If you were a tech giant that had set up data centres in the Middle East because of the cheap fossil-fuel power and permissive regulatory environment, you might now be looking wistfully at Australia.

Right on cue, the Australian government released on 23 March a set of ‘expectations’ that any new data centres built here serve the national interest on criteria such as security, jobs, local innovation, electricity and water supplies. It’s a flexing of muscle that puts the AI-infrastructure industry on notice.

But Australia should be careful not to overplay its hand, given the clear strategic imperative to become an AI infrastructure powerhouse. Along with creating applications, infrastructure is our clearest opportunity to find a place in the global AI industry. We need to capitalise on it fully by striking a prudent balance between meeting our sovereign needs and sending signals that don’t scare off investment.

Yes, we’re an attractive destination: a Five Eyes intelligence partner, handily located at the southern hinge of the world’s fastest growing region, with land, renewable energy, regulatory certainty and political stability.

But we have our irritants, notably our copyright and tax laws. And we’re not the only destination. The United States might be struggling with community backlashes and supplying enough electricity, but it’s still the centre of gravity for AI. Southeast Asia might carry rule-of-law and geostrategic risks, but it has cheap power. Europe might be a regulatory headache but it’s stable and close to existing centres of economic and political power. India, meanwhile, recently declared a 20-year tax holiday for any foreign firm using Indian data centres to provide services globally, underscoring its determination to win its share of the global gold rush for compute.

The expectations are meant to provide clarity to the industry, though serious data centre applicants likely won’t have a much clearer idea than they had last week. You’d have to be living under a hunk of silica not to know these were broadly the federal government’s priorities for AI infrastructure.

The government says it will look more kindly on applications that meet these criteria, but it hasn’t defined targets. How much cooperation with regulators and suppliers on peak-load management is enough? How many apprenticeships should a project generate? How much good faith with local communities is enough?

This could be called smart negotiation, giving applicants an incentive to go further at the risk of overshooting rather than meeting only some declared minimum requirements. It’ll also have the benefit of building community confidence, especially given fraught energy politics. But it’s a demonstration of partial clarity at best.

Nor does it directly address the fact that the federal government has only a limited role in assessing data centre proposals. Its main levers are the Foreign Investment Review Board and responsibility for the national electricity rules and market, but states do much of the approvals work for AI infrastructure. Accordingly, the federal government says it will work with the states to turn the principles into concrete criteria, but how exactly it does so will determine the effectiveness. An energy ministers’ meeting in May and water ministers’ meeting mid-year should give us some insight.

Concerns about power and water usage can be addressed simply by data centres contributing enough clean-energy and water-supply infrastructure to neutralise their drain on these networks and hopefully provide a net boost.

Generating jobs is trickier, because it’s inseparable from the broader AI revolution, which will create economic winners and losers. Data centres create a lot of construction work but relatively few ongoing operational jobs. The impact of AI on downstream jobs is one of the central debates of our time, from which no clear answer has emerged.

The government’s inclusion of national-security principles is vital but will need to evolve into more concrete benchmarks. Protection of personal data, countering threats and disruptions and limiting both physical and digital access are all sensible requirements. One of the most important roles for the federal government will be to keep the Security of Critical Infrastructure Act relevant and manage how it applies to data centres.

Australia is happily out of range of Iranian drones and missiles, but it’s very much in range of Chinese intercontinental ballistic missiles. And if AI’s disruption of work turns out to be severe, how long will it be before economically marginalised groups decide that the most visible and exposed manifestations of the AI era make tempting targets for sabotage attacks?

There’s also work to do on aligning data centres with our geopolitical interests. Do we want our data centres providing cloud services to Chinese app TikTok, which is widely regarded as a vector for political influence? That is hardly in our national interest.

The most useful expectation in this week’s statement is that some computing power should go to Australian companies and researchers on favourable terms as a public interest to drive innovation and build the smart end of the Australian economy. Infrastructure is a great way to create clusters of innovation which can then become locomotives for the economy overall. It’s a good proposal, though again more detail will be needed.

Inevitably, the national interest will grind against the industry’s interest in making returns on huge capital outlays. This week’s announcement sets some helpful direction, but more work is needed to find the right balance. The biggest blow to the national interest would be if we failed to make the most of this chance to establish ourselves in the global AI value chain.