China’s industrial-internet pilots lift firms’ green productivity but increase total pollution: digital connectivity boosts technical efficiency yet fuels output expansion that, on balance, raises emissions—especially in state-owned and manufacturing firms.

Main Finding

The Industrial Internet Construction (IIC) pilot program significantly raised firms’ green total factor productivity (GTFP), primarily via improvements in green technical efficiency (management and process optimization), but it also increased firms’ total pollution emissions—the result of a scale rebound (capacity expansion) that outweighed per‑unit intensity reductions. Green technological progress (frontier-shifting R&D) did not show a significant contribution in the study period. Heterogeneous effects: stronger GTFP and rebound effects in state-owned and manufacturing firms; muted effects in high‑tech firms. IIC generally stimulates green innovation, but that response is weaker among firms under faster growth pressure or with weaker market power/long-term orientation.

Key Points

• Dual, paradoxical effect: IIC delivers measurable “new quality productivity” gains (lower emissions intensity and higher efficiency) while producing a firm‑level “green paradox” where total emissions rise.
• Mechanism decomposition: GTFP gains come mainly from green technical efficiency (catch-up), not from green technological progress (frontier breakthroughs).
• Rebound mechanism: Efficiency gains lower marginal costs and free capacity is largely converted into higher output; the scale effect dominates the intensity (per‑unit) effect, producing higher total pollution (backfire).
• Heterogeneity:
  • Stronger policy effects (both efficiency gains and emissions rebound) in state‑owned enterprises and manufacturing firms.
  • High‑tech industries see smaller effects (less rebound, weaker emission increases).
  • Firms with weaker market power or shorter long‑term orientation face larger rebound/externality risks.
• Green innovation: IIC tends to promote firm green innovation overall, but firms under rapid growth pressure exhibit a weaker green‑innovation response.
• Policy relevance: Evaluating digital infrastructure only by intensity or efficiency metrics can miss hidden environmental costs from scale expansion.

Data & Methods

• Quasi‑natural experiment: Multi‑batch Industrial Internet pilot demonstration projects announced by China’s MIIT since 2017 used as staggered policy shocks.
• Unit of analysis: Firm‑level panel (multi-period), exploiting variation in timing of firms’ inclusion in IIC pilots.
• Identification strategy: Multi‑period difference‑in‑differences (DID) with firm and year fixed effects, controlling for firm-level covariates (Xit) to estimate causal impacts of IIC on outcomes.
• Outcome measures:
  • Green total factor productivity (GTFP) computed using an SBM‑ML (Slack‑Based Measure — Malmquist‑Luenberger) index that incorporates undesirable outputs (pollutants).
  • Total pollution emissions (aggregate firm pollutant quantity) rather than only intensity or carbon per unit output.
• Decomposition: GTFP decomposed into green technical efficiency (catch‑up) and green technological progress (frontier shift) to identify which channel drove productivity changes.
• Additional analyses: Heterogeneity tests by ownership (SOE vs non‑SOE), industry (manufacturing vs others, high‑tech vs non‑high‑tech), firm market power, and long‑term orientation; examination of green innovation responses and interaction with growth pressure.

Implications for AI Economics

• Rebound risk is a central economic externality of AI/digital adoption: Productivity and efficiency gains from AI (or industrial internet) can trigger output expansion that raises total environmental harms unless externalities are internalized.
• Measurement matters: Evaluations of AI’s environmental impact should combine intensity metrics (efficiency) with total‑volume measures and decompose productivity gains into catch‑up vs frontier effects to understand short‑ vs long‑term implications.
• Policy design: Digital infrastructure and AI diffusion policies must be paired with environmental instruments (e.g., emissions caps/pricing, stricter monitoring, sectoral limits) to prevent scale rebound that negates per‑unit efficiency gains.
• Incentives for green frontier innovation: Because short‑term AI/IIC gains are mainly efficiency/catch‑up, promoting longer‑horizon green R&D (subsidies, tax credits, public‑private partnerships) is essential to shift the technological frontier and achieve sustained emission reductions.
• Targeting and heterogeneity: AI deployment strategies should be sector‑sensitive (manufacturing more prone to rebound) and ownership‑aware (SOEs behaved differently). Policies can be tailored—e.g., impose stronger environmental constraints or green innovation incentives where rebound risk is highest.
• Firm incentives and governance: Strengthening firms’ long‑term orientation and market incentives for green products (green procurement, carbon disclosure, ESG-linked finance) can reduce the likelihood that efficiency gains translate into emission backfires.
• Research guidance: Empirical AI economics should use firm‑level causal designs (staggered DID, careful staggered-treatment inference) and incorporate environmental outcomes at both intensity and scale levels to capture hidden externalities of digital adoption.