The Albanese government has embraced the extraordinary expansion of data centres as evidence that an artificial intelligence revolution will ride to the rescue of Australia’s weak economy.
More than $130 billion is slated for data centres and supporting infrastructure by the end of the decade. It sounds transformational, but look beyond the headlines and the economic payoff is vastly smaller than promised.
Goldman Sachs estimates all the construction will add just 0.3 percentage points to GDP growth, before actually becoming a drag as the build-out slows down. Across the entire time, it amounts to roughly 1 per cent of GDP. To put the paltry benefit in context, at its height, mining investment reached 7 per cent of GDP.
To make matters worse, Treasury estimates that two-thirds of this expenditure is absorbed by imported servers, storage and the astronomically expensive chips that power AI. The mining boom also required imported machinery, to be fair, but it left Australia with rivers of iron ore, coal and gas that generated massive royalties and tax revenues for decades past and decades to come.
When the data centres are finished, what exactly will Australia sell? Treasury concedes the answer isn’t clear. There’s no guarantee of an export payday to follow.
Instead, Australia will supply the dirt, water, electricity and labour. Foreign tech giants will supply the hardware, control the AI models themselves, and send much of the profits earned offshore.
We are essentially volunteering to be the dumb landlord.
Even the construction benefits need to be weighed against reality. Australia isn’t suffering from a surplus of idle tradies or cheap building materials that need to be put to use. We’re in the grip of a housing crisis and a severe infrastructure backlog alongside it.
Goldman Sachs estimates all data centre construction will add just 0.3 percentage points to GDP growth, before becoming a drag as the build-out slows down (Jim Chalmers above)
More than 70 hectares of vacant green space beside Sydney Airport is set be transformed into an AI data centre
An AI-led construction boom doesn’t fall from the sky, it cannibalises already strained resourcing. Data centres will compete directly with apartment blocks for electricians and with hospitals, schools and transport projects for concrete, copper and electrical equipment. The boom might flatter the investment data, but it risks deepening the shortages and bottlenecks Australians are suffering from already.
Once the dust settles, the employment dividend also evaporates. These facilities are essentially ghost towns: enormous, humming sheds requiring thousands of labourers to build them, but vastly fewer people to operate them afterwards.
The Australian Energy Market Operator expects data centres to gorge on electricity, surging from 5 terawatt-hours today to 34 less than a decade from now.
That’s a sevenfold increase, accounting for 13 per cent of electricity supplied through the National Electricity Market precisely at the time coal-fired power stations are closing and replacement generation and transmission is struggling to keep pace.
The thirst for water is equally alarming. Sydney Water warns that proposed data centres could consume up to 250 million litres a day by 2035, roughly one-sixth of Greater Sydney’s current supply. Because large-scale recycled water infrastructure takes many years to build, these facilities could draw heavily on existing drinking water systems while recycled infrastructure is trying to catch up.
Switching to waterless cooling merely shifts the burden back to the overtaxed electricity grid, so that’s no solution either.
None of this means Australia should ban data centres or hide from the AI revolution. Of course not. But hosting foreign servers isn’t a sovereign AI strategy worth very much.
A data centre planted on Australian soil doesn’t make its data or AI models Australian. Unless we demand enforceable conditions alongside playing the role of landlord, we risk socialising the infrastructure costs of Silicon Valley giants.
Andrew Charlton has warned that simply building data centres doesn’t mean Australia captures the economics of AI
That would deserve a slow clap for incompetence.
Governments really do need to stop rubber-stamping these projects just because the price tags look good in a media announcement.
Operators must be forced to fund their own additional power, water and grid connections. Local communities shouldn’t be left to underwrite private tech monopolies making more money.
Furthermore, Australia should extract guaranteed computing access for local start-ups and researchers as the price of hosting the rollout.
Let’s be frank here: who trusts politicians to get this right, especially Labor? What, with their abundance of entrepreneurial experience and business acumen?
There is one source of hope: Andrew Charlton. The Oxford-trained economics PhD is also a self-made multimillionaire, having built and sold his own analytics business before entering parliament. More importantly, he understands the dangers here. Charlton has warned that simply building data centres doesn’t mean Australia captures the economics of AI and wants operators to provide Australian start-ups and researchers with computing access on favourable terms.
It says everything about Labor’s factional system that Charlton remains an assistant minister while lesser factional operators occupy Cabinet. Albo has at least made him Cabinet Secretary and placed him at the centre of the government’s AI strategy, effectively asking him to babysit Jim Chalmers through the revolution in the hope that he can save the Treasurer (and the rest of us) from Chalmers’ incompetence.
Chalmers calls AI the biggest economic transformation of our lifetime. If that’s true, which it probably is, the government can’t afford to confuse erecting sheds with capturing the real economic benefits.
Get this wrong and Australia will absorb all the physical strain of the AI revolution, only to rent the actual technology back from overseas.
