Why our productivity problem won’t be solved by one big idea

Australia, like most advanced economies, has struggled in recent decades to resurrect the stronger productivity growth of the 1990s. The issue is quickly acquiring a new complication: the country is investing heavily in AI infrastructure before there is much evidence that AI itself is lifting productivity.

In the United States, where the data centre boom is most advanced, labour productivity has improved, but from very low levels. Much of the growth still appears to reflect capital being poured into the sector, rather than a fundamental change in what businesses can produce or how efficiently they operate.

That leaves an awkward question for policymakers: is the current investment boom laying the foundations for higher productivity growth, or is it building an enormous amount of infrastructure around an unproven promise?

These tensions were clear during a recent policy discussion at UNSW Sydney and the e61 Institute’s Firm Productivity and Dynamism Workshop 2026, which brought together Dr Alex Heath, Deputy Secretary for Economic Strategy and Productivity at NSW Treasury, and John Haltiwanger, Professor of Economics at the University of Maryland and a senior research associate at the US Census Bureau.

The US is not yet a productivity model

Moderated by Richard Holden, Vice-Chancellor’s Professor and Chief Societal Economist at UNSW, the discussion began with a look at whether Australia’s productivity challenges are unique or part of a broader trend. 

“Up until 2019, it was global,” Dr Heath said. “Now, the US looks different, and I think the interesting questions are: is it real, and is it sustainable? Is there a risk of overcapacity in certain industries that seem to be driving the productivity growth? How long is it going to take to actually see the productivity growth – not from the building bit, but from the technology being embedded in other processes?”

Even the improved productivity growth in the US is an “increase relative to a really poor trend” that had run since 2005, Prof. Haltiwanger said. He noted that the faster US growth in labour productivity (output per hour worked)  has been attributed partly to AI and partly to other factors, including the post-pandemic increase in job switching.

Learn more: How AI is changing work and boosting economic productivity

But the picture is less encouraging when measured by total factor productivity (TFP), which assesses how efficiently labour and capital are used together. Adjusted for capacity utilisation, TFP has not accelerated, suggesting that much of the apparent improvement may instead reflect the capital being poured into data centres, Prof. Haltiwanger said.

“Putting on an optimistic hat, there’s a reason we’re investing so heavily in data centres: because there’s a projection that AI is about to do amazing things to productivity growth – both TFP and labour productivity growth,” he said. “This isn’t the intangible capital that may also be accelerating; the data centres reflect physical capital investment that may be necessary to generate that growth.

“You could say the stock market believes that story,” he added. “Could this be a positive signal for the next 10 years? Good question.”

When technology changes the system

For AI to have a more systemic effect on productivity, it will need to change the way organisations and systems actually operate.

Prof. Holden referred to research comparing AI with the rise of electrification. Factories became more efficient by replacing steam engines with electric generators, but the more profound gains came later, when businesses reorganised production around the new technology.

"Everyone’s looking at AI as a way of potentially getting productivity gains in the delivery of services"

ALEX HEATH

Because factories no longer had to be organised vertically around a central steam source, production could be decentralised, making new forms of layout – and ultimately the assembly line – possible. That transition took decades because while the technological change was relatively simple, changing the factory's structure was not.

AI is already beneficial for some tasks, including information processing, coding, and document processing. But the larger gains will depend on discovering “literally, new ways of organising business”, Prof. Haltiwanger said. “That’s going to take time.”

For Dr Heath, the challenge also stems from the implications for large, long-term infrastructure investments. Even the comparatively simple changes involved with electrification required “huge amounts of fixed investment”.

“The business case of changing it, or the ease with which you can change it, is made difficult by the fact you’ve invested in this thing that’s going to be there 50 years,” she said. “The optimal time to change it is at the end of that 50 years – not now, when the new technology is here.”

With the data centre boom, this is not just a hypothetical problem. Dr Heath suggested that a profound technological shift may already be looming over the investment case: quantum computing could eventually deliver far greater computing efficiency, changing the economics of data centres.

She pointed to the UNSW Fundamental Quantum Technologies Laboratory’s work in silicon-based quantum computing as part of the broader research focus that could drive such a change. “All these data centres that are being built – how are they going to be future-proofed for this disruptive, wildly more efficient technology that’s apparently close to being possible?” she asked. “It’s partly about investment cycles, and you don’t want to get into the trap of thinking it’s the same as just getting a new software update.”

Government faces a slower learning curve

Government will be an important test of whether AI can deliver economy-wide productivity gains. With the public sector accounting for a substantial share of Australia’s economic activity and employing millions of people, even modest efficiency improvements could have a significant impact. But they will take time to emerge, Dr Heath cautioned.

“Everyone’s looking at AI as a way of potentially getting productivity gains in the delivery of services,” she said. “But there’s quite a significant learning curve before that’s going to turn into substantially higher productivity growth in the non-market service delivery space.”

Governments also need to ensure the workforce has skills that are “complementary to the new technology”, Dr Heath noted. “For a state government that delivers education, there’s a big piece of work on how you make sure all the kids are coming out of the education system with the right skills. There are a lot of legacy ways of doing things in education that will need to be reimagined.”

Subscribe for the latest research, analysis and insights from UNSW Business School

Prof. Haltiwanger agreed, adding that AI is “coming at us incredibly quickly and already disrupting our previous approaches. We’re in very deep water,” he said. “There’s a challenge for us in terms of just being able to continue doing what we’ve been doing – to figure out, how do we generate the kind of learning that we did in the past?”

According to Dr Heath, that is a problem for graduates and for those who employ them. “Marks are possibly not a great indicator of skills now, and you can’t assume that students have been taught the critical thinking skills you would expect; it changes how you have to train people for the workforce,” she said. “How do you make sure you’re training them to be complementary to AI – to use AI as their assistant to leverage up, rather than as a crutch?”

Measuring better care

As Australia’s population ages, healthcare and aged care are becoming a larger part of the economy – non-market sectors where productivity gains have traditionally been difficult to achieve because so much of the work depends on direct human care.

Prof. Holden suggested there may be more scope for productivity gains in healthcare than the headline figures imply, pointing to questions about how the sector is measured and the potential for complementary technologies, such as pill-dispensing robots.

Dr Heath agreed that technological change aimed not at replacing people but at complementing their skills can improve productivity. The measurement challenge, she said, is capturing the quality of healthcare outcomes and developing a “broader concept of quality-adjusted productivity”.

The NSW Government’s Performance and Wellbeing Statement, she said, reflects the need to look beyond point-in-time measures from the national accounts. “We actually want people to get well faster and get out of hospital faster – these are outcomes that I think about as a sort of quality adjustment to the productivity measure that no national accounts are going to pick up easily.”

At the same time, Prof. Haltiwanger pointed to concerns around people’s growing reliance on AI chatbots for information on health, relationships and mental health. “On the one hand, it seems incredibly powerful, and I think we’re awestruck,” he said. “You give it perhaps more credibility than you used to, even a few years ago, but it is still subject to errors.”

Small changes to prepare for big shocks

A broader policy question was whether policymakers should concentrate on a few potentially transformative interventions or pursue a long list of smaller reforms. “You need to work on the little things, because what you’re trying to do is make sure that there aren’t unnecessary rigidities in the system,” Dr Heath said.

While standards and regulation remain important, some rules reflect an earlier economic environment and may no longer serve their original purpose. Removing those rigidities would help the economy respond to larger changes.

"If you’re getting this big step change in the productivity of what you’re doing, are you getting a fair share of the rewards in a labour-negotiation sense?"

ALEX HEATH

“The reason we do macroeconomic policy is to make sure the economy is able to adapt in a more effective way when a shock comes along – a COVID, a financial crisis, a fuel crisis – your economy is able to adapt in a more effective way,” she said.

Prof. Haltiwanger agreed that economies, firms and individuals need to be able to pivot as circumstances change. “The evidence is pretty overwhelming that you need the economy and firms and individuals to be able to pivot with changing circumstances,” he said. “Promoting dynamism and flexibility and mobility and not having barriers to that on as many dimensions as possible is really important.”

For Dr Heath, the quality of the public debate around productivity and economic policy is another barrier. She pointed to the 2018 changes to the GST distribution, which were introduced in response to a cyclical problem but, in her view, embedded longer-term structural changes in the way GST revenue is shared between the states and territories.

The reforms effectively guaranteed Western Australia enough funding to provide 113% of the standard level of services, while other states and territories remained below that level unless topped up by the federal government.

Learn more: Why fixing Australia’s productivity problem is a ‘game of inches’

Dr Heath said the Productivity Commission had clearly set out the implications of the changes, but some of the public debate focused on denigrating the commission rather than the trade-offs involved in the distribution system. “The No Worse Off Guarantee is supposed to run out in a couple of years, so unless something changes, that impost is going to end up on all the other states and territories,” she added. “The quality of the conversation about this is just not sufficient to really grasp where the real trade-offs are and what the real policy issues are.”

Who gets the productivity gains?

Those trade-offs will become more difficult if AI does deliver the gains currently being promised. “If you are using AI efficiently and effectively, you’re generating a lot more surplus for the number of hours you’re working,” Dr Heath explained.

“If you’re getting this big step change in the productivity of what you’re doing, are you getting a fair share of the rewards in a labour-negotiation sense?” she added. “Things are changing quickly, and it is early days for thinking about how this will play out.”

One possible outcome is that productivity gains translate into shorter working hours without reducing pay. But that may require a change in the way the surplus is distributed, rather than assuming the benefits will flow through to workers. “There’s a lot about what’s fair in a labour market that’s up for grabs,” Dr Heath said.

Prof. Haltiwanger cautioned against governments getting too prescriptive about how businesses approach productivity. “I’m incredibly sceptical of any government-imposed structure of how businesses ought to organise their production,” he said. “Economies that have tried to do that are exactly the economies where productivity really suffers; government’s not well-suited to do this.”

He also argued that concerns about AI job displacement have not yet materialised at scale. In the US, he said, there had been “essentially no displacement” among the most AI-intensive users. “Historically, the job destruction often comes not from the successful adopters of new technology, but from the businesses that either don’t adopt or do it badly.”

Australia may eventually achieve the promised productivity gains, but technology alone will not determine the outcome. The infrastructure may take years to build, the systems around it may take longer to change, and the distribution of the gains will require choices that have barely begun to be discussed.