Shared ownership, public-private partnering and revenue apportionment for AI training infrastructure
This report has been prepared by Per Capita. It is informed by a roundtable process Per Capita conducted with a range of sector stakeholders, including research teams, industry and professional associations, sector organisations and individual experts. The report does not necessarily reflect the views of any of those participants, and no participant should be taken to endorse its analysis or recommendations.
Per Capita’s research team examined the plausible approaches to capturing the national interest in AI training infrastructure through the lens of the UK Government Office for Science’s AI 2030 scenario framework, and the recommendations set out represent one analytically defensible position intended to be tested in good faith against alternatives.
Investment in AI infrastructure continues to rise rapidly, with technology companies committing more than US$650 billion to AI infrastructure in 2026.1 Within the U.S. it represents the second largest annual investment in history, since the railroad network. While this interest is both cause for consternation and optimism it is apparent that this is a significant moment. It is anticipated (and the current construction pipelines suggest) that the relative level of investment in Australia will be (broadly) comparable, albeit smaller (as a percentage of GDP). The scale of investment is significant greater than any recent investment in domestic infrastructure notwithstanding the mining boom.
Notably interest exists in expanding Australia’s inference asset base but also in the strategically important compute infrastructure base. AI training infrastructure, often called ‘compute’ infrastructure, is much more concentrated in a small number of locations than the computing infrastructure used to run everyday AI applications. This is because building and operating this infrastructure requires significant investment, energy, and resources, many countries may not be able to host it at scale.
Australia is well positioned to become a major hub for AI compute infrastructure. It offers several advantages, including abundant renewable energy resources, political and regulatory stability, a highly skilled workforce, and a strategic location close to Southeast and South Asian markets.
The key question is not whether AI infrastructure should operate in Australia, but on what terms of its operation, so that the economic value generated flows back to Australians.