Last month, federal Treasurer Jim Chalmers released the 2026 Intergenerational Report, and its productivity assumption raised eyebrows. Despite productivity growth averaging only around a quarter of a per cent annually over the past decade, the IGR assumes average annual labour productivity growth of 1.2% over the next four decades. To justify this contentious productivity assumption, Chalmers cited AI’s potential. One of Australia’s leading economists on the economics of new technologies, RMIT’s Chris Berg, called this Australia’s “big bet” when I spoke with him on the Australian Taxpayers’ Alliance (ATA) livestream in late September.
You can also listen to the relevant parts of the livestream, which I featured on my Economics Explored podcast:
Chris Berg is on the optimistic side of the debate about AI’s potential, though he cautions against assuming we’ll see a major economy-wide productivity uplift. It’s undeniable that it has boosted productivity already in some areas. For instance, most new computer code is now partly generated by AI. Chris also points to recent AI advances in mathematics, biology and medicine. I went away to have a look at what advances Chris may have had in mind, and these AI advances include:
The debate over AI’s impact on productivity hinges on whether AI is a normal or abnormal technology. Is it an abnormal technology that will profoundly change the trajectory of our recent productivity performance? This is a difficult question to answer as of 2026. As Chris noted in our conversation, ChatGPT is about four years old, and if this were the internet, we’d be in 1972-73.
My fellow host of the ATA livestream, ATA Chief Economist John Humphreys, brought up the Solow paradox, based on an observation by MIT Nobel laureate economist Robert Solow in the eighties: computers were everywhere, yet they didn’t show up in productivity statistics. John argued that much of the internet’s benefit showed up in consumer surplus, the surplus consumers get because they can buy something much cheaper than they were willing to pay.
The internet generates large consumer surplus because many applications (e.g., Google, Facebook) are free to consumers but still deliver significant benefits. John wondered if this were the case with AI, which was fine from a consumer perspective, but if it didn’t generate a productivity uplift, it wouldn’t boost GDP and tax revenue to help reduce future budget deficits.
Chris countered John’s argument by noting that the impact of computers eventually appeared in productivity statistics, as it arguably did in the 1990s. However, the effect may have been overshadowed by microeconomic reform (e.g., see Dean Parham’s research, Australia’s 1990s Productivity Surge and its Determinants).
What is clear is that a productivity uplift from AI shouldn’t be taken for granted. We’ll likely see productivity gains, and many of us have already boosted our own productivity with AI, but we shouldn’t bank on AI getting Australia out of its productivity slump.
One constraint on AI’s potential economic gains is the huge capital investment (i.e. upfront costs) required. Recently, Anthropic announced that it would build a $32 billion data centre near Dalby on Queensland’s Darling Downs. This data centre would require enough electricity to power 1.5 million homes, plus large quantities of water for cooling.
In my view, AI’s power needs will wreck the Government’s desired energy transition, which it promoted in the IGR as putting downward pressure on power prices, a controversial proposition in itself. The Government is counting on AI to boost productivity, but this will require substantial investment in data centres and add to electricity demand. This raises the question of how quickly Australia can transition away from fossil fuels. It means the IGR is internally inconsistent. In my opinion, the AI revolution is incompatible with the rapid energy transition the Government plans.
In our conversation, Chris identified the worldwide shortage of computer processing capability (aka compute) as a major constraint on AI uptake. Furthermore, the added energy demand will require major capital expenditures in energy generation and network infrastructure. “Welcome to economics”, as Chris said. This compute shortage, and the implied need for huge investment in data centres and energy infrastructure, should reduce our anxiety about AI’s recursive self-improvement (i.e. AI building a better version of itself, and so on), AI taking over our economies and societies, and AI posing an existential threat. That is, Chris is less worried about the Skynet scenario (from the Terminator films) than others are, including John on the livestream.
In the livestream, Chris made a persuasive argument that AI is a trajectory-changing technology, but even he doesn’t see this as supporting the Government’s optimistic 1.2% long-run productivity growth assumption in the IGR. The baseline assumption should have been lower, given what appears to be a significant stepdown in productivity trend growth over the last decade. The baseline 1.2% productivity growth scenario should instead have been presented as an optimistic scenario. If we actually achieve it, we get a surprise and welcome upside to the economy and the budget.
Gene Tunny, Director, Adept Economics
Published on 13 October 2026. For further information, please contact us at contact@adepteconomics.com.au or call us on 1300 169 870.