Digital deepening and carbon outcomes in China's production network using an integrated framework combining...

China's digital transformation is increasingly embodied in digital goods and services used as intermediate inputs in production. Yet existing research still provides limited evidence on whether the carbon implications of digital deepening are stable across regions and sectors, through which mechanisms they evolve, and how such heterogeneity translates into future emissions risk. To address this gap, we conceptualize digitalization as production-embedded digital inputs and examine how these inputs reshape production-based emissions across heterogeneous provincial roles and interindustry stages in China. We harmonize Asian Development Bank input-output tables with CO₂ emissions accounts to construct a national-province-sector panel for 2009-2022, and combine Tapio decoupling diagnostics, logarithmic mean Divisia index decomposition, and Monte Carlo-based STIRPAT projections for 2023-2035. Four main findings emerge. First, digital deepening and CO₂ emissions remain spatially misaligned, separating the locations where digital demand is generated from those where emissions are recorded. Second, decoupling regimes are structured rather than random. Coastal provinces with strong service-manufacturing linkages more often move toward stronger decoupling, whereas northern provinces specialized in resource extraction and heavy industry more frequently remain locked in coupling regimes. Third, the carbon consequences of digital inputs depend on shifting bundles of transmission channels. Digital inputs often intensify scale pressure, but mitigation arises when diffusion is absorbed as efficiency-enhancing capital that strengthens substitution and improves energy performance. By contrast, in upstream energy and materials systems, digital tools can raise utilization and reinforce carbon lock-in. Fourth, faster diffusion widens upper-tail emissions risk in upstream production bases, whereas metropolitan service economies stabilize earlier and display narrower uncertainty bands. Overall, the marginal carbon effect of digital inputs is not fixed in sign, but is instead a governable system outcome shaped by network position, absorption mode, and binding constraints.