyers. If Canberra waits for a local model to be demonstrably world-beating before it will buy one, the developers argue, it will wait forever, because world-beating is a destination you reach by being bought, not a precondition for the first sale.
That tension, between value-for-money orthodoxy and industry-building intent, is the real substance of the dispute. It is also not new. Similar arguments have played out in defence manufacturing, in shipbuilding and in critical minerals, where Australia has repeatedly discovered that a domestic capability it did not nurture is a capability it later has to import at a premium.
Why it matters for Australia
The stakes here run well beyond one company’s order book. Foundation models are becoming infrastructure in the literal sense, the layer on which banks, hospitals, insurers and government agencies build their own tools. A country that owns none of that layer is a country whose most sensitive systems ultimately depend on decisions made in boardrooms it does not sit in, under laws it does not write. The recent debate about sovereign AI capability, canvassed by bodies including the Australian Strategic Policy Institute, has kept returning to exactly this vulnerability.
Procurement is where the abstract argument becomes concrete. Australia does not need to out-build Silicon Valley to have a meaningful sovereign position, but it does need at least a few local players with the scale to keep improving, and scale in this industry is bought with contracts. Xaana.ai’s intervention is really a claim that the government has the tool to build that capability sitting in its own hands and is declining to use it, even as it insists the capability matters. For Canberra, which hosts a growing cluster of AI firms precisely because of its proximity to federal buyers, the credibility of the whole sovereign pitch rests on whether the words are matched by orders.
There is a political dimension too. The Office of AI and the government’s broader framework have already drawn criticism from the Coalition and from parts of the industry as arriving late and light on detail. A charge of hypocrisy from a local developer, rather than from a rival party, is harder to wave away as point-scoring, because it comes from precisely the constituency the policy is meant to help.
What’s next
The pressure now is on the government to show its procurement behaviour matches its language, whether through a sovereign preference in tenders, a carve-out for local models in agency buying, or simply a public accounting of how much of its AI spend actually stays onshore. None of those steps is guaranteed, and each carries a cost that Treasury will want justified. But the developers have reframed the question in a way that is difficult to ignore. The test of a sovereign AI ambition is not the number of speeches it generates. It is the number of Australian invoices it pays. Until those two figures move closer together, expect the local industry to keep pointing at the gap.
Sources: The Australian Financial Review.
