The short answer is yes, Australia has a real AI industry, and it is smaller, more concentrated and more research-heavy than the funding announcements suggest. The report, the third in NAIC and CSIRO's ecosystem series, identifies a sample of 1,533 AI companies headquartered in Australia: 1,121 private and 412 public, with 110 of the private companies founded in 2023 or 2024. The report is explicit that this is a sample built by structured search, not a registry, so treat the totals as a floor with known method rather than a census in the strict sense.
Where the companies are, and how small they run
Geocoding placed 858 companies, 68 per cent of those with mappable addresses, into 25 distinct clusters. Melbourne's CBD is the country's largest single AI concentration with 188 companies, ahead of clusters in Sydney, Brisbane and Perth. The specialisations are distinct enough to be useful: Canberra's cluster leans to government, security and defence, Perth's to resources, and Maroochydore on the Sunshine Coast has built a genuine niche in AI for media, social platforms and marketing, 11 companies, and on the report's figures the strongest concentration in that category of any cluster in the country.
The structural fact that shapes everything else: 85 per cent of private AI companies employ fewer than 50 people, a pattern the report calls "vibrant but potentially fragile". Private companies concentrate in business processes and support services (494 companies, 44 per cent of the private sample) and IT infrastructure and hosting (346, 31 per cent). The public companies in the sample are mostly AI adopters rather than developers, led by energy, raw materials and utilities (82 companies) and healthcare (77).
The research engine, and the gap at the end of it
Source: NAIC/CSIRO ecosystem report: 93,302 AI-related scholarly publications and 4,075 AI patents, 2015-2024, "nearly 23 research publications for every patent". Bars are proportional to the counts. Accessed 11 July 2026.
The growth numbers are real. AI patents rose from 170 in 2015 to 629 in 2024, close to a fourfold increase, and AI research went from 5.3 per cent of all Australian scholarly output to 11.6 per cent over the same decade. The report's honest caveat is the global denominator: worldwide AI publications rose 218 per cent in that window, from 206,160 to 655,454 a year, so Australia's share of global AI research fell from 2.6 per cent to 1.9 per cent even as its absolute output grew. Australia is running faster and losing ground, which is what happens when the track fills with sprinters.
The commercialisation ratio is the finding we would put on the minister's desk. Twenty-three publications per patent is not itself a scandal, patents are a narrow measure and plenty of good research becomes open-source code or method rather than intellectual property. But the report pairs it with a sharper number: Australia produces about 2 per cent of the world's AI research and, on the report's own analysis, 0.18 per cent of global AI patent inventions. The pipeline narrows precisely at the point where research becomes product, which is also the point where the 85-per-cent-under-50-staff structure bites: small companies do not run patent departments.
The labour signal
Demand-side data in the report points the same direction as its company counts: 1,532 organisations sought AI-skilled workers in 2024, 3.8 per cent of hiring organisations, up from 483 (2.7 per cent) in 2015, and AI requirements rose from 0.2 per cent to 0.9 per cent of all job postings. Concentration is the story inside the growth: 100 companies account for 58 per cent of all AI job postings, and inner Sydney, Melbourne, Brisbane and Perth account for 64 per cent of AI position locations. If you are hiring for or seeking AI work outside those four inner cities, you are operating in the thin third of the market.
Read alongside NAIC's separate SME AI Pulse, which put small-business AI adoption at 43 per cent for the December-February quarter with a February rebound to 44 per cent, the picture is consistent: broad, shallow adoption across the economy, deep capability in a few square kilometres of four cities, and a research base still looking for its commercial gearbox. That is a more useful map than either the boosters or the doubters usually draw, and it comes from the government's own count.