Can digital transformation curtail carbon emissions? Evidence from a quasi-natural experiment

China contributes over one-third of global carbon emissions and has pledged to peak carbon emissions by 2030. Traditional environmental regulations can impose economic costs and may be less effective in economies heavily reliant on fossil fuels, creating a need for policies that support both growth and decarbonization. The digital economy and digital infrastructure construction (DIC) are proposed as potential avenues to reduce emissions by enabling intelligent energy management, dematerializing consumption and work patterns, and supporting environmental monitoring and regulation. However, ICT infrastructure is energy-intensive and prior work is mixed on digitalization’s environmental effects, with limited causal evidence for DIC specifically. This study asks whether China’s Broadband China Pilot (BBCP) policy—a large-scale DIC initiative—reduces urban carbon emissions, and examines mechanisms via industrial structure upgrading and green technological innovation. Hypotheses: H1, BBCP reduces urban COE; H2, BBCP reduces COE by upgrading industrial structure; H3, BBCP reduces COE by advancing green technological innovation.

Prior research links population growth and energy consumption to rising carbon emissions. Studies on the digital economy (DE) and carbon emissions (COE) report heterogeneous effects: inverted U-shaped relationships where DE raises COE initially then reduces it; potential carbon rebound effects; DE agglomeration reducing COE; and cross-country evidence of reduced carbon intensity but higher per capita emissions. Much literature examines traditional infrastructure’s impacts on COE, whereas DIC’s impact remains underexplored, with some arguing that ICT’s energy intensity could undermine long-term carbon neutrality. Existing studies may capture emissions generated by DE itself while overlooking spillovers whereby DIC improves efficiency in other sectors (e.g., energy). Emerging evidence suggests DIC promotes industrial ecology, regional innovation, firm digital transformation, and public health, but its role in fostering low-carbon cities and improving low-carbon ecological efficiency has received limited attention. Theoretically, DIC can reduce COE by: (1) enabling digital energy-saving solutions (smart grids, smart transport, AI-based energy scheduling, better energy structure favoring renewables), (2) dematerializing production and lifestyles (e-commerce, teleconferencing, paperless processes), and (3) strengthening environmental monitoring and regulatory enforcement. Indirectly, DIC may reduce COE by upgrading industrial structure (shifting from resource- and fossil-intensive secondary industries to services and ICT) and by spurring green technological innovation (GTI) through improved information flows, R&D capabilities, and diffusion of clean technologies. These foundations motivate H1–H3.

Study design: A quasi-experimental Difference-in-Differences (DID) approach evaluates the causal impact of the Broadband China Pilot (BBCP) policy on urban carbon emissions. Sample and period: Panel of Chinese cities from 2009–2019. BBCP pilot cities were announced in three batches (2014–2016), totaling 117 cities. Data sources: China City Statistical Yearbook, China Urban Construction Statistical Yearbook, China National Development and Reform Commission, and China National Intellectual Property Office. Key variables: - Outcome: City-level carbon emissions (COE), computed by converting annual consumption of coal gas, methane, LPG, electricity, and heat using 2006 IPCC Guidelines, aggregated to total COE (kilotons), then log-transformed. - Treatment: BCPP, a dummy equal to 1 for a city in and after the year it becomes a BBCP pilot, 0 otherwise. - Controls: Economic development (lnGDP), population (lnPOP), foreign direct investment (FDI/GDP), fiscal expenditure (FinExp/GDP), and infrastructure (per capita urban road area). Econometric specification: Two-way fixed effects DID with city and year fixed effects and clustered standard errors at the city level. The coefficient on BCPP captures the average treatment effect on treated cities’ COE. Event-study specifications assess parallel trends. Robustness strategies include a placebo test with 500 random policy shocks, propensity score matching DID (1:1 nearest-neighbor using all controls), and a local projection DID to address potential confounding in staggered adoption settings. Extensions: - Heterogeneity analyses by green fund share (high vs. low based on median), resource dependence (resource-based vs. non-resource cities), and fiscal expenditure (high vs. low based on median). - Additional outcomes: Low-carbon ecological efficiency (super-efficiency slacks-based measure with inputs: energy, capital, labor; desirable outputs: GDP, per capita green space; undesirable outputs: COE, SO2, dust, wastewater) and COE intensity (COE/GDP). - Mechanisms: Industrial structure (Structure = tertiary/secondary output ratio) and green technological innovation (GreenPat = green patent applications per 10,000 people). Interaction models test whether BBCP’s COE effects operate via Structure and GreenPat, alongside effects of BBCP on these mediators.

- Baseline DID: BBCP significantly reduces urban COE by about 9.4% (BCPP coefficient −0.094, t = −2.159) with rich controls, city and year fixed effects (N = 2701; R2 = 0.925). Among controls, lnGDP is positively associated with COE (0.555, t = 5.396); FinExp is positive (0.781, t = 2.131); infrastructure is slightly negative (−0.005, t = −1.800). - Parallel trends: Event-study shows no significant pre-trends; post-policy, COE declines in pilot cities, with effects strengthening in years 3–4. - Placebo test: 500 random policy simulations yield coefficients substantially higher than the actual estimate; 100% of placebo estimates exceed the true effect, implying the observed COE reduction is unlikely due to random shocks (>99% confidence). - PSM-DID: After 1:1 nearest-neighbor matching, BBCP remains significantly negative on COE (−0.105, t = −2.351; N = 1998; R2 = 0.933). - Local projection DID: Treatment effects remain significantly negative across horizons, indicating persistent COE reductions following BBCP implementation. - Heterogeneity by green finance: Interaction BCPP × High GreenFund is negative and significant (−0.096, t = −1.894), indicating stronger COE reduction where the share of green funds is higher; main BCPP effect remains negative (−0.096, t = −2.222). - Heterogeneity by resource dependence: No significant effect in resource-based cities (−0.032, t = −0.369; N = 1079; R2 = 0.877); significant reduction in non-resource cities (−0.115, t = −2.348; N = 1622; R2 = 0.945). - Heterogeneity by fiscal expenditure: Insignificant in high fiscal expenditure cities (0.029, t = 0.373; N = 1339; R2 = 0.912); larger and significant reduction in low fiscal expenditure cities (−0.161, t = −2.908; N = 1337; R2 = 0.914). - Low-carbon ecological efficiency: BBCP significantly increases low-carbon eco-efficiency. - COE intensity: BBCP significantly reduces COE intensity (COE/GDP) in pilot cities. - Mechanisms: • Industrial structure: BBCP increases Structure; interaction BCPP × Structure is significantly negative, indicating that industrial upgrading is a channel through which BBCP lowers COE. • Green technological innovation: BBCP raises GreenPat; interaction BCPP × GreenPat is significantly negative, indicating GTI is a channel reducing COE.

The findings confirm that digital infrastructure policy (BBCP) causally reduces urban carbon emissions while supporting economic activity, addressing the study’s central question about whether digital transformation can contribute to decarbonization. The results are robust to multiple checks and show increasing effects over time, indicating that DIC can help cities transition toward sustainable, low-carbon development. Policy implications include: leveraging DIC to build low-carbon cities; using tax incentives and regulatory support to attract private investment into digital infrastructure; aligning DIC with green finance to amplify emissions reductions; and prioritizing supportive environments for GTI to sustain long-term decarbonization. The heterogeneity results suggest that DIC may be especially effective in non-resource-dependent cities and where fiscal capacity is limited but private capital can be mobilized, whereas resource-dependent cities may face barriers (resource curse) that constrain DIC’s environmental benefits. Enhancing low-carbon ecological efficiency and reducing COE intensity indicate that digital transformation can contribute to economic decarbonization and more efficient resource use.

Using city-level data from 2009–2019 and a DID design, the study finds that the Broadband China Pilot policy significantly reduces urban carbon emissions. The effects are robust across parallel trends, placebo, PSM-DID, and local projection DID tests. BBCP’s emissions reductions are stronger in non-resource-dependent cities, cities with higher proportions of green funds, and cities with lower fiscal expenditure. BBCP also improves low-carbon ecological efficiency and lowers COE intensity. Mechanism analyses indicate that BBCP reduces COE by upgrading industrial structure and fostering green technological innovation. Overall, digital infrastructure construction emerges as a viable green development strategy that can support economic growth while curbing emissions, particularly by catalyzing services and ICT sectors and stimulating green innovation.

Data limitations preclude separating industrial versus household carbon emissions; future work could assess sector-specific impacts, especially given manufacturing’s likely responsiveness to DIC. The study also does not analyze how DIC affects the integration of the digital economy with the real economy, which may further influence COE as digitalization matures. Exploring these dimensions and addressing the resource curse constraints on DIC’s environmental effectiveness are important avenues for future research.