AI-led digital transformation is linked to substantial declines in manufacturers' carbon intensity—coefficients imply large reductions and up to 42% lower carbon intensity in high-DTI firms—with green innovation accounting for nearly 39% of the benefit and supply-chain spillovers amplifying effects.

Main Finding

AI-facilitated digital transformation substantially reduces firm-level carbon intensity in Chinese manufacturing supply chains. Using a composite Digital Transformation Index (DTI), the authors find statistically significant reductions in carbon intensity (coefficients between −0.689 and −1.245, p < 0.01). Green innovation mediates roughly 38.6% of the DTI effect. There are also sizable spatial spillovers through supply‑chain and geographic networks (≈21.7% of spillover effects). Heterogeneous effects favor firms with stronger regulatory pressure and higher absorptive capacity; high‑DTI firms experienced a 42.3% reduction in carbon intensity versus 18.7% for low‑DTI firms.

Key Points

• Sample and scope: 450 Chinese manufacturing firms, 2015–2024, 4,500 firm‑year observations; Scope 1–3 emissions included.
• DTI construction: composite index of AI adoption, automation, digital maturity and tech investments (PCA used to combine indicators). AI‑enabling tech categories: ML/predictive analytics, NLP/text mining, intelligent automation, smart logistics, sensor monitoring, real‑time analytics.
• Main quantitative results:
  • DTI → lower carbon intensity: coefficients −0.689 to −1.245 (p < 0.01) across model specifications.
  • Mediation: green innovation accounts for ~38.6% of the environmental benefit from digital transformation.
  • Spatial spillovers: ~21.7% of DTI effects propagate via supply‑chain and geographic networks.
  • Heterogeneity: high‑DTI firms cut carbon intensity by 42.3%; low‑DTI firms by 18.7%; regulatory stringency and absorptive capacity amplify gains.
• Sectoral coverage: automotive (15%), electronics (18%), chemical/pharma (12%), textiles (10%), food & beverage (20%), metals/steel (10%), other manufacturing (15%).
• Controls: firm size, R&D intensity, industry energy intensity and other firm characteristics included in models.
• Framing: authors present an Integrated AI‑Driven Sustainability Analytics Framework (IASAF) — five modules linking multi‑source data, AI processing, econometric modeling, performance assessment, and feedback.

Data & Methods

• Data sources:
  • Firm supply‑chain operational data (inventory, supplier networks, transport/logistics metrics).
  • Organizational carbon records (internal disclosures, CDP, emissions databases) covering Scope 1–3.
  • Surveys and investment records for AI/digital technology metrics.
• DTI measurement: multi‑indicator composite via Principal Component Analysis (PCA) combining AI adoption, automation level, digital maturity, and investment measures.
• Empirical approaches:
  • Panel econometrics on 4,500 firm‑year observations (fixed effects and robustness checks reported).
  • Mediation analysis to quantify the contribution of green innovation intensity to carbon reductions.
  • Spatial econometric analysis to capture network and geographic spillovers (authors report supply‑chain/geography‑based spillover share).
  • Machine learning methods used in data processing, variable selection, and validation (no single ML algorithm specified in the summary; “state‑of‑the‑art” ML applied).
  • Robustness checks and iterative validation embedded in IASAF.
• Limitations noted or implicit:
  • Observational panel data—causal inference is supported by controls and robustness tests but not randomized assignment.
  • DTI relies on PCA and survey/administrative indicators → potential measurement error.
  • Sample limited to Chinese manufacturing; external validity to other countries/sectors may be limited.

Implications for AI Economics

• Externalities and network effects: AI-driven digitalization produces positive environmental externalities that spill across supply‑chain links and geography; policy design should account for these network spillovers (e.g., incentives that target lead firms can propagate benefits).
• Mechanism: a large share of the carbon‑reduction benefit operates via green innovation (≈38.6%), implying returns to complementary investments in green R&D and knowledge diffusion.
• Heterogeneous returns: firms with greater absorptive capacity and under stricter regulation capture larger benefits — targeted capacity‑building (skills, data infrastructure) and regulatory pressure can increase aggregate effectiveness of AI adoption.
• Policy levers:
  • Subsidies/tax incentives for AI + green innovation investments.
  • Encourage digital standards and data sharing across supply chains to amplify spillovers.
  • Strengthen carbon accounting (Scope 1–3) and transparency to better measure impacts and inform regulation.
• Firm strategy:
  • Prioritize AI investments that link operational optimization with green R&D (logistics optimization, predictive maintenance, sensor data for energy use).
  • Build absorptive capacity (training, data governance) to capture larger environmental returns.
• Research agenda for AI economics:
  • Causal identification: randomized or quasi‑experimental interventions on AI adoption to isolate causal effects.
  • Cost‑benefit analysis: quantify financial returns and transition costs across firm types and labor markets.
  • Cross‑country comparisons: test generalizability across regulatory regimes and infrastructure contexts.
  • Distributional effects: study how AI‑driven decarbonization affects smaller suppliers, labor demand, and regional inequality.
  • Model integration: link micro‑level firm adoption models with macro carbon accounting to simulate policy scenarios.

Overall, the study provides empirical evidence that AI‑enabled digital transformation is a consequential lever for industrial decarbonization, with substantial mediated benefits through green innovation and meaningful spillovers across networks — important considerations for economists designing policy and firm strategies in the transition to low‑carbon production.