There is a fashionable certainty in some circles that Artificial Intelligence (AI) will simply arrive, transform the economy, create untold wealth, and leave Australia richer, smarter, and more competitive.
The data centres will be built, the models will train, the servers will hum, and the future will arrive regardless.
This is the language of the fait accompli – AI is coming whether we like it or not, so the only question is how quickly we can embrace it.
That confidence is misplaced.
AI is not a software update. It is an industrial process of extraordinary energy intensity. The servers that train and run large models do not tolerate intermittent power. They demand constant, reliable, dispatchable power at scale. Without it, the servers are constrained, the training runs fail, investment moves elsewhere, and the promised Australian AI boom becomes someone else’s industrial story.
The shortcomings of our infrastructure are enormous, and it is likely they are not being addressed with the seriousness the opportunity demands.
Start with the physics of demand.
A modern AI data centre is not a conventional commercial building with variable energy load. It is closer to a continuous industrial process that must be fed electricity every hour of the day. The Australian Energy Market Operator (AEMO) estimates that data centres currently consume roughly 2 per cent of grid-supplied electricity, with that figure expected to rise towards 6 per cent by 2030 and potentially much further in the decades beyond. Internationally, the International Energy Agency expects global data-centre electricity consumption to roughly double by the end of the decade.
AI is therefore arriving at precisely the moment Australia is already attempting an enormous transformation of its electricity system.
Unlike households or many factories, these facilities cannot simply wait for the wind to blow or the sun to shine. When power quality or supply falters, the economic consequences are immediate and severe. Dispatchable baseload energy is the skipping rope of the modern grid. Picture two people turning a long rope at a steady, unvarying cadence while others jump in the middle. The rope’s speed and consistency determine whether the activity continues smoothly or ends in a tangled failure. If the turners slow, accelerate unevenly, or stop, the jumpers fail. Intermittent renewable energy, however cheap their marginal cost is on a blowy afternoon, cannot perform that role alone. Batteries and pumped hydro can smooth short gaps, but they don’t create continuous high-volume energy for the multi-year training runs and always-on inference loads that define competitive AI infrastructure. Relying on weather-dependent generation is an invitation to catastrophic economic disruption – lost investment, deferred projects and the migration of compute capacity to jurisdictions that can keep the rope turning.
Australia does not currently possess the firm capacity, nor a credible pathway to grow it at the scale and speed required. Coal plants are closing or being constrained. Gas is treated as a transitional embarrassment rather than a strategic asset. Large-scale nuclear remains politically radioactive in more than one sense. The result is a system that is increasingly optimised for the integration of variable renewable energy, and is poorly suited to AI, which requires high reliability. It is entirely reasonable to conclude that under present settings, Australia lacks both the existing capacity and growth development required to support a serious domestic AI industry.
One technology that must be seriously considered is the small modular reactor (SMR). SMRs offer one possible means of supplying this firm power. Rather than constructing a single conventional reactor of around a gigawatt or more, an SMR system uses substantially smaller generating units, typically no greater than 300 megawatts of electricity. Their attraction lies partly in their modularisation as major components can be manufactured under controlled factory conditions, transported to site and assembled as part of a larger generating precinct. The underlying technology remains nuclear fission which means what changes is principally the scale and method of construction.
Both large and small reactors increasingly use modular construction techniques where major components are assembled in factory environments before being shipped to site. This improves quality control, shortens on-site construction, and can reduce costs. Components or entire modules are designed for factory fabrication under strict quality control. Designers aim for serial production to capture economies of scale similar to those which have long been achieved in aerospace. These modules are then shipped and assembled on site, potentially shortening construction timelines. Many designs allow phased rollout, so capacity can be added in stages, which reduces financial risk and allows flexibility to match demand growth.
The usual objection is safety, and it deserves to be addressed seriously. Many advanced SMR concepts are designed around passive safety systems which rely on physical processes such as gravity, natural circulation, and convection rather than requiring powered pumps or constant operator intervention during an emergency. Some designs also envisage much longer fuel cycles than conventional reactors, in certain cases extending for decades between refuelling. That does not mean such reactors operate for decades without maintenance, however, it does illustrate how far modern reactor design has moved from the popular image of nuclear technology during the 20th Century.
Many advanced SMR designs are expected to use High-Assay Low-Enriched Uranium (HALEU). HALEU enables smaller cores, longer operating cycles and improved performance. At present, commercial-scale HALEU production outside Russia and China remains limited. There is another advantage that deserves greater attention, which is energy density. A group of modular reactors can produce very large amounts of electricity from a comparatively small industrial footprint, potentially reducing the need for new transmission corridors, land clearing and widely dispersed generation. Existing coal-fired power stations such as Tarong are particularly interesting because they already possess transmission connections, industrial land, and much of the infrastructure required for large-scale generation, making them obvious candidates for future energy precincts incorporating SMRs and the industries that require their power.
Australia sits on some of the world’s largest and highest quality uranium resources. Olympic Dam, in South Australia, alone contains an extraordinary concentration of uranium and other valuable minerals. We are therefore not lacking raw material, rather, what Australia lacks is the downstream capability, such as conversion, enrichment and fuel fabrication. Australian companies such as Silex have developed advanced laser enrichment technology with HALEU potential, but commercial roll out is occurring primarily through overseas partnerships. The geological resources are suitable however, the industrial and regulatory conditions are not yet in place.
Even if the policy setting were changed tomorrow and SMR designs licensed, the physical plants would still need to be built. That will require rebuilding the skilled construction capacity that Australia has allowed to decline. Three trades stand out as particularly critical for the civil and nuclear-grade infrastructure involved. These include nuclear welders, form workers and steel fixers.
Nuclear welding is not ordinary fabrication. It demands the highest standards of process control, non-destructive testing, documentation and metallurgical understanding. Welds in primary systems or containment structures must meet strict codes because the consequences of failure are unacceptable. Form workers create the exact temporary moulds into which concrete is poured for foundations, containment buildings and massive structural elements. Steel fixers place and secure the reinforcing steel that gives those concrete structures their strength and stretch under load, seismic events and thermal stress. All three trades operate in an environment of rigorous quality assurance. Errors are expensive, and shortcuts are career-ending.
Now go and ask any high school career advisor to tell you all they know about these jobs and what career pathways a school leaver would need to follow in order to secure long-term employment in these roles. Being a form worker or a steel fixer is not a ‘shiny’ job, but without them major construction projects cannot proceed.
Automation may reduce the amount of bespoke welding undertaken onsite, and modular construction may shift a greater proportion of work into controlled factory environments. Yet those factories will still require highly skilled people, and the reactors still need to be transported, assembled, connected, inspected and maintained. Data centres themselves also require enormous quantities of electrical, mechanical, and civil construction. The industrial workforce does not disappear simply because some of its tools become more sophisticated.
Australia already faces severe shortages across welding and construction trades. Industry bodies have warned of a potential shortfall measured in the tens of thousands of welders by the end of the decade. Shortages that already constrain defence, shipbuilding, and major infrastructure projects, including elements of the Aukus submarine program. Formwork and steel-fixing capacity is similarly stretched by the volume of ordinary commercial and civil work, let alone the specialised, high-specification requirements of nuclear-grade construction. Training pipelines have not kept pace with retirements or the growing complexity of modern projects. Apprenticeship completions in critical metal and construction trades have lagged demand for years. Elevating the status, training pathways, and pay of these occupations is not about romanticising Australia’s industrial past. It is essential to building the energy system that any serious AI strategy requires.
And this is ultimately the larger point. Australia risks approaching AI in the same manner that it has approached too much of modern industrial policy, which is, assuming that because the technology exists somewhere in the world, we will inevitably participate in its prosperity. We continue to enjoy the comfort of policy targets and enormous mineral exports while too often allowing the difficult business of turning raw resources into energy, infrastructure and advanced industry to occur somewhere else.
None of this is inevitable. The immediate barriers are as much political and legislative as they are industrial. Australia still prohibits nuclear power generation, but policy can change, and regulatory conditions can be modernised. Training places can be expanded and made more attractive. Uranium resources can be developed further down the value chain. SMRS or other low-emissions technologies can be given a genuine pathway. Changing the law would not create an industry overnight, but neither can an industry emerge while the technology itself remains prohibited. None of these steps are automatic, and the window is not infinite. Other countries are already securing power purchase agreements, building gas and nuclear capacity and training the workforces needed to host the next generation of computing infrastructure.
If Australia continues to treat dispatchable power as an afterthought, nuclear as untouchable, and high-skill trades as lesser careers, then the AI boom will indeed be a fait accompli – just not here.
The servers will be built where the electricity is reliable, a capable workforce exists, and the policy settings are stable. The skipping rope will keep turning somewhere else.
The question is whether Australia still has the will to grab the end of it.
