The Reserve Bank of Australia’s Sarah Hunter says AI investment is already adding demand and some price pressure, while broad productivity gains and job-market effects remain uncertain.

Artificial intelligence has begun to change the composition of investment and demand, but the productivity revolution often promised by its advocates has not yet appeared clearly in economy-wide data. That was the central tension in a speech delivered in Sydney on 9 November 2026 by Sarah Hunter, Assistant Governor (Economic) at the Reserve Bank of Australia (RBA).

Hunter’s address, “What Happens to the Economy as ‘Intelligence’ Gets Cheaper?”, argues that the transition may initially add to demand faster than supply can respond. Over time, expanded capacity and genuine productivity improvements could ease those pressures. The sequence, size and timing are not settled. Reuters’ independent report on the speech similarly highlighted limited evidence of aggregate productivity gains or widespread job losses so far.

The distinction matters: Hunter was outlining a framework for assessing a fast-moving technology, not announcing a new interest-rate decision or promising a particular inflation outcome. Much of the speech’s discussion of future effects is explicitly conditional.

Adoption is moving faster than measured productivity

The RBA’s business liaison observations show a sharp reported shift in adoption between 2025 and 2026. In the chart accompanying the speech, the share of firms reporting no AI adoption falls from about one-third to almost zero; limited use becomes the most common category, while moderate adoption also rises. Deep or transformative use remains a much smaller share, increasing only modestly, from roughly 8 per cent to 10 per cent.

These are reported adoption levels among firms in the Bank’s liaison evidence, not a census of every Australian business and not proof that AI use has improved output. The speech separates the spread of tools from their economic payoff. Workers may already complete some tasks more quickly or handle tasks that previously were impractical, Hunter said, citing examples such as document review and customer-service responses.

But the RBA finds limited evidence so far that AI has materially lifted aggregate productivity. Its accompanying survey chart indicates that about 35 per cent of firms reported no productivity gains; approximately 30 per cent reported minor gains, 23 per cent moderate gains and around 11 per cent significant gains. These figures are business-reported experiences, not a measured national productivity contribution. They are consistent with an early implementation phase in which firms are experimenting, investing and reorganising rather than harvesting fully embedded gains.

Hunter compares that transition with the “productivity J-curve” associated with earlier general-purpose technologies. Installing a new technology is not the same as redesigning a business around it. Electricity delivered its largest gains when factories changed their layouts and production processes, not merely when steam engines were replaced by electric motors. AI could likewise require complementary investment, new workflows, staff training and risk controls before productivity statistics capture durable benefits.

That lag helps explain why forecasts vary so widely. Hunter cited estimates for AI’s effect on annual productivity growth over the coming decade ranging from more than 2 per cent to just 0.1 per cent. The range reflects different assumptions about how deeply AI changes tasks and industries; it is not an RBA forecast or a claim that either endpoint is assured. The speech also notes a study suggesting markets expect a substantial software-engineering productivity effect, but expectations in asset prices are not proof of economy-wide realized gains.

Investment is the visible economic channel

While the supply-side payoff remains uncertain, spending on the infrastructure needed to develop and run AI is already visible. Data centres require buildings, servers, advanced semiconductors, electricity generation and transmission. Hunter said the RBA’s estimates of realized Australian data-centre investment had increased tenfold over three years, though from a very low base.

The composition of that expenditure matters. Much of the equipment is imported, so investment can raise domestic demand without contributing an equivalent amount to Australian domestic production. The RBA nevertheless reports modest indirect spillovers to construction, engineering and energy infrastructure through its liaison program and other data. The Bank also points to overseas investment and stronger machinery and equipment exports in technology-producing economies as part of the wider cycle.

Some effects are potential rather than observed. The speech says AI infrastructure may raise demand for electricity, energy resources and critical minerals, while Australian data centres might eventually export computing capacity. Hunter cautions that current commodity effects are difficult to separate from broader energy-transition demand, and the possible export channel is a future possibility, not an established outcome.

Household spending could also respond if people expect future income or wealth gains. Hunter describes a scenario in which households consume more now, saving less or borrowing more in anticipation of higher future productivity. She cited RBA staff estimates that about 5 per cent of Australian households’ financial wealth is invested in AI-related companies, directly or through superannuation. In the United States, AI-company share prices had roughly quadrupled since the end of 2022, according to the speech’s comparison. Neither statistic means all households share those gains equally: the RBA stresses that financial returns have so far been concentrated among a relatively small group of firms and owners.

Why prices may rise first—and later behave differently

Hunter’s near-term inflation argument is about timing. If infrastructure investment and associated spending lift demand while available supply is constrained, the imbalance can put upward pressure on prices. She said this pattern appears consistent with current evidence: surging global investment has contributed to capacity pressures in supply chains for AI-enabling products, including semiconductors, with some upstream costs passing through to downstream goods in advanced economies.

The observed price effects described in the speech are narrow. Price increases are concentrated in a limited set of consumer-electronics categories that have small weights in consumer price indexes. Some RBA liaison contacts also say data-centre construction is adding demand for already-stretched skilled construction trades. That is not the same as evidence that AI has broadly raised Australian consumer-price inflation; the speech presents these as specific pressures and emerging channels.

Over time, the supply response could change the balance. More data-centre capacity and related infrastructure may relieve bottlenecks. If AI also produces broad productivity gains—more output from the same inputs—the initial inflationary effect of investment could fade. Hunter’s longer-run assessment is not that AI must continually lower inflation: stronger productive capacity could be accompanied by stronger demand, leaving the technology’s long-run effect on inflation broadly neutral, as with prior general-purpose technologies.

There is substantial uncertainty in both directions. The speech says it is too early to quantify the magnitude or timing of AI’s effects. AI investment could be larger or more persistent than expected; productivity could arrive slowly, unevenly or not to the extent some projections imply. Near-term price pressure and a possible later easing are analytical scenarios grounded in supply and demand channels, not a precise forecast path.

Employment signals are mixed and tentative

On jobs, Hunter finds little evidence of large aggregate employment effects across advanced economies to date. Australian businesses in the RBA’s liaison program mostly said AI had not changed their headcount, and most did not expect a headcount impact over the following year. In the speech’s chart, around three-quarters reported no headcount change in the past year; roughly 12 per cent reported slower hiring or unfilled roles, about 8 per cent reported cuts, and about 6 per cent reported increases.

Those aggregate figures do not rule out disruption for particular workers or occupations. Hunter points to tentative signs of weaker job advertisements and employment outcomes in more AI-exposed occupations, alongside relatively stronger growth in less-exposed areas, particularly for younger workers. She emphasizes that the differences are modest and that it is too early to tell whether they reflect AI adoption or other forces. This caution is essential: a correlation between occupational exposure and hiring trends cannot by itself establish that AI caused the change.

Exposure also does not mean replacement. Jobs and Skills Australia estimates cited by Hunter classify about 4 per cent of Australian workers as highly exposed to AI automation, compared with 31 per cent highly exposed to AI complementing or augmenting their work. These figures describe potential task exposure, not actual AI use, certain job losses or the eventual design of roles. A tool can automate some tasks, assist with others and create new responsibilities in the same occupation.

The adjustment could still be difficult for people whose tasks or positions shrink, even if total employment remains resilient. The RBA notes that concentrated gains and uneven transitions could affect incomes and consumption. It says it will monitor job markets, wages and income distribution as adoption progresses rather than infer a settled labour-market outcome from early indicators.

What this does—and does not—say about rates

The speech is an economic analysis, not a policy announcement. It does not set a new cash rate, provide a decision timetable or commit the Monetary Policy Board to a response. Reuters reported that the RBA had raised rates five times that year to 4.85 per cent, but that policy context should not be misrepresented as a new decision made in Hunter’s address.

For monetary policy, the important question is how demand, supply capacity, inflation and employment evolve together. If investment adds to demand before capacity catches up, that may matter for inflation assessment. If productivity later expands supply, the balance may look different. The speech also discusses a possible rise in the neutral interest rate as firms compete for global savings, but Hunter calls the scale hard to quantify and says the Bank will continue assessing the evidence.

There are financial-stability considerations too. Hunter warned that rapid AI advances could worsen cyber and operational risks, while disappointing returns on the investment boom could expose equities and corporate debt to sharp repricing. These are risks for ongoing monitoring, not evidence that a market correction is imminent. The breadth of the speech reflects the RBA’s effort to map multiple channels rather than reduce AI’s macroeconomic consequences to a single productivity number.

The practical takeaway is a two-part one: investment and adoption are already changing spending patterns, but promised aggregate productivity gains and large employment effects have not yet been established. For households, businesses and policymakers, the distinction between what is measured now and what is projected for later is the key to reading the speech accurately.

Frequently asked questions

Did the RBA say AI has already boosted productivity?

No. Hunter said there is limited evidence that AI has materially lifted aggregate productivity so far. Some firms report task-level improvements, but widespread measured gains remain uncertain.

Does the RBA expect AI to raise inflation?

The speech describes possible near-term upward pressure because investment demand is arriving while supply is constrained. It also says added capacity and productivity could ease that pressure later. This is an analysis of channels, not a fixed inflation forecast.

Did Hunter announce an interest-rate change?

No. The 9 November speech was not a formal rate decision. It discussed how AI could influence economic conditions that central banks monitor.

Does AI exposure mean a worker will lose their job?

No. Exposure measures the potential for tasks to be automated or augmented; it does not establish that a job will disappear. The RBA describes early labour-market signals as modest and inconclusive.

References