“You are creating policy in an environment of great uncertainty.” — Andrew Charlton, the assistant minister with responsibility for AI.
Andrew Charlton’s roadmap from the ANU Crawford School and the Joe Walker Podcast
In a speech to the Australian National University’s Crawford School and a two-hour appearance on the Joe Walker Podcast, Andrew Charlton sketched what the report describes as the most detailed roadmap Australia has seen for responding to rapid AI change. Charlton acknowledged the deep uncertainties: AI leaders believe the technology might be able to improve itself without human help within a year, but “they’re not sure.” Likewise, experts disagree on whether AI will match virtually any human skill by 2028 or by 2035, and economists remain split on whether AI will primarily destroy jobs, create unprecedented new demand, or both.
Data centres, computing power, and foreign investment conditions
The piece frames AI as three interlocking elements: talent; data; and computing power. Charlton was explicit that Australia needs to build computing capacity in the form of data centres — not simply to run models but to train them. The government, the article notes, plans to attach conditions to foreign investment in computing infrastructure, including a demand that investors reserve some of the capacity for Australian companies and researchers. As Charlton put it in his Crawford speech: “Our goal ... is to convert physical investment into national AI capability.”
This site is the portfolio.
OSINTSights runs on Cloudflare Workers, D1, R2, and Vectorize, with an AI pipeline on Hetzner ARM. Nubivance designed, built, and operates it. We do the same for clients.
See what we buildOpen-weight models, Chinese offerings, and strategic risk
The article contrasts blockbuster “frontier” models — GPT, Claude and Gemini — with a burgeoning industry of smaller, task-optimised models that do not require the massive compute of frontier systems. It highlights the strategic complication that the best open-weight models today are Chinese, and that open-weight weights can be downloaded and adapted cheaply. That cheap accessibility raises “security and strategic risks”: the piece says Australian banks, telcos and power companies “can’t rely on Chinese open-weight models.” It also notes a broader ecosystem: there is an open-weight push in the US, reasons to think China’s government may crack down on companies’ open-weight proclivities, and a notable European competitor — France’s Mistral — which the article calls “the best model builder outside the two superpowers.”
Model-building skills: fine-tuning, post-training and distillation
Charlton’s central policy contention is that data centres alone will not capture the biggest returns: the intellectual property — models and applications — will. Breakthroughs in fine-tuning, post-training and distillation, which allow users to take an existing model and tweak it fairly cheaply, strengthen the case for Australia to participate in model-building rather than cede that space entirely. The article argues model-building skills will be vital whether organisations build from scratch, improve existing models or assess the strengths and risks of new offerings. Attracting AI labs and big tech investment in data centres capable of training models, rather than merely running them, is presented as a way to seed research clusters that grow and retain skilled workers.
What this means for Australian banks, telcos and power companies; researchers and universities; and AI labs and investors
- Australian banks, telcos and power companies — will face strategic and security choices about whether to permit Chinese open-weight models in critical infrastructure; the limits on such use are described as “one of the big unanswered questions.”
- Researchers and universities — stand to gain if government conditions on foreign investment ensure reserved capacity, and if public funding supports centres of excellence and “elite training colleges” in cooperation with industry.
- AI labs and investors — are the target of policy incentives: attracting them to build training-capable data centres in Australia would, the article argues, create the research clusters and skilled talent that underpin national capability.
The piece ties these threads to a strategic verdict: models are becoming a resource at the heart of the technology stack, and the people who know how to make them form a sought-after global elite. Given the choices between a risky China and a capricious US, the report’s judgment is blunt — Australia’s best bet at sovereignty and agency is home-grown skills in the decisive technology of our lifetimes.
