The Commonplace
Home Papers Evidence Explore Trends Syntheses Digests References Docs 🎲 Workforce Futures
← Papers
Direction, evidence grade, and study type are AI-generated labels (gpt-5-mini), not human-verified. Syntheses are LLM-written. "Tensions" are machine-detected candidates, not confirmed contradictions. A research-acceleration tool, not peer review. How this is built →

Countries with stronger government AI readiness show higher national energy use; the build-out of AI infrastructure, data capacity and compute appears to raise aggregate energy demand in advanced economies, suggesting AI expansion currently outweighs short-term efficiency gains.

The influence of artificial intelligence readiness on energy consumption patterns: Panel evidence from high-income countries
Sirinya Padpuy, Chatchai Khiewngamdee · January 01, 2026 · E3S Web of Conferences
openalex correlational low evidence 7/10 relevance Full text usable extracted full text DOI Source PDF

Structured author observations

Linked only from stored provider relations; the raw author line above is never matched by name.

OpenAlex

Latest observation:

  1. Sirinya Padpuy provider ID
  2. Chatchai Khiewngamdee exact ORCID

Semantic Scholar

Latest observation:

  1. Sirinya Padpuy provider ID
  2. Chatchai Khiewngamdee provider ID
Using a balanced panel of 30 high-income countries (2019–2023), the study finds that higher Government AI Readiness scores are positively and significantly associated with greater national primary energy consumption, controlling for GDP per capita and structural characteristics.

Citation observations

Cumulative provider counts captured on specific dates; providers are never combined.

This study examines the effect of AI readiness, proxied by the Government AI Readiness Index developed by Oxford Insights, on primary energy consumption in high-income countries. Using balanced panel dataset covering 30 high-income economies over the period 2019-2023, the analysis employs panel regression techniques to assess how government-led preparedness for artificial intelligence adoption influences aggregate energy demand while controlling for income levels and structural characteristics. The empirical results reveal a positive and statistically significant relationship between AI readiness and primary energy consumption, indicating that the expansion of AI-related infrastructure, data capacity, and computational intensity currently outweighs potential energy-efficiency gains associated with AI applications. In contrast, GDP per capita exhibits a negative association with energy consumption, consistent with efficiency improvements and structural transitions toward less energy-intensive activities in advanced economies, while industry share does not display a systematic effect. These findings suggest that, even in relatively energy-efficient high-income countries, advancing AI readiness may exert upward pressure on national energy demand. Consequently, policies promoting AI development should be closely aligned with energy efficiency strategies, low-carbon electricity deployment, and sustainable digital infrastructure planning to ensure that technological progress supports long-term environmental and climate objectives.

Summary

Main Finding

The paper finds a robust, positive association between national AI readiness and aggregate primary energy consumption in 30 high-income countries (2019–2023). The preferred random-effects estimate implies an elasticity of roughly 0.30: a 1% increase in government AI readiness is associated with about a 0.30% increase in primary energy consumption. By contrast, GDP per capita is negatively associated with energy use (consistent with efficiency/structural change), while the industry share of GDP is not systematically related to energy consumption once AI readiness and income are controlled for.

Key Points

  • Sample and period: Balanced panel of 30 high‑income countries, 2019–2023 (150 country-year observations).
  • Core result: AI readiness (Oxford Insights Government AI Readiness Index) raises national primary energy use; RE elasticity ≈ +0.301 (FE estimate similar at ≈ +0.289).
  • Controls: log(GDP per capita) (from WDI) and industry share of GDP. GDP per capita shows a negative association in pooled OLS and in discussion, but its coefficient in panel RE/FE estimates is small and not economically large; industry share is statistically insignificant in panel models.
  • Econometric diagnostics: IPS panel unit-root tests indicate variables are stationary in levels (I(0)). F-test rejects pooled OLS (country effects present). Hausman test (χ² = 0.285) fails to reject RE consistency, so random effects used as baseline.
  • Interpretation: At current stages, energy costs of expanding AI infrastructure/data centers/computation and rebound/scale effects outweigh AI-driven energy-efficiency gains at the national level in advanced economies.
  • Policy emphasis in paper: coordinate AI policy with energy-efficiency strategies, low-carbon electricity deployment, and sustainable digital infrastructure planning.

Data & Methods

  • Dependent variable: Primary energy consumption (Our World in Data).
  • Key explanatory variable: Government AI Readiness Index (Oxford Insights) — composite of government, technology sector, and data & infrastructure pillars.
  • Controls: GDP per capita (constant international $) and industry value‑added share (% of GDP) from World Bank WDI.
  • Specification: Log-linear panel regression: ln(Energy) = α + β1 ln(AI readiness) + β2 ln(GDP per capita) + β3 IndustryShare + country effects + error.
  • Pre-tests: Im–Pesaran–Shin (IPS) panel unit-root tests for stationarity (all variables I(0)).
  • Estimators compared: pooled OLS, fixed effects (FE), random effects (RE). F-test indicates country fixed effects; Hausman test favors RE (χ² = 0.285 → fail to reject null of no systematic difference).
  • Main robustness: coefficient on ln(AI readiness) is positive and significant across specifications (bivariate, pooled OLS, FE, RE), with FE ≈ 0.289 and RE ≈ 0.301.

Implications for AI Economics

  • Aggregate energy footprint: Macroeconomic models of AI adoption should incorporate the energy costs of infrastructure (data centers, cloud compute, edge devices) and potential rebound/scale effects; micro-level efficiency gains may not translate into lower national energy demand.
  • Externalities & policy design: AI-related energy consumption represents a growing environmental externality. Policymakers should pair AI promotion with:
    • Investments in energy-efficient computing hardware and algorithms,
    • Grid decarbonization and expanded low-carbon electricity supply,
    • Planning for sustainable digital infrastructure (cooling, siting, renewable procurement),
    • Regulatory and market instruments (e.g., carbon pricing, green procurement standards for public AI deployments).
  • Heterogeneity and modelling: Effects may differ by country energy mixes, data-center concentration, and sectoral AI adoption patterns. AI economics work should allow for heterogeneous effects across contexts and incorporate infrastructure-level constraints.
  • Research directions: prioritize causal identification (instrumental variables, difference-in-differences around policy shocks), longer time series, disaggregated energy outcomes (electricity demand, data-center footprints), inclusion of energy prices and renewables share, and firm- or sector-level studies linking AI deployment intensity to energy use and emissions.
  • Short-term takeaway for economists and policymakers: advancing AI readiness in high-income countries can raise national energy demand; to ensure AI supports sustainability goals, technological and industrial policy must be integrated with energy and climate policy.

Assessment

Paper Typecorrelational Evidence Strengthlow — The analysis is associative rather than causal: it relies on observational panel regressions without an exogenous source of variation, instrumental variables, or a credible quasi-experimental design; risks include omitted variable bias, reverse causality (energy-intensive economies may invest more in AI readiness), measurement error in the readiness index, and limited time variation (2019–2023). Methods Rigormedium — Appropriate use of panel data and controls for key covariates increases credibility relative to simple cross-sections, but the short time span, modest sample (30 countries), lack of a clearly stated fixed-effects or other robustness checks in the description, and absence of strategies to address endogeneity weaken methodological rigor. SampleBalanced panel of 30 high-income countries observed annually from 2019 to 2023 (country-year units); main variables are Oxford Insights Government AI Readiness Index, national primary energy consumption, GDP per capita, and sectoral composition (industry share), with additional structural controls reported. Themesadoption innovation IdentificationPanel regression on a balanced country-year panel (30 high-income countries, 2019–2023) using the Oxford Insights Government AI Readiness Index as the key explanatory variable and controlling for GDP per capita and structural characteristics (e.g., industry share); identification relies on cross-sectional and time variation with standard covariate adjustment (no instrumental variables or natural experiment described). GeneralizabilityRestricted to high-income countries — results may not apply to low- or middle-income contexts, Short time window (2019–2023) captures early AI deployment dynamics and may not reflect longer-run effects, National-aggregate analysis masks sectoral and firm-level heterogeneity in AI adoption and energy use, The Government AI Readiness Index is a proxy for AI adoption/compute intensity and may not capture actual deployed AI workloads or their energy efficiency, Energy mix and policy contexts differ across countries, limiting transferability of point estimates

Claims (8)

ClaimDirectionOutcomeConfidence & EvidenceDetails
AI readiness (proxied by the Government AI Readiness Index) is positively and statistically significantly associated with higher national primary energy consumption in high-income countries. Fiscal And Macroeconomic positive primary energy consumption
Reading fidelity high
Study strength medium
n=150
0.3
The expansion of AI-related infrastructure, data capacity, and computational intensity currently outweighs potential energy-efficiency gains from AI applications, contributing to higher energy demand. Fiscal And Macroeconomic positive primary energy consumption
Reading fidelity high
Study strength speculative
n=150
0.05
GDP per capita exhibits a negative association with primary energy consumption, consistent with efficiency improvements and structural transitions toward less energy-intensive activities in advanced economies. Fiscal And Macroeconomic negative primary energy consumption
Reading fidelity high
Study strength medium
n=150
0.3
Industry share does not display a systematic effect on national primary energy consumption in the estimated models. Fiscal And Macroeconomic null_result primary energy consumption
Reading fidelity high
Study strength medium
n=150
0.3
Even in relatively energy-efficient high-income countries, advancing AI readiness may exert upward pressure on national energy demand. Fiscal And Macroeconomic positive national energy demand / primary energy consumption
Reading fidelity high
Study strength medium
n=150
0.3
Policy implication: Policies that promote AI development should be closely aligned with energy-efficiency strategies, low-carbon electricity deployment, and sustainable digital infrastructure planning to ensure technological progress supports long-term environmental and climate objectives. Fiscal And Macroeconomic positive mitigation of AI-induced increases in national energy demand / environmental outcomes
Reading fidelity high
Study strength speculative
not reported
0.05
The study uses a balanced panel dataset covering 30 high-income economies over the period 2019–2023. Other null_result data coverage (countries and years)
Reading fidelity high
Study strength high
n=30
0.5
The empirical analysis employs panel regression techniques and controls for income levels and structural characteristics to assess the relationship between AI readiness and aggregate energy demand. Other null_result estimation approach / relationship between AI readiness and energy consumption
Reading fidelity high
Study strength high
n=150
0.5

Notes