Make It Make Sense

What happens when AI meets Singapore's economy?

Singapore Today

A small nation with outsized ambitions in artificial intelligence.

S$0BGDP (2024)
0%AI Adoption Rate (IMDA 2024)
0MWorking Population

Singapore has staked its future on artificial intelligence. With one of the world's highest GDP-per-capita figures, a tightly managed labour market, and aggressive public investment through the National AI Strategy 2.0, the city-state is a unique laboratory for understanding how AI reshapes a modern economy — for better or worse.

Four Possible Futures

From AI-powered prosperity to technological displacement.

Bull — AI Augments

AI augments productivity across all sectors. Wages rise, new jobs emerge, government revenues grow.

Click to explore →

Median — Cautious Optimism

Mixed effects — some sectors gain, others face disruption. Cautious policy keeps things stable.

Click to explore →

Bear 1 — Displacement

AI displaces workers faster than retraining can absorb. Government response is too slow.

Click to explore →

Bear 2 — AI Disappoints

AI underdelivers on promises. Massive CapEx spending yields disappointing returns.

Click to explore →

The Jobs Question

AI doesn't affect all sectors equally.

AI's impact on Singapore's employment landscape varies dramatically depending on the speed of adoption, government responsiveness, and the ability of workers to retrain. In the best case, AI augments existing roles and creates new ones; in the worst, displacement outpaces adaptation across entire sectors.

Total Employment Across Scenarios

202520262027202820292030203120322033203420353500K4000K4500K5000KTotal Employment

Bull: AI augments workers across every major sector, adding over one million jobs by 2035. Employment rises to 5.2 million with unemployment holding at 2.0% — essentially full employment driven by new roles in AI development, data engineering, healthcare, and personal services.

Median: The economy adds 433,000 jobs, but beneath a steady 2.0% unemployment rate lies significant churn. Workers displaced from routine roles in retail, manufacturing, and administration find new positions — often after retraining and frequently at lower initial wages.

Bear 1: Singapore could lose over 530,000 jobs by 2035 as AI displacement outpaces retraining. Unemployment surges to 12% — a level the city-state has never experienced — hollowing out the middle of the labour market while high-skill and low-skill roles survive.

Bear 2: AI underdelivers, but the damage is stagnation rather than crisis. Employment edges up by only 286,000, unemployment drifts to 3.5%, and the high-skill AI jobs that were supposed to emerge simply don't materialise in sufficient numbers.

Sectoral Employment Breakdown

20252026202720282029203020312032203320342035200K250K300K350K400K450KEmployment

Winners and Losers

Who benefits and who falls behind in each future.

AI's wage impact is far from uniform. In sectors where AI augments human productivity — finance, professional services, and tech — wages could rise substantially as output per worker grows. But in sectors vulnerable to automation, downward pressure on wages compounds the displacement effect, widening inequality between industries and skill levels.

Average Compensation Across Scenarios

20252026202720282029203020312032203320342035S$60KS$80KS$100KS$120KS$140KS$160KS$180KAverage Compensation

In the Bull scenario, average compensation rises as AI augments high-skill work and lifts productivity across services and finance. But in Bear 1, wage stagnation compounds job losses — displaced workers re-entering the labour market push down wages in the sectors still hiring, creating a vicious cycle of lower incomes and weaker consumer demand.

Sector Wage Change by 2035

Percentage change in average sector wages from 2025 base to 2035. Toggle scenarios to compare outcomes.

-100%-50%+0%+50%+100%+150%+200%+250%+300%Wage Change (2025 to 2035)ManufacturingConstructionWholesale& Retail TradeTransportation& StorageAccommodation& FoodInfo& CommsFinance& InsuranceReal Estate& Professional AdminCommunity, Social& Personal

The Government's Dilemma

How AI reshapes revenue and spending.

AI's effects flow through to government finances via tax revenues and spending. On the revenue side, corporate income tax rises or falls with profits, personal income tax tracks wages and employment, and GST follows consumer spending. On the expenditure side, retraining programmes and social safety nets expand when displacement accelerates. In the best case, a productivity-driven economy grows the tax base; in the worst, falling employment and wages erode revenue while welfare costs surge — creating a fiscal squeeze from both sides.

Government Revenue Composition

Breakdown of tax revenue sources under each scenario. Select a scenario to see how the revenue mix shifts.

Corporate Income Tax
Personal Income Tax
Goods & Services Tax
Stamp Duty
Other Revenue
20252026202720282029203020312032203320342035S$0.0BS$50.0BS$100.0BS$150.0BS$200.0BS$250.0BS$300.0BRevenue

In the Bull scenario, surging corporate profits and higher wages grow the tax base by over 40% by 2035 — CIT expands as AI-powered firms capture global market share, while rising incomes lift PIT and GST collections. But in Bear 1, tax revenue stagnates as unemployment rises and real incomes fall, even as government social spending must increase to cushion displaced workers — creating a fiscal squeeze that limits the state's capacity to invest in the very retraining programmes needed to reverse the spiral.

Fiscal Balance Across Scenarios

Government budget balance (revenue minus expenditure) under each scenario. Negative values indicate deficit.

20252026202720282029203020312032203320342035S$-20.0BS$0.0BS$20.0BS$40.0BS$60.0BS$80.0BFiscal Balance

The fiscal trajectories diverge sharply after 2028. In the Bull case, expanding revenues outpace spending growth, moving the budget toward surplus. In Bear 1, the combination of eroding tax receipts and rising social expenditure drives the deficit wider each year — the very definition of a fiscal doom loop where the government has less money precisely when it needs to spend more.

Deep Dive: Four Futures

What each scenario means for Singapore — in numbers and in practice.

Bull — AI Augments

AI supercharges productivity, wages, and fiscal health.

+1090.9K
Employment Change
S$165.3K
Avg Compensation (2035)
S$1.3T
GDP (2035)
2.0%
Unemployment (2035)
S$68.9B
Fiscal Balance (2035)

AI augments productivity across sectors, driving wage and employment growth simultaneously. Government revenues surge from corporate profits and rising incomes, enabling investment in infrastructure and education. Singapore cements its position as a global AI hub, attracting talent and capital.

Employment Trajectory

Bull — AI Augments highlighted — total employment (thousands)

202520262027202820292030203120322033203420353500K4000K4500K5000KEmployment

Most Affected Sectors

Largest employment change (2025 to 2035)

Community, Social & Personal+368.3K
Real Estate & Professional Admin+207.2K
Construction+171.6K

Explore the Data

Pick a metric, toggle scenarios, see the numbers.

GDP

All scenarios, 2025–2035. Toggle scenarios in the legend below.

20252026202720282029203020312032203320342035S$800.0BS$900.0BS$1.0TS$1.1TS$1.2TS$1.3TGDP
Bull — AI Augments
S$1.3T
+57.1% from 2025
Median — Cautious Optimism
S$1.1T
+36.0% from 2025
Bear 1 — Displacement
S$932.4B
+16.8% from 2025
Bear 2 — AI Disappoints
S$896.4B
+12.3% from 2025

How We Built This

The model, the data, and the assumptions behind the numbers.

The forecast engine uses 8 log-linear partial adjustment models estimated on Singapore historical data (2001–2024). Each equation captures how AI adoption, economic activity, and lagged adjustment interact to determine outcomes across employment, wages, government revenue, and public spending.

All models use OLS with Newey-West HAC standard errors to handle heteroskedasticity and autocorrelation. Variable selection was guided by AIC/BIC criteria, with structural break dummies for COVID-19 and policy changes. Pre-period out-of-sample validation was used to test forecast stability.

The AI variable is a hybrid construct: a global capability proxy (based on compute scaling and benchmark performance) weighted by sector-specific domestic adoption coefficients derived from IMDA enterprise survey data. This ensures the model captures both the pace of technological progress and Singapore's actual adoption patterns.

Each equation is estimated at the level that best captures the economic mechanism. Sectoral equations (1–2) are panel models with sector fixed effects; fiscal equations (3–8) are aggregate time series.

Eq 1 — Sectoral Employment: ln(E_it) = α_i + β1·ln(Y_it) + β2·A_it + β3·ln(E_it-1) + β4·D_covid + ε
Eq 2 — Sectoral Wages: ln(W_it) = α_i + β1·ln(VA/L_it) + β2·ln(U_t) + β3·A_it + β4·ln(W_it-1) + ε
Eq 3 — Corporate Income Tax: ln(CIT_t) = α + β1·ln(GDP_t) + β2·ln(CIT_t-1) + β3·A_t + β4·D_covid + ε
Eq 4 — Personal Income Tax: ln(PIT_t) = α + β1·ln(W_agg) + β2·ln(E_agg) + β3·ln(PIT_t-1) + β4·D_covid + ε
Eq 5 — Goods & Services Tax: ln(GST_t) = α + β1·ln(PCE_t) + β2·ln(GST_t-1) + β3·D_rate + β4·D_covid + ε
Eq 6 — Property Prices: Δln(P_t) = α + β1·Δln(Income_t) + β2·Δln(Activity_t) + β3·Δln(P_t-1) + β4·D_cooling + ε
Eq 7 — Social Expenditure: ln(SOC_t) = α + β1·ln(U_t) + β2·ln(GDP_t) + β3·ln(SOC_t-1) + β4·D_covid + ε
Eq 8 — Retraining Expenditure: ln(RET_t) = α + β1·ln(A_t) + β2·ln(U_t) + β3·ln(GDP_t) + β4·ln(RET_t-1) + ε

All variables in natural logs except dummies and rates. Subscript i denotes sector, t denotes year. A = AI adoption variable, U = unemployment, D = dummy variable.

Bull — AI Augments

AI augments human productivity across all sectors. Firms adopt AI to complement workers rather than replace them. Wages rise as output per worker grows, new job categories emerge, and government revenues expand with the growing tax base. Scenario multipliers amplify positive AI coefficients (wage +20%, CIT +50%).

Median — Cautious Optimism

Mixed effects — some sectors gain while others face disruption. Government policy is adequate but not transformative. The economy grows moderately as AI adoption proceeds at historical rates. Baseline scenario multipliers (1.0x across the board).

Bear 1 — Displacement

AI succeeds technically but displaces workers faster than retraining can absorb. Government response is slow and poorly targeted — retraining expenditure rises but too slowly, for the wrong skills, at insufficient scale. Unemployment hits 12%, wages fall in vulnerable sectors, and eroding tax revenue meets rising social costs in a fiscal squeeze. Employment AI multipliers are 0.80–1.50x (displacement).

Bear 2 — AI Disappoints

AI underdelivers on its promises. Massive capital expenditure yields disappointing productivity gains — a dot-com bust 2.0. CIT AI multiplier turns negative (-0.50) as firms write down AI investments. Growth stalls, but displacement is limited since the technology itself underperforms.

The AI variable is a hybrid construct combining two components:

  • Global AI capability index — a proxy for frontier AI performance based on compute scaling, benchmark scores, and model capability trajectories
  • Sector-specific adoption coefficients — derived from IMDA's enterprise digital adoption surveys, capturing how quickly each Singapore sector actually adopts new AI capabilities

To prevent unrealistic out-of-sample extrapolation, the forecasting engine applies a dampening function beyond the estimation sample maximum (AI index = 6.12 in 2023). Beyond this threshold, only 30% of additional AI capability is credited, reflecting uncertainty about whether future gains translate as directly as historical ones.

Scenario multipliers then scale the AI contribution differently: Bull amplifies the positive effects, Bear 1 amplifies displacement effects, and Bear 2 reduces the overall AI impact to reflect technological disappointment.

The v3 forecasting engine runs 4 scenarios × 11 years (2025–2035) × 9 sectors, producing 44 aggregate rows and 396 sectoral rows. Key design features:

  • Growth caps — wage growth capped at 8% annually, retraining expenditure caps vary by scenario (Bull 15%, Bear 1 only 8%), preventing explosive trajectories from compounding lag coefficients
  • Labour market floors — unemployment floor at 2.0% (historical minimum ~1.9%), ceiling at 12.0%; sector employment cannot fall below 5% of 2023 levels
  • Lag coefficient capping — estimated lag coefficients exceeding 0.98 are capped to prevent non-stationary explosive paths (Manufacturing wages had 1.14, Info & Comms 1.07)
  • Base year normalisation — 2025 values are averaged across all scenarios so every scenario starts from the same point, ensuring divergence reflects only the scenario assumptions, not estimation noise

All data is sourced from official Singapore government statistical agencies:

SourceData Used
Department of Statistics (DOS)GDP, employment by sector, compensation, government revenue & expenditure
Ministry of Manpower (MOM)Labour force, unemployment rates, wages
SingStat Table BuilderNational accounts, fiscal data, property price indices
IMDAEnterprise digital adoption rates, AI usage surveys
HDB / URAResale price index, private property price index, transaction volumes
CPF BoardOrdinary Account balances (housing affordability driver)

Estimation window varies by equation: FY2001–2024 for fiscal series, 2005–2023 for sectoral panels, 2004 Q1–2025 Q3 for quarterly property models.

  • Out-of-sample AI extrapolation — AI capability beyond 2023 levels has no historical precedent in the estimation sample. The dampening function is a modelling choice, not an empirical finding.
  • No general equilibrium feedback — equations are estimated independently. Falling employment does not endogenously reduce consumer spending or housing demand beyond what the individual equations capture.
  • Policy is exogenous — government response (retraining scale, tax changes, new programmes) is encoded as scenario parameters, not as endogenous policy reaction functions.
  • Sector classification is fixed — the 9-sector SSIC breakdown cannot capture new job categories or sector restructuring that AI might catalyse.
  • Nominal values — fiscal equations use nominal values throughout. No inflation adjustment is applied, which may overstate real growth in revenue/spending.
  • Singapore-specific — the small, open, managed economy has unique characteristics (high foreign worker share, active government) that limit generalisability.