Every fresh wave of artificial intelligence lands on the same unglamorous bottleneck: electricity. Training frontier models and running the data centres that serve them consumes staggering amounts of power, and the companies building that capacity have started shopping for firm, round-the-clock generation wherever they can find it. In the United States, that hunt has pulled nuclear energy back from the margins, with technology giants signing deals to restart mothballed reactors and bankroll small modular designs. Australia is heading in a conspicuously different direction.
A new piece from Environmental Leader frames the local picture bluntly, arguing that Australia is building its AI grid without a nuclear option. The country’s compute expansion, in other words, is being wired to solar, wind, gas and large-scale batteries rather than to the reactors increasingly favoured by hyperscalers abroad. It is partly a choice and partly a constraint, because nuclear power generation has been prohibited under federal law since the late 1990s.
Why nuclear is off the table here
The legal picture is unusual by developed-world standards. Commonwealth legislation bans nuclear power stations outright, and several states layer their own prohibitions on top. That means even if an AI operator wanted to follow the American playbook and strike a power purchase agreement with a reactor, there is no fleet to buy from and no near-term pathway to build one. The question became briefly political at the last federal election, when the Coalition campaigned on lifting the ban and constructing a handful of government-owned nuclear plants on the sites of retiring coal stations. That platform was rejected at the ballot box, and the re-elected Labor government has stuck with a renewables-led strategy anchored by firming gas and storage.
So the AI build-out is colliding with an energy system already in the middle of its own transition. Coal plants are closing on a published timetable, the share of renewables on the National Electricity Market keeps climbing, and the grid operator is racing to add transmission and storage fast enough to keep the lights on through the changeover. Into that already stretched system arrives a new and hungry customer.
The scale of the demand
Data centres are not a rounding error on the grid. Industry and market forecasts have repeatedly warned that the sector’s electricity appetite could roughly double over the back half of this decade, with AI workloads doing much of the pushing. A single large AI campus can draw as much power as a small city, and operators want that supply to be constant, which is precisely the attribute renewables struggle to guarantee on their own. That tension sits at the heart of the Australian debate: can a grid leaning heavily on variable solar and wind reliably feed a load that never sleeps?
The optimistic case says yes, provided the firming is done properly. Australia has abundant sun and wind, a fast-growing battery pipeline, and a gas fleet that can cover the gaps while storage scales. Proponents point out that data centre operators are among the most motivated buyers of clean energy on the planet, because their own customers and investors demand it, and that their long-dated, high-volume contracts can actually underwrite new renewable and storage projects that might otherwise struggle to reach financial close. On this view, AI demand becomes an accelerant for the energy transition rather than a threat to it.
The sceptical case is harder-edged. Critics argue that bolting enormous new loads onto a grid mid-transition risks pushing up prices for households and businesses, straining reliability during the evening peak when solar drops away, and leaning on gas for longer than anyone would like. They note that the firming promises depend on batteries and transmission arriving on schedule, and Australia’s recent track record on big infrastructure timelines is mixed at best. Some in the energy sector quietly concede that without a dispatchable, zero-emissions baseload option, which is the role nuclear plays elsewhere, the maths gets tighter the more compute the country tries to host.
What it means for Australia
The stakes are not abstract. Australia has set itself up to be a serious destination for data centres and, increasingly, for sovereign AI capacity that keeps sensitive workloads onshore. That ambition only works if there is power to run it, delivered at a price that does not make the country uncompetitive against regional rivals such as Singapore, Japan and the United States. Energy is fast becoming the deciding factor in where the next generation of AI infrastructure gets built, and Australia is effectively betting that renewables plus storage can carry the load that others are handing to reactors.
There is a genuine upside if the bet pays off. Cheap, abundant clean power has long been floated as Australia’s natural advantage in a decarbonising world, and hosting energy-intensive AI compute is one way to turn sunshine and wind into export-grade economic value without digging anything up. It also keeps the industry aligned with the country’s climate commitments rather than reviving a fuel source that would take a decade or more to stand up even if the ban were lifted tomorrow.
The risk is equally real. If firm supply lags behind demand, the first casualties are ordinary bill-payers and the reliability of the grid they share with these hyperscale tenants. Households already nervous about power prices will not take kindly to the idea that server farms running chatbots are competing with them for electrons. Managing that political sensitivity, and being seen to add generation rather than simply consume it, will be central to keeping the social licence for the whole AI build-out intact.
What’s next
Watch three things. The first is the pace of new firming capacity: battery projects reaching final investment decision, pumped hydro progress, and whether transmission upgrades land anywhere near their promised dates. The second is how data centre operators structure their power deals, because large, long-term renewable contracts that bring new supply online are a very different proposition from operators simply drawing down existing capacity. The third is the policy conversation itself, which is unlikely to stay settled; expect the nuclear question to resurface whenever reliability or prices wobble, even if the near-term answer remains no.
For now, Australia has made its choice by circumstance as much as by design. The AI grid being built here will run on renewables, gas and storage, and the country is about to find out, in real time and at real scale, whether that is enough.
Sources: Environmental Leader, via Google News.
