Artificial Intelligence, Green Innovation, and Firm Performance: Evidence from Emerging Economies

Abstract

There are two types of artificial intelligence (AI) – sector-specific and technological – that can contribute to environmental innovation, but each one is context-dependent. Firm-level data are analyzed for firms in selected BRICS Plus countries. The country coverage is based on 2024 technology classifications, 2020-2024 averaged controls and 2015-2024 secondary performance panels. The design combines topic and category classification, averaged firm controls, and panel performance results. The difference in the probability of green-innovation classification was not statistically significant for broad AI classification. Large average differences are rejected by equivalence tests only at the broader pre-specified margin, and matching and weighting estimates are close to zero after improved observed covariate balance. The near-zero average masks important technological and sectoral variation. Industrial automation, as well as the Artificial Intelligence category, machine learning, predictive analytics and process automation, are positively associated with green innovation, whereas the latter is negatively associated after false-discovery-rate adjustment. One possible explanation for this is technological proximity to production: while the AI that relies more on technology than on software can have a more direct impact on energy, materials, process control, and maintenance, the AI that relies more on software has more general functionality and may be used more for commercial, financial, risk, and administrative purposes. The evidence of country interaction is not statistically significant, while the country-level inference is constrained by the common support, and the heterogeneity is more pronounced at the sectoral level. R&D proxies and intangible-capability proxies do not materially strengthen the association between AI-related technological orientation and green-innovation classification, even though AI-classified firms are more R&D-oriented. These findings support a conditional-deployment view: the environmental relevance of AI-related technological orientation depends on application domain, sectoral production architecture, and strategic orientation. Because classifications are undated, the results should be read as conditional associations rather than causal adoption effects.