1 · The research question
Could Australia more than double its output per hour by 2035 if AI capabilities advance exponentially and the country simultaneously transforms health, housing, energy, industry and public administration?
The fourth scenario is intentionally far beyond normal institutional projections. It explores an outcome worth testing, not an outcome already demonstrated. Favourable GDP arithmetic alone cannot establish the availability of skilled workers, capital, electricity, resilient housing, social consent or fair distribution.
The main evidence for examining an exceptional case is exponential improvement in AI task capability, measured by METR; it does not imply the same exponential improvement in whole-economy output. Separately, CEPR researchers have modelled an extreme AI case with GDP growth reaching 15% per year by 2030, accompanied by serious labour displacement. Such work demonstrates why distribution and transition costs must be examined alongside GDP.
Modelling principle: GDP per capita and output per hour remain the core economic results. Housing affordability, care outcomes, distribution, emissions and trust are separate pass/fail measures, not invented as dollar multipliers.
2 · Four 2035 pathways
1. Business as usual
Gradual diffusion, limited institutional redesign and mounting housing, demographic and climate pressures. Closest in spirit to Treasury’s existing assumptions, but not the official Treasury forecast.
- Underlying non-AI labour-productivity growth: 0.9%/yr
- AI additional output/hour by 2035: +2.5%
- Hours per resident: −2.5% relative to 2026
2. Reform and diffusion
Better skills, business dynamism, targeted capital investment and faster adoption of existing technology. Still recognisable as a conventional productivity-reform programme.
- Underlying non-AI growth: 1.3%/yr
- AI additional output/hour: +10%
- Hours per resident: −1.2%
3. Integrated renewal
AI, housing construction, prevention-first health, care redesign, climate adaptation, infrastructure and institutional reform are treated as one programme.
- Underlying non-AI growth: 1.8%/yr
- AI additional output/hour: +30%
- Hours per resident: +1.2%
4. Exponential AI + system transformation MOST OPTIMISTIC
AI capability compounds rapidly while Australia reorganises many industries, manages displacement, industrialises housing, expands healthy working years and reinforces critical systems.
- Underlying non-AI growth: 2.2%/yr
- Additional realised AI output/hour: +90% by 2035
- Hours per resident: +4% through health, care and participation
- Capability stress test: six-month doubling—not a proven nine-year extrapolation
All pathway assumptions beyond observed statistics are scenario inputs created for this exploratory model. The underlying non-AI rate includes the potential contribution of non-AI innovation and productive capital. The AI uplift is measured relative to the model’s non-AI counterfactual, avoiding an additional AI increment on top of Treasury’s AI-inclusive 1.2% rate.
3 · Interactive scenario explorer
Change the gold-scenario assumptions to test whether an extraordinary outcome requires equally extraordinary rates of adoption, capital formation and labour-market adaptation. All four scenarios are shown in the comparison.
2035 economic outcomes, all values in constant 2025–26 AUD
| Pathway | GDP A$ trillion | GDP vs 2026 | GDP/person A$ | GDP/person vs 2026 | GDP/hour vs 2026 | GDP above BAU A$ trillion |
|---|---|---|---|---|---|---|
| Calculating scenarios… | ||||||
Year-by-year pathway 4
| Year ending | GDP A$ tr | GDP/person A$ | GDP/hour index 2026=100 | AI boost vs non-AI | Estimated adoption share of activity |
|---|---|---|---|---|---|
| Calculating years… | |||||
4 · What produces growth, and what does not
Exponential task-horizon index: 2^(12×years/doubling_months). Benchmarked progress is domain-specific; a capability doubling is not a GDP doubling.
An S-shaped diffusion factor and a deliberately small translation elasticity turn capability into a bounded scenario-specified uplift in domestic output/hour.
Labour hours/person incorporate ageing, health and care. Climate effects reduce output/hour. Capital and organisational redesign sit within non-AI productivity and realised AI gains.
How the optimistic case is calculated: First, non-AI output/hour grows at the chosen rate. Second, AI supplies a separate, cumulative productivity multiplier that reaches the selected value in 2035; its path depends on exponential capability and S-shaped organisational deployment. Third, a scenario-specific climate discount is applied. Finally, output/hour is multiplied by hours/person and population.
Capital is not added again on the spending side of GDP. Investments are required to make gains possible, but this first-pass model does not yet endogenise depreciation, capital stocks, foreign ownership, energy systems or financing. Extra investment therefore appears in the feasibility tests below and is not subtracted automatically from GDP. “GDP above BAU” in the table denotes the annual 2035 difference, not cumulative 2026–35 output or spendable government revenue.
Why the model does not claim to have solved resilience
The GDP calculations do not estimate the distribution of income, care quality, actual housing costs, emissions, public trust or debt sustainability. A large GDP gain could still coincide with mass displacement or fiscal pressure. The model has to pass separate social and physical tests before any pathway should be called successful.
5 · Feasibility of the most optimistic pathway
| System | What must become possible by 2035 | What would make pathway 4 fail | Data to verify the change |
|---|---|---|---|
| AI deployment | AI reliably performs complex multi-stage workflows, with independent verification in large parts of the economy. | Capability improves, but implementation cost, rework, errors or weak organisational redesign eliminate most gains. | Sector-level value added/hour; randomised deployment trials; quality and error rates; effective AI use. |
| Capital and energy | New compute, connectivity, power, storage, infrastructure and housing capacity support the boom. | Grid connection, water, financing, land use or construction bottlenecks restrict adoption. | Energy availability and costs, capital stock, equipment costs, investment returns, approvals and construction schedules. |
| Ageing and care | AI reduces administrative care tasks; prevention delays illness; productivity rises without reducing service quality. | Demand growth exceeds labour saved, or tasks cannot be automated safely and care quality worsens. | Need-adjusted care hours, healthy life years, avoidable hospitalisations, clinical outcomes and unpaid-care burden. |
| Housing and mobility | Faster, higher-quality construction plus well-located supply improve access to jobs and stable housing. | Lower build costs are absorbed by land prices; planning and delivery fail; households remain priced out. | Homes per construction hour; completions near employment; rent/income; first-buyer access; commute time. |
| Employment and distribution | Benefits reach workers and young people via wages, services, prices and ownership, alongside rapid reskilling. | GDP grows while labour income share collapses, long-term unemployment rises or younger cohorts lose ground. | Real median disposable income/person, labour share, transitions out of displaced roles, unemployment duration, wealth by age. |
| Resilience + trust | Lower expected climate losses and more effective institutions create space to reinvest. | Disasters, cyber attacks, concentrated power or low trust break adoption and investment coordination. | Loss-adjusted GDP, disruption days, climate insurance access, institutional trust, outages and cyber incidents. |
Capital cost: a deliberately unanswered question
The four cases assume illustrative additional annual public-and-private investment intentions of 0%, 0.7%, 1.5% and 3% of GDP at full scale, respectively. At the default settings in the fourth pathway, 3% of the modelled 2035 GDP would be roughly A$232 billion per year in 2026 purchasing-power terms—public and private combined. This is an extraordinary capital commitment and is not included in the calculated outputs or independently costed. A proper macroeconomic model would test the effect on energy, construction prices, interest rates, import dependence, public debt and private consumption. Until then, no GDP gain can be described as available government revenue or resilience funding.
6 · Evidence-gated research programme
Start with a working national accounting identity, then replace assumed coefficients with Australian sector and cohort evidence. Require independent challenge at each gate.
| Gate | Action and deliverable | What would support the optimistic case? | Falsification criterion |
|---|---|---|---|
| 2026–27 Measurement | Publish an ABS-linked versioned model card. Reconstruct GDP/hour, hours/person, sector investment and age trends, 2015–26. | Model reproduces accounting identities and out-of-sample historical trajectories. | Unexplained errors or revisions materially exceed published uncertainty bands. |
| 2027–29 Early deployment | Matched trials in care administration, building approvals, construction design, energy forecasting and SME work. | Net quality-adjusted output/hour improves markedly over control groups after integration costs. | Measured net value is small/negative, or safety/quality outcomes deteriorate. |
| 2029–32 Scale | Test national adoption, grid and workforce capacity, delivery timetables and the distribution of benefits. | Independent deployment grows across labour-intensive sectors, with adequate capital and rising median real income. | Large displacement without transition, blocked infrastructure, unaffordable energy or low domestic value capture. |
| 2032–35 Transform | Compare GDP/hour, hours/person and wellbeing with pre-registered pathway targets. | Productivity acceleration comes with healthier years, affordable housing, rising median living standards and lower losses. | GDP grows while care quality, housing security, labour share or climate exposure deteriorates beyond agreed thresholds. |
Suggested numerical go/no-go tests (not published forecasts)
Pre-register measurable thresholds in three diverse Australian pilots during 2027–2029. These are proposed decision criteria, not estimates of what the pilots will achieve:
| Pilot | Test | Continue only if… |
|---|---|---|
| Care operations | Compare AI-enabled administration and care scheduling against matched non-AI services for 12 months. | At least 20% fewer admin hours per client, with no worsening in safety, client outcomes or unpaid caregiver hours. |
| Housing production | Test AI planning/design plus modern off-site production against closely matched developments. | At least 25% greater quality-adjusted output per labour hour and evidence of lower delivered housing costs, not just land-price inflation. |
| Regional SMEs | Randomise access to AI workflow redesign and capital support, measuring value added—not only task speed. | At least 20% greater quality-adjusted value added per hour net of AI subscriptions, integration, energy and error-correction costs. |
If these results are not replicated in independent settings, reduce assumptions or reject the aggressive 2035 diffusion timetable. Passing them would support more ambitious modelling; it would not validate a 90% economy-wide AI productivity uplift.
Minimum independent data for a credible second version
ABS: annual capital inputs, industry output and hours, real GDP and demographic series. ATO/Treasury: income distributions, effective taxes, investment and budget flows. AI: audited task benchmarks plus industry adoption and net output. JSA/AIHW/NDIA: paid and unpaid care needs, workforce shifts, outcomes and cost. Housing: completions, land prices, location, rents and time-to-approval. Climate: geolocated asset risk and expected annual damage.
Model development should add (a) capital stocks and depreciation; (b) sector input–output relations; (c) cohort-based work, health and care; (d) income and ownership distribution; (e) climate damage and adaptation; and (f) uncertainty ranges and behavioural feedback. A single-factor GDP extrapolation cannot replace those modules.
7 · Evidence and source ledger
Sources establish observations, historical mechanisms or external comparisons; none authorises the invented 2035 pathway-4 parameters as a forecast. Dates and jurisdictions matter. Original sources linked below.
- ABS annual national accounts, 23 Oct 2026
Real GDP +2.3%, GDP/person +0.8%, GDP/hour +0.5% in 2025–26; next quarter 2026 release due December. - ABS quarterly national accounts, June 2026
Nominal GDP baseline $2.930326 trillion is the sum of 2025 Sep, Dec and 2026 Mar, Jun quarters from September 2026 publication; may be revised by 23 October annual benchmarks. - ABS population, March 2026
Population 27,921,150 at 31 March 2026. Simplified annual average population model anchor of 27.86m is approximate. - Parliamentary Budget Office, 2026 Intergenerational Report guide
Treasury 1.2% p.a. long-term labour-productivity projection already includes AI; sensitivity 0.8–1.6%. Population averaged 0.9% over 40 years, not 2035. - METR Time Horizon 1.1, January 2026
Historical exponential task-horizon performance; P50 all-period 196.5 days, post-2023 130.8 days, post-2024 88.6 days. Task-specific, not GDP growth. - CEPR, Korinek et al, September 2026
Transformative AI extreme model: GDP growth reaches 15% yearly by 2030 with major labour displacement; US calibration, not Australia. - ABS business AI adoption, June 2026 release
12% of businesses reported AI use in 2024–25; survey measure is not weighted share of tasks transformed. - ATSE + Kearney, AI Investment Blueprint, Nov 2025
Australian AI programme could lift 2034 GDP by 6–8%, or A$160–235 billion, with unquantified breakthroughs possible; report not measured outcome. - CSIRO Australian National Outlook 2019
Integrated national transformation scenario; GDP/person up to 36% above slow-decline by 2060 (not 2035). - Jobs and Skills Australia 2025–2035
Care/social assistance employment +22.9%; total employment +13.3% through 2035; projection excludes generative AI. - Productivity Commission, housing construction
Half as many new homes completed per hour as in 1995; quality-adjusted construction labour productivity −12%. - National Climate Risk Assessment 2025
Eleven priority climate risks; no support for specific scenario climate-loss percentages, which are assumed here. - Productivity Commission, Five Pillars final reports
47 final reform recommendations, December 2025; ingredients, not proof of an integrated 2035 scenario. - OECD, Foundations for Growth and Competitiveness 2026
Illustrative G7 AI productivity effects range 0.2–1.3 percentage points a year by diffusion scenarios; not Australian projections. - Australian Academy of Science, Science 2035
Australian scientific capability gaps across technology, demography, climate and sovereign capability.