AI-driven productivity gains in Vietnam’s heavy industry lift GDP by roughly 1–2% through 2035, while rapid IT hardware investment strains electricity capacity and temporarily reduces household consumption; together the twin transition is macro-feasible but shifts output away from textiles and leather into heavy manufacturing and IT, imposing uneven adjustment costs on the manufacturing workforce.
AI-Driven Energy Efficiency versus AI-Induced Energy Demand: A Dynamic Computable General Equilibrium (CGE) Analysis of Vietnam’s Twin Transition
openalex
theoretical
low evidence
7/10 relevance
Summary only
summary available; pdf_status=not_found
DOI
Source
A Vietnam-calibrated dynamic CGE simulation finds that AI-driven TFP gains in heavy industry raise GDP and consumption substantially by 2035, while large AI-related IT hardware investment increases electricity demand and temporarily crowds out household consumption, and the combined Twin Transition is roughly macro-additive but causes sectoral displacement and uneven labor adjustment.
Artificial Intelligence (AI) acts as a double-edged sword in the energy transition: its deployment in heavy industry generates productivity dividends that reduce energy intensity (the “Green AI” effect), while the data centers and hardware required to run it surge electricity demand and strain emerging-economy power grids (the “Brown AI” effect). This paper quantifies both effects and their interaction for Vietnam using a 23-sector recursive dynamic Computable General Equilibrium (CGE) model calibrated to the 2019 Input-Output Table and simulated through 2035. Three scenarios are evaluated: Green AI (S2), modelled as a Total Factor Productivity (TFP) shock to heavy manufacturing and electricity; Brown AI (S3), modelled as an exogenous investment surge in Information Technology (IT) hardware and services; and a combined Twin Transition (S4). Green AI delivers a compounding GDP dividend of +0.98% by 2030 and +1.79% by 2035, with consumption rising +1.17% and +2.03%. Brown AI is macro-neutral in GDP terms but imposes a consumption cost of 0.42% by 2030 as infrastructure investment crowds out household expenditure. The Twin Transition is approximately macro-additive (GDP +1.06% by 2030 and +1.95% by 2035), with consumption recovering to +1.70% above baseline by 2035. The dominant structural trade-off is distributional: real exchange rate appreciation generated by Green AI’s export surge displaces Textiles ( 5.5%) and Leather and Footwear ( 16.1%), even as Heavy Manufacturing (+12.9%) and IT Hardware (+10.9%) expand substantially. Under S3 alone the Electricity sector expands only +0.10% by 2030 against a 14.87% per year IT investment surge, indicating a binding generation capacity that the Green AI productivity shock relaxes in S4. The Twin Transition is macro-feasible, but its adjustment costs fall unevenly on the manufacturing workforce.
Summary
Main Finding
AI creates both “Green” and “Brown” effects in a middle-income, export-oriented economy. Productivity-boosting AI in heavy industry (Green AI) raises GDP and consumption materially by 2030–2035. Rapid AI-driven investment in IT hardware and services (Brown AI) is roughly GDP-neutral but temporarily crowds out household consumption and stresses electricity capacity. Combining both (Twin Transition) is broadly macro-feasible and approximately additive in GDP gains, but produces important sectoral displacement and distributional adjustment costs—especially for labour in export-oriented light manufacturing.
Key Points
- Modelled scenarios (Vietnam, horizon to 2035):
- S2 (Green AI): TFP shock to heavy manufacturing and electricity.
- S3 (Brown AI): exogenous surge in IT hardware & services investment (14.87%/yr applied to IT investment).
- S4 (Twin Transition): both shocks together.
- Aggregate outcomes:
- Green AI (S2): GDP +0.98% by 2030; +1.79% by 2035. Consumption +1.17% by 2030; +2.03% by 2035.
- Brown AI (S3): GDP ≈ neutral, but consumption −0.42% by 2030 (investment crowds out household spending).
- Twin Transition (S4): approximately macro-additive — GDP +1.06% by 2030; +1.95% by 2035. Consumption rebounds to +1.70% above baseline by 2035.
- Sectoral and distributional effects:
- Real exchange rate appreciation from Green AI export gains displaces light export industries: Textiles −5.5%, Leather & Footwear −16.1%.
- Expansion concentrated in capital- and skill-intensive sectors: Heavy Manufacturing +12.9%, IT Hardware +10.9%.
- Under Brown AI alone, Electricity expands only +0.10% by 2030 despite the large IT investment surge, indicating binding generation capacity; the Green AI productivity shock relaxes this constraint in the combined scenario.
- Overall interpretation: the Twin Transition is feasible at the macro level but imposes uneven adjustment costs across sectors and workers.
Data & Methods
- Model: 23-sector recursive-dynamic Computable General Equilibrium (CGE) model of Vietnam.
- Calibration: 2019 Input–Output Table (base year).
- Simulation horizon: 2019 baseline projected through 2035.
- Scenario implementation:
- Green AI implemented as sectoral Total Factor Productivity (TFP) improvements in heavy manufacturing and electricity generation/distribution.
- Brown AI implemented as an exogenous, sustained surge in IT hardware & services investment (14.87% per year in IT investment).
- Twin Transition combines both shocks to capture interaction effects.
- Outcomes reported: GDP, consumption, real exchange rate, sectoral output changes; attention to capacity constraints (electricity generation) and crowding-out via the investment-savings identity.
Implications for AI Economics
- Complementarity vs trade-off: Productivity-enhancing AI (Green AI) and infrastructure-heavy AI deployment (Brown AI) have complementary long-run benefits but cause short- to medium-term distributional trade-offs. Policy sequencing matters: productivity gains help absorb infrastructure-driven demand shocks by relaxing supply constraints.
- Energy and grid planning: Rapid IT/data-center investment can stress generation capacity even when macro impacts are modest. Emerging economies should pair AI deployment plans with upfront electricity generation and network investment to avoid bottlenecks and consumption losses.
- Distributional policy: Exchange-rate-driven reallocation hurts export-oriented light manufacturing and associated labour. Targeted labour market policies (retraining, social protection, local adjustment assistance) are needed to manage transitional unemployment and inequality.
- Investment finance and macro effects: Public or private financing of large IT investment surges crowds out household consumption in the short run absent offsetting fiscal/monetary measures. Choice of financing instruments and timing can materially change welfare outcomes.
- Measurement and modelling guidance:
- Distinguish TFP/productivity channels from pure demand-side (energy/hardware) channels when assessing AI’s macro and energy impacts.
- Incorporate energy-sector constraints explicitly (generation capacity, transmission limits) when evaluating data-centre and hardware investments.
- Run sensitivity analyses on TFP magnitudes, investment growth rates, and financing assumptions; update calibration beyond 2019 when possible.
- Generalizability: Results likely apply to other emerging, export-oriented economies with constrained power systems; magnitudes will vary with sectoral structure, trade openness, and the flexibility of electricity supply.
- Research priorities: endogenous modelling of electricity expansion, carbon and emissions impacts of AI deployment, spatial distribution of data centers, and firm-level heterogeneity in adopting productivity-enhancing AI.
Assessment
Paper Typetheoretical
Evidence Strengthlow — Findings come from model-based counterfactuals rather than empirical causal identification; results depend heavily on calibration, assumed magnitudes and timing of TFP and investment shocks, elasticities, closure rules and other structural assumptions, and appear not to be validated against observed post-treatment data or natural experiments.
Methods Rigormedium — Use of a dynamic recursive CGE calibrated to a detailed 2019 IO table and multi-decade simulations is standard and appropriate for economy-wide scenario analysis; however, the approach relies on strong modelling assumptions (representative agents, functional forms, fixed coefficients in some sectors), specific shock parametrizations, and likely limited microfoundations and empirical validation, which constrain inference robustness.
SampleVietnam economy calibrated to the 2019 Input-Output Table with 23 sectors; dynamic recursive CGE simulations run from 2019 through 2035; scenarios: Green AI (TFP shock in heavy manufacturing and electricity), Brown AI (exogenous IT hardware and services investment surge — e.g. ~14.87% p.a. IT investment), and combined Twin Transition; outcomes reported include GDP, consumption, sectoral outputs, real exchange rate and sectoral employment/adjustment implications.
Themesproductivity labor_markets adoption
IdentificationComparative counterfactual simulation in a 23-sector recursive dynamic Computable General Equilibrium (CGE) model calibrated to Vietnam's 2019 input-output table: Green AI is implemented as an exogenous TFP shock in heavy manufacturing and electricity; Brown AI is implemented as an exogenous surge in IT hardware and services investment; impacts are inferred by comparing scenario outcomes (S2, S3, S4) to a baseline projection through 2035.
GeneralizabilitySingle-country (Vietnam) calibration — results may not generalize to economies with different sectoral structure, energy mix, or grid robustness, Calibrated to the 2019 IO structure; post-2019 structural changes are not captured, Results sensitive to assumed sizes/timelines of TFP shocks and IT investment surges and to elasticities and closure rules, Standard CGE assumptions (representative agents, market-clearing, limited firm/worker heterogeneity) omit micro-level distributional dynamics, Simplified representation of electricity grid constraints and data-center energy demand may under- or overstate Brown AI effects, No modeled policy responses (e.g., targeted retraining, grid investments, subsidies) that could materially alter adjustment costs
Claims (12)
| Claim | Direction | Outcome | Confidence & Evidence | Details |
|---|---|---|---|---|
| Green AI delivers a compounding GDP dividend of +0.98% by 2030 and +1.79% by 2035. Fiscal And Macroeconomic | positive | Gross Domestic Product (GDP) relative to baseline |
Reading fidelity
high
Study strength
medium
|
+0.98% by 2030; +1.79% by 2035
|
| Under Green AI, consumption rises by +1.17% by 2030 and +2.03% by 2035. Consumer Welfare | positive | Household consumption (real consumption) relative to baseline |
Reading fidelity
high
Study strength
medium
|
+1.17% by 2030; +2.03% by 2035
|
| Brown AI (S3) is macro-neutral in GDP terms but imposes a consumption cost of 0.42% by 2030 as infrastructure investment crowds out household expenditure. Consumer Welfare | mixed | GDP (macro-neutral assertion) and household consumption (consumption change) |
Reading fidelity
high
Study strength
medium
|
Consumption -0.42% by 2030; GDP described as macro-neutral (no reported percent change)
|
| The Twin Transition (combined S4) is approximately macro-additive, producing GDP +1.06% by 2030 and +1.95% by 2035. Fiscal And Macroeconomic | positive | GDP relative to baseline under combined scenario |
Reading fidelity
high
Study strength
medium
|
+1.06% by 2030; +1.95% by 2035
|
| Under the Twin Transition, consumption recovers to +1.70% above baseline by 2035. Consumer Welfare | positive | Household consumption relative to baseline under S4 |
Reading fidelity
high
Study strength
medium
|
+1.70% by 2035
|
| Green AI’s export surge causes real exchange rate appreciation that displaces output in Textiles by 5.5% and in Leather and Footwear by 16.1%, while Heavy Manufacturing expands by 12.9% and IT Hardware by 10.9%. Market Structure | mixed | Sector real output changes relative to baseline |
Reading fidelity
high
Study strength
medium
|
Textiles -5.5%; Leather & Footwear -16.1%; Heavy Manufacturing +12.9%; IT Hardware +10.9%
|
| Under S3 alone the Electricity sector expands only +0.10% by 2030 despite a 14.87% per year IT investment surge, indicating binding generation capacity that the Green AI productivity shock relaxes in S4. Market Structure | mixed | Electricity sector real output change by 2030 and IT investment growth rate |
Reading fidelity
high
Study strength
medium
|
Electricity +0.10% by 2030; IT investment surge 14.87% per year
|
| The Twin Transition is macro-feasible, but its adjustment costs fall unevenly on the manufacturing workforce. Employment | mixed | Distributional/adjustment costs borne by manufacturing workforce (qualitative model outcome) |
Reading fidelity
high
Study strength
medium
|
not reported
|
| Green AI is modelled as a Total Factor Productivity (TFP) shock applied to heavy manufacturing and electricity in the CGE experiments. Other | other | Model specification (TFP shock to specified sectors) |
Reading fidelity
high
Study strength
high
|
not reported
|
| Brown AI is modelled as an exogenous investment surge in IT hardware and services in the CGE experiments. Other | other | Model specification (exogenous IT investment surge) |
Reading fidelity
high
Study strength
high
|
IT investment surge 14.87% per year (reported)
|
| The analysis uses a 23-sector recursive dynamic CGE model calibrated to the 2019 Input-Output Table and simulated through 2035 for Vietnam. Other | other | Model architecture and calibration period |
Reading fidelity
high
Study strength
high
|
not reported
|
| Brown AI’s infrastructure investment crowds out household expenditure, causing the reported consumption cost. Consumer Welfare | negative | Mechanism: crowding-out effect on household consumption due to higher investment |
Reading fidelity
high
Study strength
medium
|
-0.42% consumption by 2030 attributed to crowding-out
|