Low-carbon innovation plays an essential role in carbon reduction worldwide. This study investigates how low-carbon innovation affects carbon emissions by the Dynamic Spatial Durbin Model based on 30 Chinese provinces from 2007 to 2017. The empirical results show that: First, there exists provincial agglomeration of carbon emissions. High emission provinces concentrate in major economic zones and energy extraction areas. Second, low-carbon innovation decreases carbon emissions in general. The spillover effects are higher than the direct effects in the short run, but the spillover effects are not significant in the long run. Third, the environmental Kuznets curve (EKC) hypothesis is valid both in the short-run and long-run. Ninety percent of the provinces' GDP is above the inflection point by 2017. The summary of policy implications is as follows. First, targeted incentives for R&D in low-carbon technologies are needed; Second, the externalities of low-carbon innovation require attention; Third, energy transition need to be promoted as soon as possible.