Australia has spent the past year importing the physical substrate of the AI boom — data centres, power contracts, GPUs — while waiting for the productivity miracle meant to pay for it. On Monday, a senior Reserve Bank official said the wait is not over.

“Over time, supply will catch up with demand as data centre capacity and related sectors expand, easing current bottlenecks,” Sarah Hunter, Assistant Governor (Economic) at the RBA, told a UBS conference in Sydney. “If AI delivers broader productivity gains, economies will be able to produce more with the same inputs. As a result, the initial inflationary pressures from AI investment are likely to dissipate.”

The conditional is doing a lot of work. Hunter said there is limited evidence so far that AI has materially lifted productivity, or that it has had large effects on employment. What the RBA can see is the demand side: a boom in domestic data-centre investment that has added to price pressures in an economy already short of supply — one reason the central bank has raised rates five times this year, to an 18-year high of 4.85%.

That sequencing matters beyond Australia. The standard argument for AI is disinflationary: cheaper cognition, faster work, more output per hour. Hunter's account describes the opposite in the short run. Building the infrastructure requires labour, copper, transformers and electricity, all of them scarce; the productivity dividend arrives later, if it arrives at all. Central banks in importing economies face the same trade-off — an investment shock now, a supply shock maybe.

There are hints that something is shifting in the labour market, though not towards mass unemployment. Hunter pointed to “tentative signs of adjustment, through weaker job ad posting and employment outcomes in more AI-exposed occupations, and stronger growth in areas where the technology has had less of an impact, particularly for younger workers.” That describes entry-level work compressing while hands-on sectors hold up — a pattern several labour economists have flagged, now visible in a central bank's reading of the data.

The financial-stability warning is the sharpest part. If AI investment returns fall short of expectations, Hunter said, asset prices are exposed to sharp repricing “or even disorderly adjustments.” That is a polite way of describing what a market already nervous about AI capex looks like when the returns do not show up: the Nasdaq closed last week 11% below its record high, and the RBA is not the only institution raising the question. IMF chief Kristalina Georgieva said last week that any slowdown in AI-driven growth would be a factor for central banks; Fed officials have been pricing the same uncertainty into rate decisions.

None of this means the productivity gains are imaginary. Firm-level studies keep finding real time savings on specific tasks; the question is whether they aggregate — whether the gains survive process reengineering, organisational friction and the fact that much of the work AI speeds up was not the bottleneck. Australia's data are a useful test case precisely because the country is small, open, and heavily exposed to both the commodity and the compute side of the boom.

What to watch: whether RBA rhetoric shifts from “investment shock” to “supply-side gain” in coming statements; whether job postings in exposed occupations keep weakening while overall employment holds; whether data-centre construction costs start to fall, the mechanism Hunter says will ease bottlenecks; and whether equity markets treat a disappointing AI return year as a repricing event or a buying opportunity.

For now the ledger reads: costs are measurable, benefits are promised, and the central bank is keeping rates high until the second column fills in.