For every $100 in data centres, $80 leaves Australia almost immediately

There’s a number the data centre industry doesn’t love talking about. For every $100 poured into a fully loaded hyperscale facility in Australia – the kind Amazon, Google and Microsoft are building to serve new AI tools – somewhere between $70 and $80 leaves the country almost immediately. It flows to semiconductor makers in Taiwan, server manufacturers in the United States and cooling equipment giants in Europe.

You wouldn’t know this from the way the industry talks about itself. The headline numbers sound intoxicating: $26 billion in baseline investment by 2030, potentially $52 billion more, according to a Deloitte report – commissioned and paid for by Google – that envisions turning Australia into Asia Pacific’s AI hub.

The stakes are unusually high. Data centre investment decisions being made in the next two to three years will determine where this infrastructure sits for a generation, and to what degree Australia actually benefits. The promise is that Australian businesses will gain cheaper access to AI computing power, attract global tech talent, turbocharge productivity in industries from mining to healthcare, and catalyse billions in new renewable energy investment.

Atlassian’s billionaire co-founder, Scott Farquhar, reckons we should be “exporting megawatts as megabytes and getting paid megabucks”. Canva co-founder Cameron Adams says that data centres are the “biggest opportunity that I can see to add on an entirely new layer to Australia’s economy and kind of start diversifying us away from resources”. Steven Worrall, a former Microsoft executive now leading Telstra’s AI push, routinely calls it “one of the great economic opportunities for Australia”.

The Albanese government, too, has been ebullient. “Really, this technological revolution driven by AI is a huge focus of the cabinet,” Treasurer Jim Chalmers said last month, crediting data centres with single-handedly driving a rebound in national business investment. “There’s a lot of investment flowing.”

They might be right. But the Australian Bureau of Statistics is telling a different story, noting this year that the surge in data centre investment was “supported by a corresponding rise in imports, leading to a reduced impact on GDP”.

Translation: the money is passing through Australia like water through a pipe.

Follow the money

The leakage is significant, and structural. When a hyperscaler (think the likes of Amazon, Google) builds and fills its own facility, the IT equipment – the servers, GPUs, networking – eats 70 to 75 per cent of total cost, according to analysis from consulting firm Alpha Matica. Australia manufactures none of it – no chips, no servers and no racks. Every dollar goes straight overseas to the likes of chip companies Nvidia, AMD, Broadcom and their East Asian foundries.

Of the remaining 25 to 30 per cent for the physical building, the Alpha Matica modelling suggests roughly another $12 to $15 of every $100 invested disappears offshore to buy power supply systems, cooling infrastructure and generators from Schneider Electric, Vertiv and their ilk. What stays? About $10 to $15. That goes towards costs such as real estate, concrete, sparky labour, project management – all of which is in high demand from other sectors of the economy.

The economics improve dramatically when Australian-founded operators like NextDC, AirTrunk and CDC are the ones building. Their model is different: they construct the physical facility – the shell, the power systems, the cooling – and then lease space to tenants who bring their own hardware.

Because the imported computing equipment isn’t part of their spend, modelled estimates from Morgan Stanley Research indicate that about $45 to $55 of every $100 stays onshore. The total investment is much smaller – you’re building the building, not filling it with GPUs – but the share captured by the domestic economy is higher.

The tax question is worse than you think

If it is a big tech multinational behind a data centre, that raises another question: what slice does the government get in return?

Take Google for example. The tech giant paid $92.6 million in income tax in Australia in 2022, the last year it reported the amount of “gross revenue” it made in the country, which totalled a staggering $8.4 billion.

Equinix, a US-headquartered data centre giant that operates major facilities in Sydney and Melbourne, has reported $246 million in Australian revenue while paying $6 million in tax. Of course, there are plenty of perfectly legal ways of minimising the amount of tax paid in a jurisdiction like Australia, such as agreements to pay related companies in lower tax jurisdictions offshore for intellectual property or debt.

The Australian Tax Office is challenging some of those arrangements, and new rules cap debt deductions at 30 per cent of a common measure of a company’s earnings. But let’s be real: these debt strategies have been a feature of multinational tech operations for two decades, and we’re still chasing the same problems.

Meanwhile, the same multinationals building data centres on Australian soil are fighting tooth and nail against proposed copyright reforms that would force them to pay Australian media companies for the content their AI models were trained on: a reminder that the struggle over who captures value from the digital economy extends well beyond tax.

For context, the mining sector paid $48 billion in company tax in 2023-24, which is more than half of all tax collected from large corporates. Industry-commissioned research from Data Centres Australia, the industry’s peak lobby group, claims data centres generate the highest gross value added per unit of energy consumed of any sector, at $12.6 billion per terawatt-hour, though that figure captures the economic activity data centres enable, not the value the facilities produce directly.

Still, the gap between value generated in Australia and value captured by Australia remains gargantuan, and will continue to widen unless something’s done about it.

The multiplier defence

The industry’s strongest argument – and the hardest to verify – is indirect returns. The Deloitte report argues that for every dollar spent on domestic construction, roughly three dollars of economic activity flows through the wider economy: through supply chains, logistics, materials, local services and wages. Applied to the full hub scenario, Deloitte models $134 billion in cumulative GDP uplift by 2050, with the biggest gains coming not from the building phase but from the productivity boost as cheaper, faster AI capability spreads across Australian industries.

Daniel Roberts, whose company Iren runs renewable-powered facilities for Microsoft, says: “All we need to do is monetise [renewable energy] into a more refined product like compute and export it.”

That’s a compelling vision, even when you consider Roberts’ biases. You can see it, if you squint a bit: Australia as a refiner of energy into intelligence, the way it once refined bauxite into aluminium. But the e61 Institute – a non-partisan economic research agency – offers a cold shower: “Data centre operators are capital-intensive and relatively small employers”, with only 11,500 workers in the entire sector.

The productivity dividend requires massive follow-through in software, research and development and skills. It doesn’t materialise just because you’ve poured a slab in western Sydney.

Then there’s energy. Data centres consume 2 per cent of grid electricity now. Australian Energy Market Operator projects 6 per cent by 2030 growing to 12 per cent by 2050. Without matched renewables, wholesale prices could jump some 26 per cent in NSW by 2035.

Put that in household terms: if you’re already struggling with your power bill, the unchecked expansion of data centres without concurrent investment in new generation capacity could make it meaningfully worse.

Farquhar points out Singapore has imposed a moratorium because it’s run out of power and land. His point is that Australia – a continent many hundreds of times Singapore’s size, with some of the best solar and wind resources on the planet – doesn’t face those constraints. Fair enough. But having advantages and monetising them are different things, and we haven’t demonstrated we can do both simultaneously.

The verdict

For every $100 in hyperscale data centre investment, Australia’s direct take as it stands currently is painfully thin. The way it shakes out, we get $10 to $15 retained, modest tax income and a handful of jobs.

The real return – multipliers, productivity, sovereign capability – is legitimate but conditional on policy choices we haven’t yet made.

“We absolutely haven’t missed it,” Roberts says. “But we absolutely could miss it.”

Casey Flint, an Australian venture capital investor who now serves as chief of staff at San Francisco-based AI start-up ReflectionAI, frets about a world where the AI and compute boom happens elsewhere, leaving Australia as a mere consumer.

“My main worry is that there’s like an offshoring event on a level with or on a magnitude we’ve never seen before,” she says.

In that scenario, the country sends its data, capital and talent to offshore data centre hubs then buys back AI capacity at a premium.

Google has signalled uncertainty about its $20 billion Australian data centre commitment, and told the government it needs clearer policy settings before proceeding. “Capital is mobile,” the company said. “And the competition to attract this investment is fierce.”

There’s widespread agreement that Australia needs access to this kind of infrastructure. But we should go in with clear eyes about who really captures the value.

The Business Briefing newsletter delivers major stories, exclusive coverage and expert opinion. Sign up to get it every weekday morning.