Sydney already concentrates one-third of AI-related job offers in Australia. The National AI Centre has mapped these positions and discovered something unexpected: the jobs emerging around AI are not all engineering posts. They form eight distinct groups of activities, several of which involve translation work between tool logic and organizational reality. These hybrid positions exist, but 64% of offers requiring AI-related skills are concentrated in four metropolitan areas, while other Australian labor markets account for 36% of job locations.

The essentials

  • The Australian map of AI-related jobs reveals eight distinct activity groups, several of which are liaison positions between tools and organizations, not just technical roles.
  • Sydney, Melbourne, Brisbane, and Perth concentrate 64% of these job locations, according to the National Artificial Intelligence Centre (June 2025 report).
  • The report establishes a metropolitan concentration of offers without establishing its effects on the spread of AI to regional areas.
  • Daron Acemoglu and Pascual Restrepo show that automation can dissipate wage rents and amplify wage losses; Simon Johnson, with Acemoglu, more broadly argues that automation can improve or worsen worker outcomes depending on tasks and complementary employment: the Australian case illustrates this tension.
  • The open question concerns training outside metropolitan areas: online training alone is insufficient if no local employer can offer the corresponding job.

Eight activity groups, not two categories

The National AI Centre report challenges a widespread assumption. The debate on AI and employment often sets two camps against each other: those who see engineers creating jobs, and those who see algorithms destroying them. The Australian mapping shows that the reality on the ground is more nuanced.

The analyzed offers reveal eight activity groups. Some correspond well to expected technical profiles: developers, data scientists, machine learning researchers. But others describe different roles: project managers capable of framing AI deployments within business teams, change management specialists trained in model constraints, data quality managers in operational contexts. These positions do not require knowledge of how to train a model. They require the ability to connect what a tool can do with what an organization can absorb.

This distinction has practical consequences. A developer position can be filled by someone trained online, from any city. An AI-to-business translator position requires presence, interlocutors, an organization that tests and deploys. It is a job anchored in a place, in a structure, in a collective dynamic. And 64% of these location-anchored jobs are found in four cities.

Productivity creates rents for some workers and excludes others

Daron Acemoglu and Paschal Restrepo have developed a reading of automation that directly illuminates this Australian map. In their work, they show that automation can simultaneously increase overall productivity and reduce salary opportunities for some workers. The mechanism: automation dissipates the wage rents of intermediate positions while concentrating gains on workers already well positioned, close to decision centers and employers capable of adopting new technologies.

The concentration in four urban areas illustrates only the geographic distribution of offers requiring AI-related skills. A worker in Sydney or Melbourne who develops AI-to-business liaison skills quickly finds an employer, a salary, career progression. The report indicates that employers frequently seek hybrid skills for AI-related positions. A worker in Townsville or Ballarat who followed the same training would find themselves with a skill that has no local market to absorb it. Both AI training and job offers exist, but offers are heavily concentrated and data on training supply remains incomplete.

This asymmetry has its own logic that simple observation of job offers is insufficient to grasp. Companies that adopt AI do so first at their headquarters, where decisions are made and tech teams are already present. Regional branches, subsidiaries, small and medium enterprises far from major cities adopt more slowly, and often without the intermediate positions that facilitate integration. Metropolitan deployment first is a logic of organization, not a deliberate choice to widen inequalities. But the effect is the same.

The limits of online training

A frequent response to this type of divide is online training. Platforms offer modules on AI applied to project management, data analysis, and change management. The supply exists, it is often of good quality, and it is accessible from any territory.

The Australian report documents a different problem. The National AI Centre has identified data gaps on AI training and its adequacy to market needs. A worker trained in Adelaide or Darwin who seeks a hybrid AI position may have the required skills and find no local company capable of creating such a position. Online training remains necessary, but its effectiveness is questionable when employers able to absorb the skill are geographically concentrated.

This limit is known in other contexts. German and Swiss dual training systems are built precisely on coupling training with the presence of employers able to complete training with employment. The Australian model of regional AI training will need a similar element: not just modules, but companies or public institutions close enough for the skill to find an outlet. Jobs and Skills Australia, the public agency in charge of workforce forecasting, identified this need in its 2026 work, without a systematic response yet in place.

On how training transforms digital skills into real professional mobility, the case of universities confronted with generative AI illustrates a comparable tension: tools arrive faster than the institutions that train people to use them.

A different reading: city competition as leverage

Another reading of the phenomenon is possible, from a different family of ideas. Economists interested in urban growth dynamics note that the concentration of skilled jobs in a few metropolises is first and foremost a phenomenon of collective productivity. Dense cities create externalities: meetings, learning by observation, rapid recruitment, access to clients and partners. The concentration of hybrid AI positions in Sydney and Melbourne reflects these advantages, not a market failure.

In this reading, the response would not be to decentralize jobs, but to invest in secondary cities’ capacity to generate their own adoption dynamics. Canberra, Adelaide, or Hobart will not become Sydney, but they could develop sectoral niches where AI integrates with specific local industries: agribusiness, defense, aged care services, natural resources. These niches would create hybrid positions anchored in regional realities, without attempting to reproduce the metropolitan fabric.

The two readings are not incompatible. Johnson and Hazan are right to point out the risk of gains being captured by already well-positioned workers. Urban growth theorists are right to point out that concentration creates value that forced dispersal would destroy. The tension between the two resolves less through a choice of doctrine than through the concrete design of training and location policies.

The next ten years

Two trajectories are plausible for Australia between now and 2035, and available data do not yet allow choosing between them.

In the first, active policies are put in place: in-person training outside metropolitan areas, incentives for public and quasi-public employers to locate in secondary cities, conditionality of aid for AI adoption by businesses. In this scenario, the share of hybrid AI positions located outside the four major metropolises falls below the 64% documented today. Regional branches of the National AI Centre, if created, play the catalyst role: they train AI-to-business translators in direct contact with local employers, reproducing on a small scale the coupling dynamic that strengthens dual systems. On how other countries address the question of sharing digital capacity between territories, the African case of data sovereignty shows that the institutional question often precedes the technical question.

In the second trajectory, online training remains the main instrument, and no policy pushes employers to locate outside the four metropolises. Concentration is maintained or intensifies. Hybrid AI skills become a metropolitan privilege in fact: theoretically accessible from any territory, but valuable only in four cities. The social contract around skilled work shrinks geographically, without anyone having deliberately chosen to shrink it.

The signal to monitor is simple to identify, difficult to collect: the evolution of the share of AI-related job offers located outside the four metropolises. If this share stagnates or decreases by 2028, the consolidation trajectory is engaged. If it increases, active policies are starting to produce their effects. Data from the National AI Centre, if updated regularly, will allow reading this signal. The 2026 report lays its foundation; subsequent editions will show whether the trend has shifted.

Training and the presence of employers capable of creating corresponding positions are two distinct elements in the territorial distribution of opportunities. A worker trained without local professional prospects may have difficulty leveraging their skill in the labor market.

The question the market will not solve alone

The observed market concentrates AI offers in certain urban markets, without the report establishing the precise causes of this concentration. Sydney and Melbourne offer both: many companies adopting AI, many candidates trained to fill intermediate positions. This concentration is a market result, not an anomaly.

The problem is that this market result produces a geography of opportunities where other Australian labor markets, which accounted for 36% of offers requiring AI-related skills, found themselves excluded from these positions. Online training offers growing access to required skills. It does not necessarily offer access to employers capable of creating corresponding positions, especially outside major metropolises. And employers who could locate outside the four metropolises have limited incentives to do so, in an environment where concentration offers documented productivity advantages.

The response to this situation will not come from the market alone. It will come from decisions about what Australian public institutions do with AI adoption in their own regional services, about the conditions they attach to adoption aid in the private sector, and about the geography of their own recruitment. These decisions are political choices, not technical ones. The way automated tools are or are not shared among actors with unequal resources poses the same question in another domain: who decides access, and by what criteria?

Australia has a precise diagnosis. It now has to choose whether to let the market prolong the concentration or whether to build the conditions for broader diffusion. The report identifies eight clusters of AI-related work and observes strong territorial concentration of offers, while suggesting that interventions could broaden access to talent and AI benefits.

Sources

  1. National AI Centre, Australia’s artificial intelligence ecosystem: growth and opportunities (2026), https://www.ai.gov.au/news-and-insights/reports/australias-artificial-intelligence-ecosystem-growth-and-opportunities
  2. Simon Johnson and Eric Hazan, Human Capital for the Age of Generative AI, Project Syndicate, May 2024, https://www1.project-syndicate.org/commentary/ai-can-boost-wages-not-displace-human-workers-by-simon-johnson-and-eric-hazan-2024-05
  3. Jobs and Skills Australia, 2026 report, [available on the Jobs and Skills Australia website, without exact verified URL]