Artificial Intelligence and firm total factor productivity: evidence from digital capital accumulation and labour structure adjustment

Drawing on an unbalanced panel of Chinese A-share listed manufacturing firms from 2010 to 2020, this study examines the relationship between firm-specific artificial intelligence capability and total factor productivity (TFP), the accompanying capital and labour adjustments, and heterogeneity across industrial contexts and AI subfields. Using a patent-based measure of AI capability, we estimate firm-level models with high-dimensional fixed effects, while a staggered pilot-zone design provides quasi-experimental corroboration. Stronger AI capability is associated with higher TFP, greater intangible investment, higher headcounts of highly educated employees, and lower headcounts of less-educated employees. The productivity relationship is stronger in high-technology manufacturing and in industries with broader digital application scenarios, while machine learning exhibits the most stable positive association among AI subfields. Reweighting the listed sample to match the official distributions of firm size, industry, and ownership among above-scale manufacturing enterprises leaves the estimate essentially unchanged. The findings document a positive within-firm AI–TFP relationship among listed manufacturers.