AI Can Fuse Digital Growth With Decarbonisation, But Not for Every Manufacturer

A new study suggests artificial intelligence (AI) can help manufacturers make digitalisation and environmental progress reinforce each other, rather than advance as separate corporate agendas. By linking production, energy, emissions and operational data, AI can help firms improve efficiency while also reducing waste, strengthening environmental management and supporting cleaner industrial processes.

Published in Sustainability, "Can Artificial Intelligence Adoption Empower Digital–Green Synergy in Manufacturing Firms? Evidence from Large Language Models," by Xiao Tang, Guizhu Tan, Da Gao and Qingshuo Wang, analyses Chinese A-share manufacturing firms from 2011 to 2022. The researchers find that AI adoption increased digital–green synergy by an amount equivalent to roughly 12.6% of the sample average, although the strongest gains appeared among high-tech, non-heavy-polluting and non-state-owned firms, indicating that technology works best when backed by skills, flexible organisations and environmental leadership.

AI's Green Value Lies in Coordination, Not Automation

Much of the industrial AI debate remains focused on automation: replacing repetitive tasks, improving predictive maintenance or lowering operating costs. The study pushes the discussion further by showing that AI's most important sustainability contribution may come from coordination.

Manufacturers often manage digital and green transformation through separate departments, budgets and performance systems. Digital projects are expected to improve productivity and speed, while environmental programmes focus on compliance, emissions or resource efficiency. The result can be technological fragmentation: a factory becomes more connected without becoming less polluting, or it adopts environmental targets without the data infrastructure needed to deliver them.

AI can help bridge that divide because it processes information across production, finance, energy use, equipment performance and emissions. Instead of optimising each function separately, firms can embed environmental constraints directly into production scheduling, investment planning and equipment control.

The study measures digital–green synergy through a coupling model that combines corporate digitalisation and greening indicators. It constructs firm-level AI adoption using a BERT-based language model trained to identify whether companies were actually applying AI, rather than merely mentioning fashionable technologies in annual reports.

Robot density, patent applications and keyword counts are imperfect proxies: they may capture technological interest or capacity without proving operational use. By analysing the management discussion and analysis sections of annual reports, the study attempts to distinguish genuine adoption from corporate rhetoric.

The baseline findings remain positive across alternative measures, lagged specifications, policy-shock analysis, propensity-score matching, entropy balancing and instrumental-variable tests. These checks do not make the research fully causal, but they strengthen the case that the observed relationship is not simply a statistical accident.

The Real Engine Is Knowledge, Capital and Organisational Change

According to the study, purchasing AI technology is not enough. The benefits emerge through three organisational pathways: wider knowledge, better resource allocation and the disruption of rigid routines.

• AI expands firms' knowledge breadth. Manufacturers pursuing green and digital transformation need to integrate information from fields that do not always interact naturally: software, materials science, energy management, production engineering and environmental regulation. AI can scan patents, technical documents and operational data to identify connections that traditional, experience-based systems may miss. The cognitive expansion can help firms discover where digital technology supports cleaner production, such as using machine learning to reduce energy waste, redesign processes or identify low-carbon materials.
• AI improves resource allocation. Green investments often have longer payback periods and less visible short-term returns than conventional productivity projects. Managers may therefore favour digital investments that cut costs quickly while delaying environmental upgrades. AI can incorporate multiple objectives, financial returns, energy use, emissions and operational risk, into a single decision framework. AI adoption is associated with lower investment inefficiency, suggesting that firms direct resources more effectively toward projects generating both digital and environmental value.
• AI can break organisational routines. Manufacturing firms frequently rely on hierarchical decision-making and departmental boundaries. Environmental teams, production managers and technology units may hold different data and pursue different targets. AI-driven systems encourage shared information, cross-functional decisions and more rapid organisational adaptation.

AI creates value when firms redesign how they learn, allocate capital and make decisions. Without those changes, advanced tools can simply automate outdated practices.

The Benefits Are Concentrated, and That Should Worry Policymakers

The study finds that AI's positive effect is strongest in high-tech firms, non-heavy-polluting manufacturers and non-state-owned enterprises. It is not statistically significant among non-high-tech, heavily polluting or state-owned firms. These differences expose a critical industrial-policy challenge. The firms that could benefit most from digital–green upgrading may be the least prepared to use AI effectively.

High-tech manufacturers typically possess stronger research capabilities, deeper talent pools and more mature digital infrastructure. They can absorb AI faster and connect it to product development, energy management and environmental innovation.

Heavy polluters face a different problem. Their AI investments may be concentrated on monitoring emissions and meeting compliance requirements rather than redesigning core production systems. That defensive use can improve reporting without transforming underlying processes.

State-owned firms may also respond more slowly because they operate under multiple policy, employment and social objectives. Their incentive structures can dilute the pressure to integrate AI rapidly into strategy and operational reform. Non-state-owned firms, facing stronger competitive pressure, may have more reason to move quickly.

For governments, the implication is that universal AI subsidies are unlikely to produce universal green gains. Support must be differentiated. High-tech firms may need incentives to push AI into advanced applications such as low-carbon materials, intelligent energy systems and circular manufacturing. Heavy-polluting industries require assistance in moving beyond end-of-pipe monitoring toward full-process redesign. State-owned firms may need performance systems that reward digital–green integration rather than treating digitalisation and environmental compliance as separate mandates.

Small and medium-sized manufacturers present another concern. They were not the study's primary focus, yet they often lack the capital, data, technical staff and governance systems needed for AI adoption. Without shared infrastructure, model-as-a-service platforms and affordable technical support, the green benefits of industrial AI may become concentrated among already advanced firms.

Technology Cannot Outrun Leadership or Policy Uncertainty

The study also shows that AI's impact depends heavily on leadership and policy stability. Firms with executives who have environmental backgrounds gain more from AI adoption. This finding is strategically important because it confirms that technology needs interpretation. AI can generate forecasts, identify inefficiencies and compare investment options, but leaders still decide which objectives matter and how results are used.

Executives with environmental knowledge are more likely to embed sustainability into AI strategy, capital allocation and performance measurement. Without that expertise, firms may deploy AI primarily to increase output or reduce labour costs, leaving environmental gains secondary.

Climate-policy uncertainty has the opposite effect. When firms cannot anticipate future emissions standards, subsidies, compliance costs or green-finance conditions, they are more likely to delay long-term investments. AI-enabled manufacturing systems often require high upfront expenditure and organisational disruption; unstable policy signals make those commitments harder to justify.

The study thus connects corporate technology strategy with public governance. Stable targets, phased implementation and predictable incentives are not administrative details; they shape whether firms use AI for short-term efficiency or long-term industrial transformation.

The research also links stronger digital–green synergy with better corporate social responsibility performance, higher "new-quality productivity" and greater supply-chain resilience. These outcomes suggest that the benefits extend beyond emissions. Firms combining digital and green capabilities may detect risks earlier, optimise logistics, innovate more effectively and respond faster to disruptions.

Still, caution is necessary. The sample includes listed Chinese manufacturers, which generally possess stronger governance, financing and disclosure than smaller firms. Annual reports may overstate implementation, even when analysed with advanced language models. The study cannot fully eliminate reverse causality: stronger and more sustainable firms may be more likely to adopt AI in the first place. It also does not account fully for AI's own environmental footprint, including the energy and material demands of data centres, computing hardware and model training.