Prime Minister Anthony Albanese’s first major speech on artificial intelligence on Wednesday emphatically, if belatedly, elevated AI to where it belongs as a national priority.
In particular, he made the crucial declaration that the government is serious about bringing frontier AI investment to Australia, indicating it was our best path to sovereignty and economic resilience.
This means building data centres for training the latest models such as Claude, GPT and Gemini. It was a welcome leap forward in ambition, but there are some significant hurdles yet to clear if Australia is to make this work. And as Albanese acknowledged, we need to move now.
Reflecting what’s at stake, Albanese pointed out that the great lesson of today’s turbulent world was ‘that if we are always dependent on someone else, somewhere else, we will always be vulnerable’.
Australia is a technological middle power with many strengths but also many limitations.
It’s hard to avoid dependency and create agency, choices and sovereignty – all ideals that Albanese name-checked in his speech – given the frontier is already so heavily dominated by two players, the United States and China.
Our natural advantage as a host of data centres for frontier training is our greatest strength, but only if the right investment conditions are set to ensure Australia doesn’t become, as Albanese put it, just ‘a data warehouse for AI products made overseas’.
That’s where the speech’s key policy announcement comes in: a national regulatory framework for large AI data centres, working with the states and territories, and backed by legislation expected early next year that would ensure data centres serve national interests.
It includes the obvious crowd-pleasers about managing energy and water use, but the bigger question, which Albanese didn’t get into, is the investment conditions that bake in sovereign interests.
Australia’s strategic logic here is twofold. First, as frontier models become more powerful, it will get harder for the US to share them even with trusted allies, as the recent Mythos saga showed. But if Australia is helping to build them on our soil, we gain serious leverage, a little like Australia’s geography gives it access to intelligence collected using American technology at Pine Gap.
Approvals by the Foreign Investment Review Board could demand commitments to provide reasonable Australian access to models trained here.
The second stage of the logic, taking account of Washington’s increasing unpredictability and willingness to use its power coercively even against allies, is that having frontier training capacity in Australia could expand our future options to build our own models. Investment conditions could include reserved compute capacity for Australian research, training and inference, along with the necessary data sovereignty requirements.
We’re unlikely to compete with the US and China on frontier models. But if the world really goes south, we would benefit from having greater capacity to build and train sovereign foundation models, which are general-purpose and adaptable, for vital purposes such as national security and critical infrastructure.
This could be done in partnership with other democratic middle powers. It’s ambitious but, if Albanese is true to the spirit of his speech, these big bets should be on the table.
Public investment will be needed at some stage. Other comparable countries are already there. Britain, Canada, Japan and South Korea all have government-funded programs for some form of sovereign computing power and domestic model development that run into the billions of dollars.
But before any of that is possible, the government needs to solve the headache about copyright reform. AI models are trained on vast amounts of data, including copyrighted content on the internet – a major complaint for the creative and media industries. Frontier labs say Australia’s strict copyright laws deter them from training here.
Albanese’s rhetoric about protecting creatives and journalists was strong, but he stopped well short of indicating he’d let copyright protection be the death of the frontier AI investment that he championed with at least as much intensity.
The government seems intent on finding a compromise whereby it upholds and respects the rights of copyright holders, but also gives frontier labs assured and finite legal liabilities without setting a precedent that would undermine their existing fair use exemptions in more permissive jurisdictions overseas.
Somewhere along the way, a lot of money needs to change hands. Anthropic’s reported comment that there is a long tail of smaller copyright holders they can’t realistically negotiate with individually indicates there’s a possible answer in dealing with the big rights holders like media companies and creating a fund for those in the long tail.
Coming back to the urgency, AI labs are already planning their compute infrastructure for 2027 and 2028. For all that Albanese talked up the many advantages of Australia as a second home for training US models, countries including Canada, Japan and Norway are promoting themselves heavily.
Albanese has finally conveyed the ambition we needed to see. Now comes the hard work of implementation. Australia has a real chance to establish itself as a global player with leverage to steer the most consequential technology in human history in ways that reflect our values and serve our interests. We must seize it.
This article was originally published in the Australian Financial Review.
Leave a comment