AI-related hiring is concentrated in a small number of organizations and metropolitan areas. According to a publicly released Australian government report in 2026, 100 organizations accounted for 58% of AI job postings in 2024, and two-thirds of these positions were located in four cities. AI adoption depends less on software availability than on local capacity to train, recruit, and integrate skills—a capacity that is distributed very unevenly.

The essentials

  • In 2024, 100 Australian organizations accounted for 58% of AI job postings, according to the Australian government.
  • Only 4% of small non-innovative enterprises used AI, compared to 19% of innovative ones—a gap of one to five separating two different worlds.
  • Geographic concentration in four metropolitan areas is established by Australian data. The data shows concentration of AI hiring in hubs and among a small number of employers; they do not demonstrate that productivity gains accumulate there causally.
  • Current concentration may complicate AI diffusion to SMEs and regions; the report suggests that targeted interventions could broaden access, without predicting certain lock-in.
  • Avenues exist: shared infrastructure, advice for SMEs, distributed training. Some fall within strategic recommendations; others, such as AI Adopt Centres, are already operational.

Four cities, a hundred companies, and everything else

Sydney, Melbourne, Brisbane, and Perth absorb two-thirds of Australian AI positions. These metropolitan areas also concentrate corporate headquarters, major universities, and capital. The degree of concentration is striking: one hundred organizations, out of the millions that make up the Australian economic landscape, capture more than half of all AI sector job postings.

This figure says something precise about the nature of AI adoption. Software is available everywhere: a text automation or data analysis tool downloads as easily in Darwin as in Sydney. What does not download is the human ecosystem that makes it useful—the data scientists to configure it, the managers to deploy it, the lawyers to manage its risks, the public procurement officers capable of writing a coherent specification. These skills form and are recruited where similar skills are already present. Concentration calls for concentration.

Australia is not an isolated case. Quantum computing follows the same polarization logic, with a few university and industrial hubs capturing most global investment and talent. AI’s geography reproduces innovation’s geography in general: dense in nodes, hollow in margins.

The gap between innovative and non-innovative firms reveals the real obstacle

The Australian report distinguishes between two categories of enterprises: those already innovating, and those that are not. Among small innovative enterprises, a significant proportion used AI in 2024. Among non-innovative ones, this rate was notably lower. In 2024–25, small enterprises active in innovation declared using AI more often than others.

This gap reflects a precise reality. In 2024–25, small enterprises active in innovation declared using AI more often than others; this correlation does not prove the causal mechanism advanced. For the others, barriers remain high: integrating an AI tool into a business process requires a culture of experimentation, piloting skills, and tolerance for uncertainty that some small structures have not developed.

This mechanism extends a broader trend documented by employment statistics. Job growth concentrates in high value-added sectors, leaving a majority of enterprises in activities where productivity gains remain weak. AI, therefore, risks accelerating this bifurcation rather than correcting it.

Rent dissipation and its concrete effects

Economist Daron Acemoglu has formalized a useful mechanism for reading this data. In his recent work on automation, he shows that when a technology improves the productivity of a category of workers or enterprises, it can simultaneously dissipate wage advantages—the “rents”—of more vulnerable workers without access to it. The result can be paradoxical: aggregate productivity rises, but distribution of gains collapses to the detriment of those unable to reposition themselves.

Applied to Australia, this framework illuminates the situation of SMEs outside major metropolitan areas. SMEs can suffer indirect effects of AI adoption without capturing direct benefits. In some contexts, SME employees might see their salary position affected by competition from better-equipped enterprises. These dynamics could lead to a progressive degradation of SMEs’ relative position.

This reading opposes a more optimistic vision, notably carried by economists close to the Schumpeterian school, such as Philippe Aghion. From this perspective, initial concentration of adoption among the most dynamic enterprises is a normal phase in the diffusion of a generic technology. Innovative enterprises blaze the trail, establish uses, train the first practitioners, and these skills then diffuse to the rest of the productive fabric via employee mobility, imitation of practices, or tool cost reduction. Delay is not a failure: it is the normal time for transition.

Australian data allow testing both readings. They show persistent concentration of AI hiring since 2015. They also signal that the gap between innovative and non-innovative enterprises concerns capacity to innovate generally, not only AI. Enterprises not active in innovation declare using AI less often; the causes of this gap are not established by this data, and the effectiveness of outside intervention remains to be demonstrated.

AI productivity gains and the question of their capture

Distribution of productivity gains merits attention when AI is adopted massively in large organizations. The modalities of distributing these productivity gains between capital and wages remain poorly established for AI, although a tendency to orient gains toward capital has been observed in previous automation waves.

In Australia, this concentration in a few large organizations compounds risks of gains being captured by capital, though the exact mechanism is not established. Large organizations adopting AI present a structure favoring orientation of gains toward capital returns, though this mechanism is not quantified by available data. Employees of these organizations may benefit if their employer shares part of the gains as salary. But nothing in market mechanics guarantees this distribution, especially when the local labor market offers few alternatives to candidates.

SMEs find themselves in a doubly disadvantaged position. They adopt AI less often and therefore can less often draw potential productivity gains from it. They may face intensified competition from enterprises adopting AI, though this effect is not measured by available data. These dynamics could exert pressure on margins and salary attractiveness.

Risks for remote regions

The map of AI job postings in Australia resembles the map of its major metropolitan areas. Northern Territory, Western Australia outside Perth, Queensland outside Brisbane figure little in the AI job market; the reasons for this concentration remain to be precisely documented.

This configuration has concrete consequences for demographic dynamics. Young graduates trained in AI at regional universities migrate to Sydney or Melbourne, where positions exist. Regions lose their talent at the moment when they would need it to conduct their own digital transition. This circle is not inevitable, but it closes quickly if nothing interrupts it.

According to the IMF, slowdown in adoption of new technologies figures among several structural factors in Australia’s weak productivity. Geographic concentration of AI is an additional symptom. Maintaining the status quo regarding unequal territorial adoption could allow regional inequalities to persist or worsen, according to several forward-looking scenarios.

What institutions to avoid lock-in

Australian data pose a precise question for the forward outlook: what institutions will enable SMEs and remote regions to capture AI benefits and broaden access and diffusion of AI beyond dominant centers.

Several mechanisms have proven themselves in other technological contexts. Shared resource centers, agencies, or consortiums that pool access to AI expertise for SMEs unable to recruit alone, functioned in the digital transitions of the 2010s in several Nordic countries. The logic is simple: a twenty-person SME cannot finance a full-time data scientist, but can finance a third of a position in a pooled structure. This approach assumes public-private coordination that Australia has experimented with in other sectors but little developed for AI.

Distributed training constitutes a second lever. Australia has a network of technical colleges and vocational education, the so-called TAFE system, present in regions. These institutions could become entry points for short courses in applied AI, targeting SME managers, accountants, and logistics managers—that is, those who decide whether to adopt a tool. This type of short, practical training presents low cost and potentially high multiplier effect.

Advice for SMEs is often underestimated. Non-innovative enterprises underestimate what they could automate, overestimate implementation complexity, and lack references in their sector. Subsidized advisory schemes, on the model of what some British or German public agencies implemented for digital transition, could reduce this informational entry cost.

These avenues are hypotheses that Australian data make plausible, and that ongoing experiments will allow testing in coming years. Two signals merit monitoring. The first is the evolution of AI adoption rate in non-innovative SMEs: if it stays at 4%, the gap widens; if it rises, diffusion is working. The second is the geography of AI job postings: if it diversifies beyond the four metropolitan areas, rebalancing institutions produce effects.

Timeline remains the open variable. Dominant hubs could attract new talent and develop new expertise, a process documented in other technological contexts. Absence of active diffusion policy could allow persisting or widening observed gaps. The tools exist, some experiments show results, and the choice to deploy them belongs to Australian public institutions and economic actors.


Sources

  1. Australian Government, Australia’s artificial intelligence ecosystem: growth and opportunities, June 2025
  2. Australian Bureau of Statistics, Characteristics of Australian Business 2024-25
  3. Daron Acemoglu, Automation and Rent Dissipation: Implications for Wages, Inequality, and Productivity (2025), MIT Economics