Carbon Reduction Impact of the Digital Economy: Infrastructure Thresholds, Dual Objectives Constraint, and Mechanism Optimization Pathways

Abstract

The synergistic advancement of “Digital China” and “Beautiful China” represents a pivotal national strategy for achieving high-quality economic development and a low-carbon transition. To illuminate the intrinsic mechanisms linking the digital economy (DE) to urban carbon emission performance (CEP), this study develops a novel two-sector theoretical framework. Leveraging panel data from 278 Chinese prefecture-level cities (2011–2023), we employ a comprehensive evaluation method to gauge DE development and utilize calibrated nighttime light data with downscaling inversion techniques to estimate city-level CEP. Our empirical analysis integrates static panel fixed effects, panel threshold, and moderating effects models. Key findings reveal that the digital economy demonstrably enhances urban carbon emission performance, although this positive effect exhibits a threshold characteristic linked to the maturity of digital infrastructure; beyond a specific developmental stage, the marginal benefits diminish. Crucially, this enhancement operates primarily through the twin engines of fostering technological innovation and driving industrial structure upgrading, with the former playing a dominant role. The impact of DE on CEP displays significant heterogeneity, proving stronger in northern cities, resource-dependent cities, and those characterized by higher levels of inclusive finance or lower fiscal expenditure intensities. Furthermore, the effectiveness of DE in reducing carbon emissions is dynamically moderated by policy environments: flexible economic growth targets amplify its carbon reduction efficacy, while environmental target constraints, particularly direct binding mandates, exert a more pronounced moderating influence. This research provides crucial theoretical insights and actionable policy pathways for harmonizing the “Dual Carbon” goals with the overarching Digital China strategy.

1. Introduction

Against the backdrop of accelerating global climate governance and China’s deepening “Dual Carbon” (Peak Carbon, Carbon Neutrality) strategy, the digital economy (DE) has emerged as a disruptive force, fundamentally reshaping production factors and economic structures. The synergistic relationship between DE and urban low-carbon transitions requires rigorous scientific scrutiny. President Xi Jinping’s commitment at the 2020 UN General Assembly (“Carbon Peak by 2030, Carbon Neutrality by 2060”) and the emphasis in the 20th CPC National Congress Report on “promoting green and low-carbon economic development as a cornerstone of high-quality growth” underscore the strategic urgency of this transition in China. However, persistent path dependency in traditional high-carbon development models necessitates investigation into whether DE can resolve the inherent trade-off between decarbonization and economic growth—a core challenge for achieving synergistic dual goals [1]. While China’s ambitious “Dual Carbon” goals and their scale make them a critical and instructive testbed, the fundamental tension between decarbonization imperatives and the pursuit of economic prosperity represents a universal challenge confronting nations worldwide, particularly those navigating rapid industrialization and urbanization. The disruptive potential of the Digital Economy (DE)—with its capacity to enhance efficiency, optimize resource allocation, foster innovation, and enable data-driven governance—offers a globally relevant toolkit for reconfiguring development pathways towards sustainability. However, a significant knowledge gap persists regarding the precise mechanisms, enabling conditions, and potential trade-offs through which DE can effectively unlock synergistic low-carbon transitions across diverse urban and national contexts. Rigorous scientific inquiry into the DE-urban low-carbon nexus holds profound significance beyond China’s borders. By dissecting the dynamics within this pivotal case, this research aims to generate transferable insights and empirical evidence that can inform policy design and strategic planning for sustainable urban development globally. Understanding how DE can help overcome persistent high-carbon path dependencies is crucial not only for China’s success but also for empowering cities and nations everywhere to achieve their climate commitments under the Paris Agreement while pursuing inclusive and resilient economic growth, contributing directly to the global Sustainable Development Goals (SDGs).

The digital economy (DE) is fundamentally reshaping carbon emission dynamics. While technologies such as smart grids and industrial IoT facilitate low-carbon transitions through energy system optimization and green innovation diffusion [2,3,4,5], the rapid expansion of digital infrastructure—including data centers and 5G base stations—drives substantial energy demand. China’s data centers alone consumed 2.7% of national electricity in 2023 [6,7,8]. This dichotomy fuels ongoing academic debate regarding whether digital progress inherently drives decarbonization. Such environmental dualism highlights DE’s dual role: it may mitigate the “Green Paradox” [9], while simultaneously risking carbon footprint amplification through rebound effects.

Current research exhibits three critical gaps: First, mechanistic opacity persists due to an overreliance on linear models linking the digital economy to emissions [10], neglecting diminishing marginal effects arising from digital infrastructure accumulation [11]. Second, contextual disembodiment occurs when China’s unique dual-target constraint framework (economic growth vs. environmental goals) is not incorporated, hindering analysis of policy distortions by local governments under “promotion tournament” incentives [12,13]. Third, methodological limitations exist, including dependence on provincial macro-data for validating direct effects [8] and a lack of quantitative dissection of the transmission chain: “technological innovation → industrial structural upgrading → Carbon Emission Performance (CEP) enhancement”.

The marginal contributions of this paper are primarily manifested in the following three aspects:

First, theoretical mechanisms and nonlinear effects are revealed. This study systematically elucidates the intrinsic mechanisms through which the digital economy influences carbon emission performance from a theoretical perspective. Utilizing an improved mediating effect model, it identifies and quantifies the dual mediating mechanisms of structural optimization effects and technological innovation effects, along with their respective contribution levels. This research further confirms that, under the moderating influence of digital infrastructure, this facilitating effect exhibits a nonlinear pattern characterized by diminishing marginal effects. This finding not only enriches the theoretical research on the environmental effects of the digital economy but also expands the literature on data technology support within the field of environmental governance.

Second, the moderating role of dual-target constraints is examined. This paper empirically tests the moderating effect of “dual-target constraints” (i.e., economic growth targets vs. environmental targets). The research specifically distinguishes between different constraint types and their impacts: economic growth targets (rigid constraints vs. flexible constraints) and environmental targets (direct constraints vs. indirect constraints).

Third, heterogeneity analysis and differentiated policy implications. Based on the significant heterogeneity among cities in terms of locational characteristics, resource endowments, and fiscal capacity, this study proposes tailored policy packages for digital carbon reduction. This provides more targeted decision-making support for the synergistic advancement of two major national strategies: “Digital China” and “Beautiful China”.

2. Theoretical Analysis and Research Hypotheses

2.1. Direct Impact: Efficiency Revolution and Governance Enablement

Digitalization and smart technologies demonstrate distinct advantages in terms of efficiency and cost-effectiveness for enabling urban low-carbon transitions, establishing the digital economy as a key driver of decarbonization [14].

Government governance benefits from enhanced low-carbon management capabilities. Digital technologies enable precise tracking of energy markets and price dynamics, and carbon trading data platforms reduce transaction costs through improved information matching. This strengthens the governmental capacity to balance energy supply and demand via pricing mechanisms and cross-subsidies, while providing critical support for emissions monitoring, carbon planning, and policy implementation [15].

Industrial transformation leverages digital network effects (“Metcalfe’s Law”) through deep integration with high-emission sectors. This diffusion of digital solutions: (1) reallocates production factors across efficiency gradients independent of spatial constraints, enabling intelligent resource management and green innovation [10]; and (2) reduces lifecycle energy consumption and emissions in key industries [16]. Exemplars include smart grids that minimize transmission losses and big data that optimize logistics resource use.

Residential behavior shifts emerge as digital exposure reshapes consumption paradigms [17]. Pervasive technology adoption elevates low-carbon awareness and green consumption preferences, fostering spontaneous adoption of sustainable lifestyles at the individual level [18].

Consequently, we propose the following:

Hypothesis 1.

The digital economy exerts a significant positive influence on urban carbon-emission performance.

2.2. Indirect Pathways: Dual Drivers of Innovation and Restructuring

As a pivotal engine driving the new wave of economic growth, the digital economy integrates the dual attributes of profound technological transformation and extensive digital empowerment. Its development can significantly enhance carbon emission performance (defined as economic output or environmental benefit per unit of carbon emission) through two core pathways: technological innovation and industrial structure upgrading.

Regarding the effect of technological innovation, the digital economy, particularly by fostering green technological progress, serves as a central mechanism for improving carbon emission performance [19]. This manifests in two primary dimensions: At the macro level, the deep integration and synergistic innovation between digital technologies (e.g., big data, the Internet of Things, and artificial intelligence) and energy development/utilization technologies accelerate the coupling process between the digital economy and the real economy. This convergence not only catalyzes the emergence of novel low-carbon transition technologies (e.g., smart grid optimization, digital twin-driven energy efficiency management, and precise carbon accounting systems) but also empowers the real economy to transition towards low-carbon, environmentally friendly production models (such as smart manufacturing and flexible production). This fundamental shift reduces carbon emission intensity and optimizes performance, aligning with the technology-driven environmental improvement mechanism proposed by the Environmental Kuznets Curve (EKC). At the micro level, low-cost, high-efficiency information networks and platforms established by the digital economy (e.g., industrial internet platforms and open innovation communities) greatly facilitate the rapid diffusion and spillover of green, low-carbon technologies and best practices along the industrial chain and across sectors. This not only drives the parallel transformation of entire industries towards digitalization and decarbonization but also systematically reshapes the energy consumption structure of traditional high-energy-intensity production models through optimized resource allocation (e.g., precise matching of energy elements and sharing of idle production capacity). Concurrently, market signals (such as green supply chain requirements and carbon trading mechanisms) and efficiency pressures compel the wider and deeper penetration and application of clean technologies, thereby achieving corporate-level carbon reduction targets [20] (exemplifying the application of Innovation Diffusion Theory in the green technology domain).

On the other hand, the industrial structure upgrading effect manifests as the structural reshaping of the traditional industrial system by the digital economy, with its core lying in the deep penetration of digital technologies and cross-industrial convergence and innovation. Leveraging the robust data acquisition, analysis, and decision-making capabilities of advanced technologies like cloud computing and artificial intelligence, the digital economy can precisely identify “technology gaps” and optimization potential points (e.g., production process bottlenecks, energy waste segments) within the industrial chain during the integration of traditional production factors (capital, labor, energy) with new elements like data and information. This identification lays the informational foundation for the efficient cross-industry and cross-enterprise flow and sharing of factor resources (e.g., supply chain coordination and shared manufacturing facilitated by data platforms). Subsequently, by leveraging scale effects (reducing unit service/transaction costs) and competitive effects (incentivizing efficiency improvements and innovation) inherent in digital platforms, the digital economy drives fundamental transformations in production modes. This transformation compels the entire industrial chain to undergo digital transformation. By enhancing overall operational efficiency, optimizing supply chain management, reducing intermediate link losses, and fostering the development of high-value-added, low-energy-intensive, and digital-intensive service industries, this process ultimately achieves a relative decline in total energy consumption and an absolute improvement in resource utilization efficiency. Consequently, this leads to a significant enhancement in the carbon emission performance [21]. In summary, Research Hypothesis 2 is proposed as follows:

Hypothesis 2.

The digital economy improves carbon emission performance through the effects of technological innovation and industrial structure upgrading.

2.3. Infrastructure Threshold: The Green Paradox Tipping Point

While the digital economy contributes significantly to reducing urban carbon emissions, its inherent challenges in energy conservation and emission reduction have garnered considerable attention [1]. The construction of digital infrastructure—particularly data centers, network equipment, and communication base stations—entails substantial electricity consumption and imposes significant carbon emission pressures.

Carbon footprints arise from three primary phases: (1) equipment manufacturing, involving energy-intensive production processes and complex supply chains; (2) data center operations, dominated by server computation and cooling system energy consumption; and (3) network transmission, where fiber optics, routers, and switches require continuous power, largely sourced from fossil fuels.

Consequently, activities spanning the infrastructure lifecycle, from manufacturing to operation, increase the carbon footprint. This indicates that expansion is not necessarily beneficial. The resulting lock-in effect of high electricity consumption and carbon emissions may impede urban low-carbon transitions, exemplifying the “Green Paradox” phenomenon within the digital economy.

Based on this analysis, we propose Hypothesis 3:

Hypothesis 3.

Digital infrastructure development exhibits a threshold effect; beyond this threshold, its marginal contribution to the reduction of urban carbon emissions diminishes.

2.4. Dual-Target Modulation: Policy Synergy in Chinese Governance

Effective low-carbon governance requires leveraging the fundamental role of market resource allocation while avoiding over-reliance on governmental promises [22]. For governments, setting targets for both economic growth and environmental protection is a crucial policy tool for achieving efficient and low-carbon governance. Appropriately designed dual-target constraints are vital for realizing China’s “dual-carbon” goals [23].

According to promotion tournament theory, local governments are highly motivated to achieve economic growth targets. When the central government assigns provincial targets, local officials often respond by “layering” weight onto these targets, exceeding them and fostering competition centered on “growth pursuit” and “investment attraction.” While this model drove rapid economic development during specific historical periods, it inevitably led to a zero-sum game dynamic. This manifests as an imbalanced investment structure, with excessive emphasis on infrastructure at the expense of service sector development, thereby contributing to high carbon emissions. Furthermore, prolonged reliance on traditional factor inputs and growth patterns has created a path dependence characterized by “high energy consumption, high emissions, and high pollution,” which further induces elevated carbon emissions.

Consequently, excessively stringent economic growth targets can weaken the effects of technological innovation and industrial structure upgrading driven by the digital economy, ultimately inhibiting low-carbon transformation and development [24].

Conversely, comprehensive environmental policies—encompassing regulations, supervision mechanisms, and strengthened governance demonstrably improve ecological quality. However, persistent challenges related to energy consumption and pollution remain significant. Recognizing this, China’s 14th Five-Year Plan explicitly imposed binding environmental targets on energy consumption and carbon emissions. Generally, environmental target constraints compel firms to adjust pollutant emissions, adopt greener production methods, and increase their investment in green technology innovation and R&D. Therefore, strict environmental target constraints, particularly direct mandates, are likely to strengthen the green technology bias inherent in the digital economy, facilitating low-carbon transformation and development [25,26].

Based on this analysis, we propose Hypothesis 4:

Hypothesis 4.

Flexible economic growth targets enhance, while rigid targets weaken, the carbon emission reduction efficiency of the digital economy. Environmental target constraints, especially direct constraints, exert a positive moderating effect.

3. Research Design

3.1. Model Building

3.1.1. Static Panel Model

This study employs the STIRPAT model as its analytical framework. This model comprehensively accounts for the influence of key socio-economic factors on environmental outcomes and is well established for investigating the determinants of carbon emission performance [27].

$$CE{P}_{it}={\alpha }_0+{\alpha }_{1}CE{P}_{it-1}+{\alpha }_{2}D{E}_{it}+{\alpha }_3{C}_{it}+{\mu }_i+{\lambda }_t+{\delta }_{it}$$

(1)

Among them, $CE{P}_{it}$ represents the carbon emission performance of city $i$ in the $t$ year; $D{E}_{it}$ is the level of digital economic development; $C_{it}$ represents the group of control variables that affect carbon emission performance, including economic development level, urbanization level, population size, fixed asset investment, foreign direct investment, and other variables. ${\mu }_i$, ${\lambda }_t$, ${\delta }_{it}$ represent the regional effect, the temporal effect, and the stochastic perturbation term, respectively.

3.1.2. Mechanism Test Model

In order to test the existence and contribution of the two major mechanisms of technological innovation effect and industrial structure upgrading effect, this paper draws on the method of Cutler & Liters-Muney (2010) [28] to add mechanism variables to Equation (1), which include the measurement of technological innovation effect per 10,000 green patent applications ($ETI$), the ratio of added value of secondary and tertiary industries to measure the effect of industrial structure upgrading ($ISU$) [29], and the other variables remain unchanged. The test steps are as follows: (1) directly test the impact of the digital economy on the effect of technological innovation and the effect of industrial structure upgrading, so as to verify the existence of the two mechanisms; (2) On the basis of the regression Equation (3), the explanatory strength of the two mechanisms is further measured, and the calculation procedure is to obtain the coefficients $\widehat{\alpha }$ and $\widehat{\phi }$ of the level of digital economy development from the regression Equations (1) and (3), and calculate $1-{\displaystyle \raisebox{1ex}{$\widehat{\alpha }$}\!\left/ \!\raisebox{-1ex}{$\widehat{\phi }$}\right.}$, that is, to obtain the contribution degree of the mechanism variables. The relevant regression equation is as follows:

$${\gamma }_{it}={\beta }_0+{\beta }_{1}D{E}_{it}+{\mu }_i+{\lambda }_t+{\delta }_{it}$$

(2)

$$CE{P}_{it}={\phi }_0+{\phi }_{1}CE{P}_{it-1}+{\phi }_{2}D{E}_{it}+{\phi }_3{\gamma }_{it}+{\phi }_4{C}_{it}+{\mu }_i+{\lambda }_t+{\delta }_{it}$$

(3)

3.1.3. Threshold Effect Model

Considering the carbon emission pressure caused by digital infrastructure construction, this paper adopts the classical panel threshold model to verify that the impact of the digital economy on urban carbon emission performance shows a nonlinear feature of diminishing marginal effect, and the equation is as follows:

$$CE{P}_{it}={\omega }_0+{\omega }_{1}CE{P}_{it-1}+{\omega }_{2}D{E}_{it}\times I(DI{F}_{it}\le \vartheta )+{\omega }_{3}D{E}_{it}\times I(DI{F}_{it}>\vartheta )+{\omega }_4{C}_{it}+{\mu }_i+{\lambda }_t+{\delta }_{it}$$

(4)

Among them, $DI{F}_{it}$ is the level of digital infrastructure, that is, the threshold variable, $\vartheta$ is the threshold value, I(·) is the indicative function, if the condition in parentheses is satisfied, the value is assigned 1, and the other variables are consistent with (1).

3.1.4. Moderating Effect Model

In order to explore the differentiation mechanism of the digital economy on carbon emission performance under the constraint of dual objectives, this paper introduces the interaction term between the development level of the digital economy and the constraints of the dual objectives on the basis of (1), and constructs the following moderating effect model:

$$CE{P}_{it}={\gamma }_0+{\gamma }_{1}CE{P}_{it-1}+{\gamma }_{2}D{E}_{it}+{\gamma }_{3}EG{C}_{it}+{\gamma }_{4}D{E}_{it}\times EG{C}_{it}+{\gamma }_5{C}_{it}+{\mu }_i+{\lambda }_t+{\delta }_{it}$$

(5)

where $EG{C}_{it}$ represents the dual target constraints, including the economic growth target constraint and the environmental target constraint, and the other variables are consistent with (1).

3.2. Variable Selection

3.2.1. Dependent Variable

Carbon Emission Performance (CEP) serves as the dependent variable in this study and is operationally defined as urban carbon emission intensity (CEI). Lower CEI values indicate a superior CEP. To ensure a positive interpretation of the digital economy’s impact on CEP (where higher values denote better performance), we apply a logarithmic transformation to CEI after adding a constant of 1 (i.e., ln (1 + CEI)). CEI is calculated as the ratio of total urban carbon emissions to GDP.

We estimated city-level carbon emissions by downscaling provincial emission data using calibrated NTL data (DMSP-OLS and NPP-VIIRS). To address spatial granularity limitations, we applied a cross-sensor calibration protocol [30,31,32,33] to harmonize NTL intensity across datasets. Specifically, DMSP-OLS data were corrected for saturation and blooming effects via quadratic regression with intercalibrated stable lights [32]. NPP-VIIRS data were filtered to remove background noise using a threshold of 0.5 nW/cm²/sr [33]. A power function model (CE = α × ${NTL}^{\beta }$) was employed for inversion, where parameters (α, β) were calibrated against provincial emission inventories from China Emission Accounts and Datasets (CEADs). The mean R² of provincial fits was 0.87 (range: 0.82–0.92), confirming NTL as a robust proxy for economic activity-driven emissions.”

To validate our NTL-based estimates, we conducted three robustness tests:

Firstly, we compared the data with official city-level data. For the 15 cities included in the national GHG reporting pilot (e.g., Beijing and Guangzhou), we compared our CEI values with official reports (2019–2021). Secondly, the Pearson correlation coefficient was 0.78 (p < 0.01), with a mean absolute percentage error (MAPE) of 12.3%. Sensitivity to NTL thresholds. Thirdly, emissions were re-estimated using alternative noise thresholds (0.3–0.7 nW/cm²/sr). Results remained stable (CEI deviations < 5%). Province-to-city disaggregation consistency. We compared summed city emissions against provincial totals from CEADs. Discrepancies averaged 8.7%, within acceptable margins for downscaling studies [33].

While NTL data provide the best available proxy for spatially explicit city-level emissions in China, two constraints are noteworthy.

On the one hand, emissions from non-economic activities (e.g., forest fires) may be underrepresented, and downscaling accuracy depends on provincial inventory quality. On the other hand, we mitigate these by focusing on energy-related CO₂ (90% of China’s emissions) and using CEADs—the most widely cited inventory in peer-reviewed studies.”

3.2.2. Explanatory Variables

The explained variable of this paper is the development level of the digital economy ($DE$). To assess the development level of the digital economy, this paper refers to the indicator systems constructed by Zhao et al. (2020) and Zhong et al. (2021) [34,35,36]. Five indicators are selected: the number of mobile phone subscribers per 100 people, per capita telecommunications consumption, the proportion of fixed asset investment in the ICT industry, the total number of unicorn enterprises, and digital finance. Data is sourced from the China City Statistical Yearbook and authoritative third-party institutions. First, the min-max normalization method is applied to eliminate the influence of different measurement units. Subsequently, the Pearson correlation coefficients between indicators are calculated to confirm the absence of significant multicollinearity risk (max |r| = 0.63 < 0.7). Then, the coefficient of variation method is used to determine the weights of the indicators. Finally, a digital economy development index is synthesized using linear weighting. Additionally, four indicators are selected to characterize digital infrastructure variables, including the total number of CN domain names, the mobile phone penetration rate, the number of IPv4 addresses, and the Internet penetration rate. The entropy method is used to measure the level of digital infrastructure ($DIF$).

3.2.3. Control Variables

By combining the literature on the influencing factors of carbon emission performance, it is found that carbon emission performance is mainly affected by economic development, population size, and technological progress; therefore, this paper selects relevant control variables (as shown in

Table 1. Description of control variables.

The Name of the Variable	Variable Symbol	Measurement
Level of economic development	RGDP	GDP per capita
Level of urbanization	UR	Urbanization rate
Population size	POP	Urban resident population
Investment in fixed assets	IFA	Investment in fixed assets per capita
outward foreign direct investment	FDI	Proportion of outward direct investment to GDP

).

3.3. Data Source

The study period spans 2011 to 2023. This timeframe was selected to capture the rapid development phase of the digital economy while ensuring adequate data is available. Cities with substantial missing data were excluded, resulting in a final sample of 278 Chinese prefecture-level and higher cities. The data primarily come from the “China City Statistical Yearbook”, “China Energy Statistical Yearbook”, various local statistical annual reports, the China carbon emission database, the green patent database, and the EPS database. To address missing data points (which constituted approximately X% of the total observations across the core variables used in the final model), we employed two methods based on the data structure and missing patterns: (1) adjacent-year averaging (proximity mean): for time-series data missing a single year, the value was imputed as the average of the preceding and following year’s values; (2) linear interpolation: for consecutive missing years within a city’s time series, values were estimated using linear interpolation between the nearest available data points before and after the gap.

All variables were logarithmically transformed to mitigate skewness. Economic and environmental growth targets were extracted from municipal government work reports. Nighttime light (NTL) radiance values were obtained from the Global Nighttime Lights Database, with China-specific city-level data calibrated prior to use.

Table 2. Descriptive statistics of the variables.

The name of the Variable	Variable Symbol	Number of Observations	Mean	Standard Deviation	Min	Max
Carbon performance	CEP	3614	0.2611	0.1302	0.0731	1.6591
The level of development of the digital economy	DE	3614	0.1344	0.0688	0.0277	0.6533
Technological innovation effect	ETI	3614	0.6652	1.6173	0.0033	7.8815
The effect of industrial structure upgrading	ISU	3614	1.7732	1.3081	0.7266	3.8891
Digital infrastructure level	DIF	3614	0.1266	0.1178	0	1
Level of economic development	RGDP	3614	4.0066	2.6653	3.3277	9.4499
Level of urbanization	UR	3614	0.6712	0.4491	0.1022	1.8823
Population size	POP	3614	4.8854	1.0977	2.2981	7.6928
Investment in fixed assets	IFA	3614	3.0166	1.7633	2.3015	6.6671
outward foreign direct investment	FDI	3614	1.8933	1.6642	0.2711	5.0172

presents the descriptive statistics for all variables. To assess the robustness of our findings with respect to the imputation methods, we will conduct sensitivity analyses as part of the revision: (a) re-estimating the main models using the dataset excluding all imputed values and (b) employing multiple imputation techniques (e.g., chained equations) to generate several complete datasets and pool the results.

4. Empirical Analysis

4.1. Benchmark Regression

Table 3. Benchmark regression results.

Variables	(1) CEP	(2) CEP
DE	0.2671 ** (0.0501)	0.2381 ** (0.0109)
L.CEP	0.0319 * (0.1022)	0.0609 ** (0.0813)
RGDP		0.1501 *** (0.0066)
UR		−0.0078 * (0.0065)
POP		0.0329 * (0.1122)
IFA		−0.0166 (0.0297)
FDI		0.0031 (0.0099)
Constant terms	1.8633 *** (0.0033)	1.8861 *** (0.0011)
AR (1)	−3.0017 ***	−2.0092 ***
AR (2)	−1.0098	−0.9917
Sargan	33.1988	39.0013
N	3614	3614

Note: *, **, and *** indicate significance at the 1%, 5%, and 10% levels, respectively; values in parentheses are robust standard errors.

presents the baseline regression results that examine the impact of the digital economy on urban carbon emission performance (CEP). Column (1) reports estimates without control variables, while Column (2) includes the control variables. From the regression results of columns (1) and (2), it can be seen that compared with the regression without control variables, the regression coefficient of the digital economy decreases slightly after the control variables are added, but the overall explanatory power of the model is improved, and the regression coefficient of the digital economy on the urban carbon emission performance is always significant at the level of 5%, indicating that the digital economy is helpful in improving the urban carbon emission performance.

4.2. Robustness Test and Handling of Endogeneity

To further test the robustness of the digital economy’s impact on carbon emission performance, this paper employs five distinct approaches: substituting the core explanatory variable, incorporating time trends and control variable interaction terms, accounting for potential confounding policy influences, applying alternative spatial weight matrices, and conducting instrumental variable regression to address endogeneity concerns.

4.2.1. Replace the Explained Variable

To incorporate energy consumption and carbon reduction into the traditional Total Factor Productivity (TFP) framework, we measure carbon emission performance using Total Factor Carbon Productivity (TFCP). This metric employs labor, capital, and energy as inputs, with GDP as the desirable output and carbon emissions as the undesirable ones. TFCP calculations are performed using the Slack-Based Measure Data Envelopment Analysis (SBM-DEA) model under a super-efficiency framework. As presented in

Table 4. Robustness tests and results of endogeneity treatment in regression.

Variables	(1) Replace the Explained Variable	(2) Control Variables * Time Trend Terms	(3) Exclude Other Policy Implications		(4) Instrumental Variable Regression
			Broadband China	Low-Carbon Pilot	Stage 1	Stage 2
DE	0.2377 ** (0.0688)	0.2199 ** (0.0490)	0.2099 ** (0.0265)	0.2206 ** (0.0397)	/	0.2697 *** (0.0031)
L.CEP	0.0401 ** (0.0300)	0.0619 ** (0.0091)	0.0866 *** (0.0001)	0.0288 * (0.0988)	/	0.0718 ** (0.0920)
T* Control variables	/	YES	/	/	/	/
Control variables	YES	YES	YES	YES	YES	YES
Tool variables	/	/	/	/	0.0082 *** (0.0088)	/
Constant terms	1.0622 *** (0.0038)	1.1379 *** (0.0066)	1.1281 *** (0.0095)	1.1388 *** (0.0061)	0.0382 ** (0.0376)	0.2231 ** (0.0911)
AR (1)	−3.1988 ***	−2.1066 ***	−3.0017 ***	−2.0092 ***	0.7662	0.6631
AR (2)	−1.2010	−0.8688	−1.5301	−0.7633	/	/
Sargan	31.0089	40.0092	35.1148	38.1922	44.1988	46.0019
N	3614	3614	3614	3614	3614	3614

Note: *, **, and *** indicate significance at the 1%, 5%, and 10% levels, respectively; values in parentheses are robust standard errors.

, row (1), substituting the core explanatory variable demonstrates that the digital economy has a statistically significant positive impact on carbon emission performance. This result further supports the baseline regression findings.

4.2.2. Add Control Variables to Interact with Time Trends

Incorporating interaction terms between temporal trends and control variables in the empirical model mitigates estimation bias by accounting for time-varying heterogeneity in the explanatory factors [14]. As evidenced in Column (2) of Table 4, the digital economy’s statistically significant positive impact on carbon emission performance remains robust in both direction and magnitude. This consistency further validates the stability of benchmark regression results.

4.2.3. Exclude the Influence of Other Policies

This analysis prioritizes controlling for the ‘Broadband China’ and ‘Low-Carbon Pilot Cities’ policies, given their significant concurrent influence on both digital economy development and urban low-carbon transformation. The ‘Broadband China’ strategy was implemented in phases: pilot cities designated between 2014 and 2016 were assigned a value of 1, with non-pilot cities coded as 0. Similarly, the ‘Low-Carbon Pilot Cities’ initiative commenced in 2010 (selecting five provinces and eight cities), followed by subsequent pilot batches in 2012 and 2017; all designated pilot areas received a value of 1, others 0. As reported in Column (3) of Table 4, excluding these potential confounding policies reveals that the estimated impact of the digital economy on carbon emission performance remains statistically indistinguishable and qualitatively unchanged. This confirms the robustness of our core findings.

4.2.4. Instrumental Variable Regression

The development of the digital economy and carbon emission performance are both potentially influenced by common factors such as the urban institutional environment, government governance capacity, and technological innovation. This shared causality may introduce endogeneity issues into their relationships. To effectively mitigate potential endogenous biases, this study adopts instrumental variable (IV) construction methods from the existing literature. Specifically, we introduce terrain relief as a natural geographical factor and construct its interaction term with year dummy variables as the instrumental variable. We then conduct a two-stage least squares (2SLS) regression analysis. The rationale for selecting this instrumental variable is threefold:

First, it satisfies the exogeneity criteria. Terrain relief is primarily determined by natural geological processes and the evolution of landforms. As an inherent geographical characteristic, it is theoretically exogenous to modern socio-economic activities, including carbon-emission performance. Consequently, terrain relief is unlikely to directly influence carbon emission performance through channels other than its effect on digital economy development, thus satisfying the core exogeneity requirement for a valid instrument.

Second, it meets the relevance criteria. Terrain relief intuitively reflects the complexity of a city’s topography. Greater relief corresponds to more complex terrain, which impedes the deployment of digital infrastructure (e.g., fiber-optic networks and base stations) by increasing construction and maintenance costs. This directly constrains the coverage and development level of the digital economy, establishing a significant correlation between terrain relief (via infrastructure difficulty) and the core explanatory variable, that is, digital economy development.

Finally, the rationale for the exclusion restriction is valid. The key design feature of this instrument is that terrain relief (and its interaction with time) is theorized to influence carbon emission performance primarily, if not exclusively, through its impact on the digital economy. Given its natural attributes, terrain relief lacks a plausible direct theoretical link to other modern economic, social, or policy factors that affect carbon emission performance (e.g., industrial structure, environmental regulation intensity, and consumption patterns). This strengthens the justification for the exclusion of the restriction assumption.

The two-stage least squares (2SLS) results are presented in Column (4) of Table 4. The first stage confirms a statistically significant positive relationship (p < 0.01) between the IV and digital economy development, supporting the relevance of the condition. In the second stage, the estimated coefficient for the digital economy remains significantly positive at the 1% level, with increased magnitude and statistical precision compared to the baseline estimates. This confirms the robust positive impact of the digital economy on carbon emission performance after correcting for endogeneity. In terms of statistical testing, the F statistic (146.9077) exceeded the critical value, and the LM statistic was significant at the 1% level. It passed the weak instrument validity test and rejected the null hypothesis of under-identification, supporting the validity of the instrumental variables. Consequently, Hypothesis 1 is empirically supported.

4.3. Heterogeneity Analysis

4.3.1. Heterogeneity of Urban Location

The Qinling-Huaihe River line serves as China’s fundamental north-south biogeographical demarcation, notably determining winter central heating provisions. Significant differences in topography, climate, economic development, and ecological conditions exist across this boundary, prompting an investigation into the potential regional heterogeneity in the digital economy’s carbon mitigation efficacy. Regression results for geographical heterogeneity (

Table 5. Heterogeneity test regression results I.

Variables	(1) Heterogeneity of Urban Location		(2) Heterogeneity of Urban Resource Endowment
	South of the Qinling-Huai River Line	North of the Qinling-Huai River line	Resource-Oriented Cities	Non-Resource-Based Cities
DE	0.1326 * (0.1681)	0.1796 ** (0.0688)	0.1301 ** (0.0968)	0.0791 ** (0.0392)
L.CEP	0.0331 ** (0.0110)	0.0700 ** (0.0082)	0.0701 *** (0.0002)	0.0310 * (0.0630)
Control variables	YES	YES	YES	YES
AR (1)	−3.3367 ***	−2.6631 ***	−3.1903 ***	−2.1179 ***
AR (2)	−1.0789	−0.7633	−1.5700	−0.7198
Sargan	36.1138	43.1908	36.1980	37.1003
N	1404	2210	1638	1976

Note: *, **, and *** indicate significance at the 1%, 5%, and 10% levels, respectively; values in parentheses are robust standard errors.

, Column 1) demonstrate that the digital economy’s impact on carbon emission performance is both statistically stronger (p < 0.01) and quantitatively larger north of the demarcation than in the south. This indicates enhanced low-carbon governance effectiveness in northern cities.

We attribute this divergence to structural economic differences: Northern cities exhibit greater reliance on emission-intensive secondary industries within a comparatively lower-end industrial structure. This industrial composition, coupled with severe environmental challenges, generates a stronger regional demand for pollution control and emission reduction technologies, which the digital economy appears better positioned to address in these localities.

4.3.2. Heterogeneity of Urban Resource Endowment

Using the official catalog of resource-based cities, we classified the sample cities into resource-based and non-resource-based categories. Regression results (Table 5, Column 2) reveal a more pronounced low-carbon governance effect of the digital economy in resource-based cities than in their non-resource-based counterparts. This suggests the digital economy effectively mitigates economic disparities between these city types, positioning it as a critical pathway for addressing the “resource curse.” These findings are consistent with those of Zou et al. (2024) [37].

4.3.3. Heterogeneity of Urban Characteristics

Effective maximization of the digital economy’s low-carbon governance effects necessitates context-specific policy implementation. Given that the development of the digital economy and environmental governance depend critically on fiscal investment and new infrastructure scale, we examine heterogeneity across city characteristics.

Using Peking University’s Digital Inclusive Finance Index to measure financial inclusion levels, we stratify cities into high- and low-inclusive finance cohorts. Similarly, cities are classified into high- and low-fiscal expenditure groups based on the ratio of fiscal expenditure to GDP (data: China Urban Statistical Yearbook).

Regression results in

Table 6. Heterogeneity test regression results II.

Variables	(1) The Level of Digital Inclusive Finance		(2) The Level of Fiscal Spending
	High	Low	High	Low
DE	0.1933 *** (0.0391)	0.0866 * (0.3785)	0.1022 * (0.3251)	0.6377 *** (0.1131)
L.CEP	0.0566 ** (0.0210)	0.0661 ** (0.0072)	0.0511 *** (0.0004)	0.0400 * (0.0830)
Control variables	YES	YES	YES	YES
AR (1)	−3.0067 ***	−2.9906 ***	−3.2099 ***	−2.6678 ***
AR (2)	−1.0087	−0.8890	−1.4677	−0.8066
Sargan	36.9908	41.1779	39.1078	37.5609
N	1404	2210	1638	1976

Note: *, **, and *** indicate significance at the 1%, 5%, and 10% levels, respectively; values in parentheses are robust standard errors.

reveal significant heterogeneity: (1) Cities with advanced digital inclusive finance exhibit stronger digital economy-driven carbon mitigation. (2) Conversely, the digital economy’s low-carbon governance effect is more pronounced in cities with lower fiscal expenditures than in cities with higher expenditures.

We posit that high-expenditure cities deploy diversified pollution-control measures, diminishing the marginal impact of digital solutions. Thus, digital economy development may generate disproportionate benefits in cities with low fiscal expenditure, where emission reduction strategies are less diversified.

5. Extended Discussion

5.1. Mediation Analysis

Empirical findings confirm the positive effect of the digital economy on urban carbon emission performance. We identify two primary transmission channels: technological innovation and industrial structure upgrades. Building on established mediation analysis frameworks, this study quantifies both the existence and relative contribution of these mechanisms.

Table 7. Regression results for testing the existence and contribution of the mechanism of action.

Explanatory Variables	(1) ETI	(2) ISU	(3) Baseline Regression	(4) Mechanism 1: Technological Innovation Effect	(5) Mechanism 2: The Effect of Upgrading the Industrial Structure
				ETI	ISU
DE	0.0463 *** (0.0377)	0.0021 *** (0.0017)	0.2381 ** (0.0109)	0.3560 *** (0.0021)	0.2931 ** (0.1826)
L.CEP	/	/	0.0609 ** (0.0813)	0.0122 ** (0.0013)	0.0301 *** (0.0001)
ETI	/	/	/	0.04833 *** (0.0391)	/
ISU	/	/	/	/	0.0026 *** (0.0021)
$1-{\displaystyle \raisebox{1ex}{$\widehat{\alpha }$}\!\left/ \!\raisebox{-1ex}{$\widehat{\phi }$}\right.}$	/	/	/	38.1120%	19.3362%
Control variables	YES	YES	YES	YES	YES
AR (1)	/	/	−2.0092 ***	−4.0013 ***	−3.9188 ***
AR (2)	/	/	−0.9917	−0.7621	−0.8097
Sargan	/	/	39.0013	28.1322	27.0065
N	3614	3614	3614	3614	3614

Note: ** and *** indicate significance at the 5%, and 10% levels, respectively; values in parentheses are robust standard errors.

shows the results of the mechanism of the impact of the digital economy on carbon emission performance, the existence of the test, and the contribution tests. Columns (1) and (2) show the regression results of the test for the existence of the mechanism. The results show that the digital economy has a significant positive impact on the two major mechanism variables, and the impact on technological innovation is greater than the effect of industrial structure upgrading, indicating that the digital economy has a positive impact on the mechanism. Columns (3), (4), and (5) present the regression results of the measurement mechanism, which show that the contribution of the two major mechanisms to the impact of the digital economy on carbon emission performance is more than 50%, among which the contribution of the technological innovation effect is the largest, reaching 33.1126%, followed by the industrial structure upgrading effect of 18.7662%. These results provide empirical support for Hypothesis 2 regarding transmission pathways.

5.2. Threshold Effect Analysis

Based on our investigation into the nonlinear relationship between the digital economy and urban carbon emission performance using the Hansen (1999) panel threshold model specified in Equation (4), we first conducted significance testing for threshold effects. Digital infrastructure, a critical enabler of digital technology development, incurs carbon emission costs during its construction and operation phases. To analyze this relationship, the estimation procedure rigorously followed established methods.

Table 8. Threshold effect test results.

Model	Threshold Value	F Value	p Value	Critical Value
				1%	5%	10%
Single threshold	0.0577	36.1077 ***	0.0031	30.0026	23.0918	19.7251
Double threshold	0.5688	7.0988	0.8266	31.2988	22.0881	16.3092

Note: *** indicate significance at the 10% levels; values in parentheses are robust standard errors.

shows the test results of the threshold effect, we performed 300 Bootstrap iterations to identify the threshold value and test its significance, resulting in an estimated single threshold value for digital infrastructure of 0.0577. The threshold effect test results confirm a statistically significant single threshold effect (p < 0.01) while indicating that no statistically significant double threshold exists.

To visualize the nonlinear effect, Figure 1 plots the marginal impact of the digital economy on carbon emission performance across threshold regimes.

Table 9. Threshold effect regression results.

Variables	CEP
$DE\times I(DIF\le \theta )$	1.1673 *** (0.1683)
$DE\times I(DIF>\theta )$	0.2865 *** (0.1016)
L.CEP	0.0700 ** (0.0019)
Control variables	Control
AR (1)	−4.0018 ***
AR (2)	−0.6519
Sargan	36.3788
N	3614

Note: ** and *** indicate significance at the 5%, and 10% levels, respectively; values in parentheses are robust standard errors.

presents the regression results of the panel threshold regression model. The impact of the digital economy on urban carbon emission performance is not an absolute linear relationship but exhibits nonlinear characteristics. Specifically, as digital infrastructure is constructed and used on a large scale, and after crossing the threshold, the regression coefficient of the digital economy decreases from 1.1673 to 0.2865. This indicates that due to the influence of digital infrastructure, the effect of the digital economy on urban carbon emission performance demonstrates diminishing marginal effects, thereby validating hypothesis 3. This may be because the “low-hanging fruits” that are easy to achieve in the early stage of digital infrastructure construction have been picked; the energy consumption problem of digital infrastructure itself is prominent (if the power is not greened); deep emission reduction requires more complex and costly system integration and cross-domain collaboration; the lag of management change and organizational adaptation restricts the development of technological potential; advancing to deep water areas and weak links faces higher transformation costs and lower return on benefits; and the expansion of the scope of popularization has led to the dilution of unit input benefits. This shows that in order to continue to improve the contribution of the digital economy to urban carbon emission reduction, we cannot only rely on the scale expansion of digital infrastructure and the simple application of the digital economy, but need to pay more attention to green energy supply (solving the emission of infrastructure itself), deepening technology integration and system integration, promoting supporting management and organizational changes, innovating policy incentive mechanisms, and promoting the deep and green integration of the digital economy with traditional industries, especially high-carbon industries. The results also indicate a ‘de-greening’ trend in the country’s digital infrastructure. Therefore, in the future, focusing on energy efficiency applications and upgrading digital infrastructure as a key goal for energy conservation and emission reduction is crucial to drive green development from the source.

5.3. Analysis of the Moderating Effect of Dual-Target Constraints

Based on the above theoretical analysis, the effects of excessively high economic growth targets and strict environmental targets on the impact of the digital economy on carbon emission performance are completely different; therefore, this paper explores the moderating effect of dual objective constraints from an empirical perspective. Based on the government work reports of China’s prefecture-level cities, we analyze the economic growth target constraints (EGC) and construct interaction terms related to the digital economy by measuring the economic growth target values published in previous reports. In addition, we also distinguish in detail between the hard constraints and soft constraints of economic growth targets [38], that is, when the formulation of economic growth targets is modified by words such as “strive”, “above” and “ensure”, this paper defines it as hard constraints on economic growth targets (HEGC), and when it is described in intervals such as “up and down” and “left and right”, it is defined as soft constraints on economic growth targets (SEGC). In terms of measuring the environmental target constraints, this paper analyzes whether the energy consumption target is explicitly mentioned as the standard for measuring the environmental target constraint (EOC) in previous prefecture-level municipal government work reports and introduces dummy variables to evaluate the interaction effect of the digital economy [39].On this basis, environmental target constraints are subdivided into direct and indirect environmental target constraints [40,41]. In addition, if the assessment results are announced in the government work report of the following year, it is determined that the direct environmental target constraints (DEGC) have been implemented, and the indirect environmental target constraints (IEOC) have been implemented in other cases. The data related to economic growth target constraints and environmental target constraints are derived from the original government work reports of prefecture-level cities over the years, and are obtained by manual collation.

Column (1) of

Table 10. Double-objective constrained regression results.

Variables	(1) Adjustment of Constraints on Economic Growth Targets			(2) Regulation of Environmental Target Constraints
DE	3.1063 ** (0.0762)	2.8066 ** (0.1177)	2.6093 ** (0.1133)	0.8792 *** (0.1502)	0.5166 *** (0.2033)	0.6122 *** (0.0375)
L.CEP	0.0233 * (0.0922)	0.0366 ** (0.0701)	0.0530 * (0.0991)	0.0522 ** (0.0018)	0.0510 ** (0.0016)	0.0600 ** (0.0011)
DE*EGC	−0.3177 *** (0.0131)	/	/	/	/	/
DE*HEGC	/	−0.3908 *** (0.0081)	/	/	/	/
DE*SEGC	/	/	0.2261 *** (0.0139)	/	/	/
DE*EOC	/	/	/	1.0611 *** (0.0722)	/	/
DE*DEGC	/	/	/	/	0.4336 *** (0.0579)	/
DE*IEOC	/	/	/	/	/	0.2013 *** (0.0100)
Control variables	YES	YES	YES	YES	YES	YES
AR (1)	−2.1166 ***	−2.6609 ***	−3.9060 ***	−2.9907 ***	−3.6651 ***	−2.0010 ***
AR (2)	−1.9877	−0.7633	−1.9908	−0.7390	−1.0098	−0.6099
Sargan	34.0366	40.2688	36.0098	36.0911	37.3466	33.1088
N	3614	3614	3614	3614	3614	3614

Note: *, **, and *** indicate significance at the 1%, 5%, and 10% levels, respectively; values in parentheses are robust standard errors.

reports the regression results for the economic growth target constraints. The interaction term between economic growth target constraints (EGC) and the digital economy is statistically significant and negative (p < 0.05), indicating that EGC exerts a negative moderating effect on the relationship between the digital economy and carbon emission performance. This outcome arises because subnational governments often prioritize short-term growth initiatives over long-term technological investments [42]. Such prioritization suppresses corporate innovation incentives, curtails urban green technology development, impedes industrial digital transformation, and ultimately reduces economic efficiency—collectively undermining low-carbon transitions. This suggests that an excessive pursuit of high economic growth targets weakens the digital economy’s effectiveness in low-carbon governance, and moderate economic growth target constraints are beneficial for enhancing the digital economy’s low-carbon governance efficacy.

Column (2) of Table 10 presents the regression results for environmental target constraints. The findings indicate that the interaction coefficient between environmental target constraints and the digital economy is positive at the 1% significance level. This suggests that environmental target constraints play a positive moderating role in enhancing the impact of the digital economy on carbon emission performance. The government’s implementation of reasonable and effective environmental governance measures may create favorable conditions for green development of the urban economy. Such actions compel high-polluting enterprises to adjust their production models and leverage digital technologies to transition their industries towards greener and smarter practices, thus improving the city’s carbon emission performance. Regarding the intensity of environmental target constraints, the moderating effect of direct constraints is more pronounced than that of indirect constraints, indicating that stringent environmental target constraints are more beneficial for enhancing the low-carbon governance performance of the digital economy. Therefore, Hypothesis 4 has been fully validated.

6. Discussion

Through rigorous theoretical model construction and multi-dimensional empirical analysis, this study reveals the significant promotion effect of the digital economy on urban carbon emission performance, as well as its complex mechanism and boundary conditions. The purpose of this discussion is to interpret the theoretical and practical implications of the research findings, position them in the context of the existing literature, and explore their broad policy implications and future research directions.

Our empirical results clearly confirm that the development of the digital economy can significantly enhance the carbon emission performance of cities, providing strong micro evidence for ‘Digital China’ to empower the construction of ‘Beautiful China’. This supports the general expectation that digital technology promotes low-carbon transformation by optimizing resource allocation, improving efficiency, and enabling green innovation. However, the more significant contribution of this study is revealing the nonlinear threshold effect of this facilitative relationship, which shows a diminishing marginal effect with the level of digital infrastructure (threshold value of 0.0577) as the boundary. This differs from the linear relationships found in some studies and transcends the assumption of a simple positive effect. This profoundly illustrates that digital infrastructure is a fundamental prerequisite for the digital economy to exert low-carbon effects, but more is not always better. Once digital infrastructure surpasses a certain threshold, its subsequent expansion may be accompanied by a deepening ‘digital divide’, a resource lock-in effect (such as a surge in energy consumption in data centers), or diminishing marginal returns, which may weaken its role in further enhancing overall carbon emission performance. This finding enriches our understanding of the ‘double-edged sword’ characteristic of the digital economy, emphasizing that the quality, efficiency, and inclusiveness of digital economy development are more critical than simple scale expansion, and resonates with discussions on emerging issues such as ‘green computing power’ and ‘digital inclusion’.

This study rigorously tests mechanisms and quantitatively identifies that the effects of technological innovation (33.11%) and industrial structure upgrading (18.77%) are the two main channels through which the digital economy enhances carbon emission performance. This strongly validates the core hypothesis of the theoretical model, which states that the digital economy primarily achieves carbon emission reduction and efficiency by driving technological advancement (such as intelligent transformation, clean technology innovation, and optimized management of digital platforms) and fostering a transition of the economic structure towards low energy consumption, high added value, and service-oriented directions (such as intelligent manufacturing, digital services, and platform economy). The dominant role of the technological innovation effect underscores the essential characteristic of the digital economy as a general-purpose technology (GPT)—its core value lies in stimulating widespread and disruptive innovation. The upgrading effect of the industrial structure highlights the crucial role of the digital economy in reshaping industrial ecosystems and value chains. This finding not only aligns closely with the literature that emphasizes the core role of innovation-driven and structural transformation in green development but also precisely depicts the relative importance of both in the low-carbon pathways of the digital economy through the quantification of their contributions, providing a basis for targeted policy interventions. Notably, other potential channels (such as improved resource allocation efficiency and changes in consumption patterns) may also play a role, which can be further explored in future research.

Despite the progress made in this study, there are still limitations that point to future research. First, although the two core channels of technological innovation and industrial structure upgrading and their contributions have been identified, more in-depth case studies or enterprise-level data analysis are still required to understand the micro-mechanisms of how the digital economy specifically triggers these processes (such as specific technology types, platform models, and changes in corporate behavior). Second, although advanced methods for correcting and downscaling nighttime light data have been employed, there remains room for improvement in the accuracy of carbon emission data. In the future, multi-source data (such as satellite remote sensing CO2 concentration, ground monitoring, and corporate emission data) can be integrated to create a more refined and dynamic urban carbon emission database. Third, further exploration can be conducted on the differences in the impact of the digital economy on carbon emission performance within cities (such as central urban areas and suburbs), across enterprises of varying sizes, and in different industries. Fourth, this study primarily relies on static and threshold models. The impact of the digital economy may have characteristics of time lag and dynamic evolution. Future work could construct dynamic panel models or apply long-term time series analysis to examine long-term effects and path dependence. Fifth, the findings from Chinese cities should be placed in a global context, comparing them with other countries/cities at different stages of development and with varying institutional environments, in order to extract more universal principles. Sixth, future research could distinguish the differentiated impacts and synergistic effects of various digital technologies (e.g., 5G, AI, IoT, blockchain) on carbon emission performance and explore the potential changes in employment structure, skills demand, and the impact on different social groups during the low-carbon transformation driven by the digital economy (issues of a just transition). Sixth, the model assumes that the impact of the digital economy on carbon emission performance is mainly transmitted through two linear pathways: the effects of technological innovation and the upgrading of industrial structure, while in reality, there may be uncaptured nonlinear threshold effects or bidirectional causal feedback. At the same time, the model implies a synchronization between the penetration of digital technology and the low-carbon transformation of industries, but the actual pace of transformation may be constrained by exogenous factors such as institutional barriers and differences in regional digital infrastructure, leading to estimation bias. At the level of variable measurement, although multi-dimensional indicators are used to construct the digital economy index, there are still issues of incomplete proxy variables in the core explanatory variables (such as the degree of marketization of data elements). In addition, the assessment of carbon emission performance relies on macro-statistical data, making it difficult to completely avoid measurement errors. Although the model mitigates endogeneity through the use of instrumental variables, potential interference from unobserved variables (such as local green innovation policy preferences) must be approached with caution. Future research can further verify the synergistic mechanism and spatiotemporal heterogeneity of the dual pathways of technology and structure by introducing non-parametric estimation methods, constructing dynamic multi-sectoral CGE models, or utilizing big data from micro-enterprises.

In summary, this study provides a solid theoretical and empirical foundation for understanding the complex mechanisms through which the digital economy enables the ‘dual carbon’ goals. This research confirms the significant positive impact of the digital economy on urban carbon emission performance, revealing its nonlinear characteristics based on infrastructure thresholds, the core transmission pathways of technological innovation and industrial upgrading, and the significant heterogeneity and moderating effects stemming from urban types and the institutional environment (inclusive finance, fiscal capacity, flexibility of growth targets, and intensity of environmental constraints). These findings not only deepen the academic community’s understanding of the environmental impacts of the digital economy but also provide crucial scientific evidence and actionable policy levers for policymakers to synergistically promote the strategies of ‘Digital China’ and ‘Beautiful China’, accurately formulate differentiated regional low-carbon development policies, and optimize the institutional environment to maximize the green dividends of the digital economy. Future research should continue to deepen the aspects of mechanism depth, data accuracy, causal identification, dynamic evolution, and social equity, in order to more comprehensively capture the grand vision of green transformation driven by the digital economy.

7. Conclusions

Amidst global decarbonization imperatives and China’s accelerating digital transformation, this study provides robust empirical evidence of the complex relationship between the digital economy (DE) and urban carbon emission performance (CEP). Our analysis reveals several pivotal insights. Crucially, while the digital economy demonstrably enhances urban carbon emission performance, this positive impact exhibits a distinct threshold characteristic associated with the maturity of digital infrastructure. Below a critical developmental point, DE expansion yields substantial improvements in CEP; however, beyond this threshold, the marginal benefits diminish significantly, highlighting potential diminishing returns and underscoring the risk of a “green paradox” where further digitalization yields progressively smaller environmental gains.

The pathways through which DE drives decarbonization are dominated by the catalytic effect of green technological innovation, which plays a substantially more prominent role than the concurrent upgrading of the industrial structure. This finding confirms that the digital economy primarily facilitates emission reduction by drastically lowering the costs associated with clean technology research and development and inducing innovation specifically biased towards environmental sustainability.

The impact of DE on CEP is far from uniform and reveals significant heterogeneity across city contexts. The effect is markedly stronger in cities located in northern China, historically reliant on resource extraction (where DE demonstrates a particularly powerful capacity to enhance efficiency), and in fiscally constrained urban centers (where it offers a potent low-cost pathway for pollution control). This underscores the potential of the digital economy to help overcome developmental constraints like the “resource curse” and enable effective environmental management even under budgetary pressures.

Furthermore, the effectiveness of DE as a decarbonization tool is dynamically moderated by prevailing policy frameworks. Flexible economic growth targets, such as those expressed as interval values, significantly amplify DE’s capacity to reduce emissions. Conversely, rigid “guaranteed” growth targets demonstrably suppress this beneficial effect. Environmental policy targets also play a critical moderating role, with explicit and binding mandates exerting a considerably stronger influence on channeling DE towards emission reduction than more indirect or suggestive constraints.

Collectively, this research illuminates the nuanced interplay between digitalization and decarbonization at the urban level, offering vital theoretical insights and actionable policy pathways for strategically aligning China’s ambitious “Dual Carbon” goals with its transformative Digital China initiative.

To optimize the enhancement effects of the digital economy’s (DE) carbon emission performance (CEP), evidence-based policy must focus on strategic infrastructure investment with threshold management, prioritizing regions below critical development levels for maximum marginal gains. Policy leverage should center decisively on green innovation support—through dedicated funds, R&D incentives, and IP reform—given its dominant contribution to decarbonization pathways, while treating industrial structure upgrading as a secondary consideration. Implementation requires spatially differentiated strategies, such as deploying advanced DE solutions in high-energy-consuming regions, leveraging DE efficiency for resource-based city transitions, and empowering fiscally constrained cities with cost-effective digital governance tools. Crucially, optimizing the dual-target constraint system is imperative by replacing rigid economic growth targets with flexible interval-based objectives to avoid efficiency suppression while simultaneously strengthening environmental governance through binding quantitative emission targets and integrating real-time DE-enabled monitoring. Finally, while China’s context offers transferable principles, global application necessitates adaptable policy toolkits calibrated to local capacities and stages, alongside advocating for international standardization of digital carbon accounting to enable scalable impact assessment and carbon market linkages.