Australia tells big tech: invest here, but on our conditions

Australia has begun telling the world's largest artificial intelligence companies that they are welcome to build data centers on Australian soil, but only on conditions that no major economy has yet managed to impose at scale. Microsoft, Google and Anthropic all face demands that their facilities run on clean electricity, account for water consumption, deliver training programmes for Australian workers and share computing capacity with local researchers and startups. It is a deliberately transactional approach, and one that Canberra knows carries risk.

A third way between American backlash and European rigidity

The United States now hosts roughly 4,000 data centers, and the pace of construction has generated a political backlash that Australian officials have watched with unease. Communities in Virginia, Texas and elsewhere have pushed back against the power demands, water consumption and noise of server farms that employ few people while straining local infrastructure. Europe, by contrast, has responded to AI with the EU AI Act, a comprehensive regulatory framework that sets rules on what AI systems can do and how they must be governed, but that critics say has done little to ensure European countries capture economic value from the technology.

Australia is attempting something different: using its attractiveness as an investment destination to extract concrete commitments before granting permission. The country's 28 million tech-enthusiastic citizens, its position as the world's twelfth-largest economy and its abundance of land make it a useful market for AI companies seeking growth outside the United States and China. Canberra's calculation is that this leverage will not last forever, and that it must be used now.

Prime Minister Anthony Albanese, who has acknowledged he does not personally use AI chatbots, has nonetheless made tech policy a priority since his government's successful implementation of restrictions on teenage social media use. That episode gave him confidence that Australia can set its own terms. In a July speech, Albanese told Australians: "We can set the terms, we can determine AI's social license. But we have to do it now." He has set early 2027 as the deadline for finalising AI legislation.

The mining boom lesson Australia does not want to repeat

The intellectual engine behind Canberra's approach is Andrew Charlton, the Assistant Minister for the Digital Economy, whom Albanese tasked with developing a national AI strategy. Charlton's argument draws directly from Australia's decades-long minerals boom, and it is not a flattering comparison.

Australia digs up some of the world's most valued critical minerals, but other countries process them and manufacture products from them. The country sits closer to the bottom of those value chains than the top, despite owning the raw material. Charlton acknowledges that the mining era did deliver good wages, government revenue and a mining-services sector, because Australia controlled what he calls the "choke point" in the value chain: the ore itself.

AI offers no equivalent choke point for Australia. There is no raw resource that the country alone controls. The value sits in the models, the intellectual property and the customer relationships, all of which are held by foreign companies. In a speech on 18 August, Charlton put the risk plainly: "Unless we take timely action, Australia is on course to be a large and permanent importer of intelligence." His ambition is for Australia to participate in roughly 80 percent of the AI supply chain, from data centers through model training to application development, describing this as "sovereignty across the whole stack" on a podcast the day before.

Computing access as the new industrial policy

Central to Charlton's thinking is the idea that computing capacity is the gatekeeper of the AI economy. Without access to high-end processors, Australian researchers and startups cannot train models, test ideas or build applications. The contestable layer of the AI economy, as he puts it, is not contestable by Australians if they cannot get compute.

This is why the government wants data center operators to share capacity with domestic researchers and startups, not simply build facilities that serve foreign companies. Data centers become instruments of industrial, security and energy policy rather than straightforward infrastructure deals. The government established an Office of AI within the Prime Minister's Office in July to coordinate this approach and give investors a single point of contact.

New South Wales, the country's largest state, has already released six principles for data center development and promised faster approvals for projects that comply. The promise of speed matters to tech companies, but the conditions attached to that speed are the price of admission.

Voters worry about power bills, not productivity gains

The political calculus underpinning Canberra's approach is rooted in public anxiety that has little to do with the abstract debates about generative AI that have dominated discussion since 2022. After years of arguments about productivity gains and doomsday predictions from technology executives, Australian voters have homed in on something more tangible: the cost of electricity.

A poll of 1,230 Australian adults conducted by JWS Research between 6 and 10 August found that two-thirds favour a cautious approach to AI, even if it means missing some economic opportunities. The government's overarching goals, as stated internally, are to protect household electricity prices and to integrate Australia into the AI supply chain. The order of those priorities is revealing.

Nicolette Boele, an independent MP representing a wealthy Sydney electorate, received more than 750 responses when she surveyed constituents about AI. More than 80 percent said they were not confident the government would regulate data centers effectively, and three-fifths worried about a repeat of what happened with the gas industry, where domestic consumers ended up paying more while exports boomed. Water and energy impacts topped the list of concerns.

The Fortescue model: stand alone or don't build

The most aggressive response to the energy question comes not from government but from the mining magnate Andrew Forrest. Fortescue Mining is constructing a 2.3-gigawatt off-grid green energy network around its operations in Western Australia, which it describes as one of the largest such networks in the world. Forrest's position is blunt: data centers should generate their own clean power and remove any risk of driving up costs for other consumers.

It is a model that appeals to politicians mindful of the backlash in the United States, where data centers have been blamed for pushing up local electricity prices and straining water supplies. If Fortescue can demonstrate that large-scale compute can be powered entirely by off-grid renewables, the argument that data centers must inevitably burden existing infrastructure becomes much harder to sustain.

What middle powers can actually demand

Australia's negotiating position rests on a premise that researchers Markus Anderljung and Stephen Clare have articulated clearly: middle powers, defined as countries with strong economies but limited cutting-edge AI capabilities, have more influence over the trajectory of AI development than is commonly assumed. The United States dominates high-end compute, private investment, frontier-lab valuations and chip design. But middle powers collectively account for more of the world's GDP than the United States and China combined, and they hold critical links in the semiconductor supply chain.

Anderljung and Clare argue that at least until 2026, middle powers arguably had more influence on frontier company behaviour than the US government did. Several of the most consequential AI governance moves since the launch of ChatGPT came from countries outside the two superpowers. The lesson for Canberra is that leverage exists, but it is perishable.

There is an unspoken dimension to Australia's strategy. A government that secures a stronger social license for AI, by demonstrating that it can manage the technology's energy and environmental impacts, gains more room to deal with the labour market disruptions that AI adoption accelerates. The political risk is not only higher power bills; it is job displacement in sectors where AI substitutes for human labour. A government seen as having extracted concessions from big tech is better positioned to manage that transition.

The gamble on timing

Australia's approach is not without tension. Demanding clean power, computing access and local training could push some investors toward jurisdictions that ask less. The United States offers scale and few conditions. Southeast Asian countries offer lower costs. Europe offers regulatory certainty, albeit with heavy compliance burdens. Australia is betting that its combination of political stability, rule of law, English-speaking workforce and geographic position between American and Asian time zones is attractive enough that companies will accept the strings attached.

The risk is that this window narrows. AI companies are making infrastructure decisions now, and jurisdictions that move faster on approvals, even with fewer conditions, may capture investment that never returns. South Australia's premier, Peter Malinauskas, announced a royal commission into AI impacts on 10 August, a move that could either sharpen policy or slow it down depending on what the commission finds and how long it takes to report.

Charlton's argument that computing access determines whether a country can participate in the AI economy is sound as far as it goes. But requiring foreign companies to share capacity with domestic startups is easier to announce than to enforce. The details of how much capacity, at what price, and with what enforcement mechanisms, have yet to be worked out. These are the terms that will determine whether Australia's approach becomes a model for other middle powers or a cautionary tale about overreach.