When Productive Capability Leaves the Balance Sheet

Australia’s uncomfortable productivity question

Australia’s productivity performance has become difficult to explain in plain language. Organisations can access dramatically more computing power, software, cloud infrastructure, automation, data capability and—now—generative AI than they could a generation ago. Much of it is available on demand, without building a data centre or employing a large internal IT workforce.

Yet measured productivity growth has remained weak. The Productivity Commission’s September 2026 update reported that labour productivity was flat in the June quarter and fell 0.2 per cent over the year, leaving overall productivity barely above its pre-pandemic 2015–19 average.[1]

That gap between felt capability and aggregate productivity invites strong stories. Some blame measurement. Others point to regulation, market concentration or industry mix. Some assume digital transformation must be productive and ask why the statistics disagree.

This essay asks a narrower question—one that sits inside the wider puzzle rather than replacing it:

Has the shift from internal labour and owned digital capital towards externally supplied cloud, software-as-a-service and AI weakened the relationship between the productive capability available to Australian organisations and the capital, value added and productivity observed within the Australian economy?

The proposition is exploratory. It does not claim that Australia’s productivity slowdown is a cloud-accounting artefact, that the Australian Bureau of Statistics has failed to count digital spending, or that strong corporate profits prove mismeasurement. It argues instead that digitalisation has changed where capability is produced, how it is recorded and when downstream outcomes become visible in productivity statistics. Those structural shifts deserve closer attention before the next wave of AI consumption.

From owning capability to accessing it

For much of the enterprise-computing era, productive digital capability appeared on balance sheets and in headcount. Organisations owned or tightly controlled infrastructure, capitalised eligible software and systems, employed specialists to build and operate them, and depreciated or amortised assets over time. Employee compensation formed part of value added when those workers were employed by the producing entity.

That model still exists. But a growing share of digital capability is acquired through a different model:

Cloud and SaaS did not invent outsourcing. They made granular, metered access to computing and software routine at scale. Generative AI extends the same logic from compute towards cognition: models, tokens and APIs can be purchased like other digital services.

Digital expenditure → access → productive capability → effective use → valuable outcome

These are not the same step. A firm can spend more on digital services and still fail to complete more valuable work. It can lower operating costs by substituting purchased services for internal labour without increasing Australian gross value added (GVA). It can also improve private profitability through pricing, mix or imported inputs without an equivalent gain in economy-wide productive capability.

The claim that “SaaS is OPEX, so productivity statistics miss it” should therefore be rejected. Purchased cloud and software services are generally captured in the national accounts—as intermediate consumption, imports, provider output, wages and gross operating surplus. The harder questions concern classification, location, timing and the link to downstream output, not whether invoices exist.

Follow A$100 million

Consider a hypothetical Australian company that spends A$100 million a year on hyperscale-style cloud infrastructure and related managed services. The sector does not matter. The example could apply to manufacturing, financial services or a government trading enterprise.

All figures in this section are illustrative. They are not financial data for any named provider and should not be read as estimates of how cloud revenue is divided among wages, imports and gross operating surplus in Australia. The purpose is to trace the accounting logic, not reverse-engineer vendor cost structures.

The internal counterfactual

Under a more internal model, the organisation might have employed engineers, architects, developers and operators; owned or leased data-centre capacity and hardware; licensed software; and developed systems in-house. Wages would have formed part of its own GVA, while eligible software and infrastructure would have been capitalised and depreciated over time. In this stylised case, total annual cost might have been A$120 million.

The cloud shift

Under the cloud model, the same organisation might spend A$100 million on infrastructure, SaaS, security, data platforms and implementation partners, while employing fewer people to run physical infrastructure and owning less capitalised IT.

Operating costs could fall by A$20 million and, all else equal, reported profit could improve. That is a real financial outcome for the firm. It does not follow that Australian GVA rises by A$20 million.

Purchased services are generally intermediate consumption for the customer: they are deducted when calculating the customer’s GVA. Employee compensation, by contrast, is part of value added for the entity that pays the wage.[2]

The A$100 million does not disappear. It may reappear as output and GVA of an Australian-resident provider; wages paid in Australia; gross operating surplus; purchases from Australian suppliers; imports of software, intellectual property, equipment or services; or intermediate inputs elsewhere in the supply chain.

Without provider-level data, assigning shares would be speculation. The structural point is enough: the same underlying activity can move across line items, industries and balance sheets without disappearing from the accounts. National accountants have explicitly recognised that cloud adoption can alter investment patterns, trade flows and measured value added, requiring statistical systems to evolve.[3][4]

The Australian cloud architect

The aggregate example becomes more concrete when viewed through labour. Imagine a solutions architect employed by an Australian-resident cloud or IT services provider spending 100 hours helping a local customer design virtual networks, API gateways, identity, security controls and operating architecture.

• The hours are not missing. If the worker is employed in Australia by a resident provider, their compensation contributes to that provider’s Australian GVA.[2]
• The architecture may benefit the customer for years through faster delivery, safer operations and reusable patterns, even if the customer does not recognise a distinct capitalised intangible for much of that know-how.
• Subsequent cloud charges remain purchased services for the customer rather than owned capital.
• The customer’s productivity effect appears only when output changes. Official measures reflect a gain when the customer ultimately produces more measured real output relative to its inputs, including service charges and internal labour.

A useful chain—though not a national-accounts identity—is:

Australian architecture labour → configured cloud environment → embedded organisational capability → recurring cloud consumption → changed workflow → valuable output → eventual measured productivity effect

National accounts measure inputs and outputs within the production boundary. They do not maintain a separate stock account for every organisational capability created along the way. That is not necessarily an error. It is a gap between what economic aggregates are designed to explain and what executives experience when capability moves off the balance sheet.

Firm-level accounting reinforces the distinction. A SaaS customer typically does not control the supplier’s underlying software. Configuration and customisation costs are capitalised only when they satisfy IAS 38 criteria; otherwise, they are often expensed even when they shape future operations.[5] Company accounts and national accounts answer related, but not identical, questions.

What national accounts measure—and what they do not

The ABS defines multifactor productivity (MFP) for an industry as gross value added relative to combined labour and capital inputs. Gross output-based MFP is measured relative to labour, capital and intermediate inputs, including services in KLEMS growth accounts.[6][7]

GVA is output less intermediate consumption. Wages and gross operating surplus belong to the producing unit; purchased services used in production are generally intermediate inputs, not part of the customer’s GVA.

That framework is coherent. It is also sensitive to where production is attributed when capability is unbundled across firms and borders. As digital delivery shifts from owned capital and internal labour towards purchased services, the customer’s capital stock may grow more slowly, its GVA may contain fewer internal IT wages, provider and import lines may rise, and measured output per input may improve only after a long and noisy adjustment.

IMF work on cloud computing distinguishes two effects. One is genuinely productive: shared infrastructure can reduce excess on-premises capacity and allow the same real output to be produced with fewer resources. The other is structural: geographic allocation, leasing boundaries and statistical systems designed for traditional IT can complicate where activity is recorded as delivery models change.[3][4]

None of this proves that Australia’s slowdown is a cloud artefact. It shows why digital production and accounting structures can diverge even as data collection improves.

The questions aggregates cannot answer

National productivity measures are not operational dashboards for institutions. They do not, by themselves, tell us whether a government agency can complete more compliant decisions each month; whether organisational knowledge was reused or merely re-documented; whether AI reduced rework or accelerated low-value activity; how much judgement remained authoritative after automation; or whether staff can access more capability than the institution owns.

Those are management and governance questions, not necessarily gaps in GDP. The distinction can be stated simply:

digital expenditure ≠ access ≠ productive capability ≠ effective use ≠ valuable outcome

KLEMS and the migration from capital to services

KLEMS decomposes industry inputs into K (capital), L (labour), E (energy), M (materials) and S (purchased services).[7]

In Information, media and telecommunications—an industry central to digital delivery—the ABS reports that the services share of gross output rose from 50.1 per cent in 2014–15 to 55.9 per cent in 2023–24, while the capital share fell from 26.2 per cent to 18.4 per cent. The ABS explicitly links those cost-share movements to increased outsourcing of cloud-based services.[7]

This is structural evidence that activities once produced through capital and labour inside firms can increasingly be accessed as purchased services. It does not prove that cloud caused Australia’s economy-wide productivity slowdown. Industry cost shares move for many reasons, including reclassifications in supply-use tables, and MFP growth varies from year to year.

The more defensible conclusion is narrower: capability can migrate across input categories even when total activity is captured. That is a classification and production-structure effect, not evidence that productivity has been mismeasured.

Digital services and the evolving statistical picture

Statistical agencies have already had to adapt as digital delivery outgrew older service categories. From the September quarter 2024 release, the ABS introduced improved estimates of digital services in international trade, with revisions back to 2009–10. The new categories draw on data from the Australian Taxation Office, Business Activity Statements, GST, ASIC and international trade in services.[16]

The ABS estimates cumulative digital-services imports of about A$87 billion from 2009–10 to 2023–24—roughly 17 per cent of other-services imports over that period. Annual imports rose from about A$119 million to about A$18.5 billion.[16]

These figures are not a count of “missing cloud bills”. The categories are broader than hyperscale cloud and include software-distribution licences, advertising and market-research services, and other electronically delivered services. The lesson is more modest: cross-border digital-service measurement is difficult and still evolving.

The revisions changed imports, intermediate use and parts of gross fixed capital formation and household consumption—not because the economy had been invisible, but because classification and data sources improved.[16] At the same time, ABS work on data centres shows that domestic hyperscale infrastructure can appear strongly in capital expenditure, building activity and equipment imports.[23] Digital capability can therefore raise measured investment in one part of the accounts and service imports in another.

The intangible investment problem

Much organisational change—process redesign, training, workflow experimentation and governance improvement—creates benefits over several periods but is treated as current expense in company accounts and often as intermediate consumption or labour cost in national accounts.

Productivity Commission research found that Australian market-sector spending on intangibles was large relative to spending on tangibles, including organisational capital, firm-specific skills and computerised information, and that much of it was not treated as investment in the national accounts of the time.[13]

This matters because cloud transitions are usually bundled with organisational change: new operating models, controls, skills and vendor relationships. The expenditure may be visible while the stock of effective practice is not.

But the counter-evidence is important. Capitalising more intangibles does not automatically improve the historical MFP story. Measured inputs rise alongside measured output, so the net effect on MFP growth is ambiguous and can be negative in some periods.[13][14] The Commission’s message was not that productivity was secretly healthy. It was that measurement choices affect both sides of the ratio.

That qualification disciplines the argument here. The hypothesis is not that hidden investment explains everything. It is that the location and timing of capability complicate the relationship between digital access and measured productivity.

Profitability is not productivity

Corporate profit can improve without an equivalent gain in economy-wide productive capability. The distinction is definitional, not cynical.

In the hypothetical cost shift, moving from A$120 million of internal IT to A$100 million of cloud services can strengthen value capture if margins improve without adding the same amount to Australian GVA. That is a story about where value added is recorded, not proof of mismeasurement.

Weak aggregate productivity can coexist with strong firm-level profits for many reasons. Research on mark-ups and declining competition in Australia identifies pricing power and misallocation as one possible channel.[15] Commodity conditions, frontier–laggard divergence, global IP ownership, network effects, industry composition and imported services may also widen the gap between value capture and economy-wide value creation.

Productive capability without capital ownership

For analysis and institutional governance, one question should sit alongside—not replace—official measures such as capital services per worker and industry capital shares:

How much productive capability can an Australian worker or institution access, regardless of who owns the underlying capital?

This is a conceptual distinction, not a proposed substitute for national-account measures. Cloud makes it visible. A mid-sized Australian firm can now access compute, storage, analytics and language models that would once have required substantial owned infrastructure.

Capability accessible can therefore diverge from capital owned or produced within Australia when services are imported, intellectual property is held offshore, or architectural knowledge sits in provider labour and customer practice rather than capitalised software.

National accounts should continue to measure domestic production and income on established definitions. Executives and policymakers also need language for access without ownership—without treating every subscription as hidden investment or automatic productivity gain.

Then come AI tokens

Cloud computing allowed firms to rent computing capability. Generative AI increasingly allows them to rent cognitive capability—classification, drafting, coding assistance, search, summarisation and reasoning support—metered as tokens or API calls.

Token and API spending is not yet large enough, in aggregate, to explain Australia’s current productivity performance. The argument is forward-looking: if cognitive work is purchased like cloud services, the same questions of classification, location and attribution may intensify even when real output eventually responds.

Australian institutional knowledge + Australian human judgement + external AI model + external compute + token consumption → Australian organisational outcome

Financially, token spending resembles other purchased digital services. Economically, it may substitute for parts of analyst, programmer, researcher, administrator and reviewer effort while creating new assurance and governance requirements.

• If AI enables an Australian organisation to produce more measured real output with the same domestic inputs, productivity measures should eventually reflect the gain.
• If AI mainly substitutes imported AI services for Australian labour without raising measured output, profitability may improve without a one-for-one rise in Australian GVA.

The next transformation may not simply move servers off the balance sheet. It may move a growing share of cognition there as well.

Measuring what happens between expenditure and output

National aggregates remain essential, but they are poor operational dashboards for individual institutions. A complementary institutional metric—not a replacement for GDP or official productivity—is:

Completed, quality-checked valuable outcomes ÷ total resources consumed

Resources may include human time, cloud and software spending, AI tokens, compute, review effort, rework, delay, correction costs and assurance overhead.

This ratio creates an operational evidence layer beneath national aggregates. It asks whether access to digital capability improved the institution’s ability to complete valuable work. Platforms such as ifCEM matter only insofar as they help institutions record what was attempted, what was established, who authorised it and what outcome followed—so that “we bought more AI” is not mistaken for “we became more capable”.

The question before the next wave

The evidence does not justify saying that cloud or intangible accounting explains Australia’s productivity slowdown.

It does justify asking whether the structure of digital production has changed faster than the intuitive chain many executives still carry:

investment → capital → capability → output → productivity

Cloud separated access to productive capability from ownership of much of the underlying capital. AI may separate access to cognition from the employment of cognition inside the firm.

Statistical agencies have revised digital-services estimates, improved trade measurement and published KLEMS accounts because the economy changed—not because the core framework failed.[7][16] Firm-level accounting likewise distinguishes subscription access from capitalised software.[5]

What remains under-examined is the middle: the stock of organisational capability created between provider labour, customer expenditure and eventual output—and whether institutions can see it clearly enough to govern AI-assisted work.

What valuable productive capability did we actually create?

That question should shape the next wave of Australian digital and AI policy more than another round of consumption metrics alone.

References

[1] Alex Robson, Productivity Commission, "Productivity update — September 2026", 3 September 2026. https://www.pc.gov.au/ongoing/productivity-insights/update-september-2026/

[2] Australian Bureau of Statistics, Australian System of National Accounts: Concepts, Sources and Methods, edition 8. https://www.abs.gov.au/statistics/detailed-methodology-information/concepts-sources-methods/australian-system-national-accounts-concepts-sources-and-methods/edition-8

[3] International Monetary Fund, Accounting for Cloud Computing in the National Accounts, Working Paper WP/20/127, July 2020. https://www.imf.org/en/Publications/WP/Issues/2020/07/17/Accounting-for-Cloud-Computing-in-the-National-Accounts-49578

[4] International Monetary Fund, Balance of Payments and International Investment Position Manual Compilation Guide — VM1 23/02, "Measurement of Cloud Computing in National Accounts", 2023. https://www.imf.org/external/pubs/ft/bop/2023/pdf/40/23-02.pdf

[5] IFRS Foundation, IAS 38 Intangible Assets. https://www.ifrs.org/issued-standards/list-of-standards/ias-38-intangible-assets/

[6] Australian Bureau of Statistics, Estimates of Industry Multifactor Productivity methodology, 2023–24 financial year. https://www.abs.gov.au/methodologies/estimates-industry-multifactor-productivity-methodology/2023-24

[7] Australian Bureau of Statistics, Estimates of Industry Level KLEMS Multifactor Productivity, 2023–24 financial year (released 28 March 2025). https://www.abs.gov.au/statistics/industry/industry-overview/estimates-industry-level-klems-multifactor-productivity/latest-release

[13] Productivity Commission, Investments in Intangible Assets and Australia's Productivity Growth, Staff Working Paper, 2009. https://www.pc.gov.au/research/completed/intangible-assets/

[14] D. T. Nguyen and M. T. Vu, "Intangible Assets and Australia's Productivity Growth", UNSW Australian School of Business Research Paper, 2014. http://research.economics.unsw.edu.au/RePEc/papers/2014-08.pdf

[15] J. Hambur and O. Freestone, Reserve Bank of Australia, Research Discussion Paper 2025-05, "How Costly are Mark-ups in Australia? The Effect of Declining Competition on Misallocation and Productivity", 2025. https://www.rba.gov.au/publications/rdp/2025/2025-05.html

[16] Australian Bureau of Statistics, "Introduction of digital services in the Balance of Payments", feature article (September quarter 2024 release). https://www.abs.gov.au/articles/introduction-digital-services-balance-payments; and "Impacts from the 2024 Annual National Accounts historical revisions". https://www.abs.gov.au/articles/impacts-2024-annual-national-accounts-historical-revisions

[23] Australian Bureau of Statistics, "Spotlight — Data Centres in Economic Statistics". https://www.abs.gov.au/articles/spotlight-data-centres-economic-statistics

Related DataMPowered research

• Are We Building Institutional Capability Inside AI When We Don't Need To?
• Are We Transforming Around the Wrong Thing?
• Information Rich, Judgement Poor
• The Knowledge We Buried
• When AI Removes Administrative Friction
• AI Isn't Just Changing Jobs — It's Exposing Why We Designed Them This Way
• The Lesson After the Bitter Lesson
• Australia Is Building AI Infrastructure. But What National Capability Are We Building?
• Judgement Governance
• Operational Intelligence

Claim verification ledger: Docs/evidence/when-productive-capability-leaves-the-balance-sheet-claim-ledger.md (WPC-01–WPC-28).