Australia should treat artificial intelligence less like a passing boom and more like the start of a new national industry, according to Peter Lewis, who argues the country should take a stronger stake in the infrastructure behind it and force companies to pay for the way they built their models.
Lewis says the chief executives of Microsoft and Anthropic have been courting Canberra with an eye to training AI models in Australia, where he says the companies could find secure places to build the massive systems they need. That push comes after, he argues, the firms have exhausted the US public’s tolerance for data centres and the water and energy they consume.
“If Australia is home to an AI gold rush, let’s not squander it. Let’s fjord a different path,” Lewis says in the piece, published by. He frames the opportunity in terms that recall the way resource booms reshaped other countries, saying Australia could produce artificial intelligence the way the Middle East produces oil. He also points to Norway, whose sovereign wealth fund grew out of North Sea oil deposits in the 1970s and is now worth more than A$3tn.
The comparison is meant to do more than flatter Australia’s ambitions. Lewis says Australians are already asking why the gas industry was allowed to operate under different rules, with little control over the asset, and he says that when huge gas deposits were discovered off the West Australian coast, the Hawke government chose not to be an upfront investor. Forty years on, he argues, that decision looks like a missed chance to shape the returns from a national windfall. The lesson, he says, is that the financial arrangements for a new industry need to be designed at the start, not after the profits have been made.
That is where the article turns from industrial strategy to copyright politics. Lewis says the AI models now being built rest on what he calls a criminal enterprise, and he argues that court cases have already established that hyper-scalers systematically and illegally scraped the web to train their systems. In that context, he says, Scott Farquhar’s pitch last year for a data-mining exemption to copyright amounted to a retrospective and ongoing legalisation of theft, a proposal that drew immediate backlash and was quickly walked back.
Lewis’s answer is blunt. “We must hold the line on copyright: a condition for operating these centres should be reparations,” he says. His argument is that if Australia is going to host the land, power and political stability that AI companies need, then the country should not simply subsidise them with cheap resources and open access. It should use its leverage to claim a share of the value and to demand payment for what has already been taken.
The debate now is whether Canberra sees AI infrastructure as just another foreign investment pitch or as the foundation of a strategic industry. If Lewis is right, the choice will not be made by the companies already lining up to train their models here, but by whether Australia decides to set the terms before the data centres arrive.
