AI adoption raises productive capability in China’s new-energy vehicle firms, driven by reorganised R&D teams and stronger innovation rather than immediate increases in R&D spending; gains are largest in Jiangsu and Zhejiang and among smaller companies.
Mechanisms and Effects of Artificial Intelligence on New Quality Productive Forces in New Energy Vehicle Firms: Evidence from the Yangtze River Delta
Using a 2001–2023 panel of NEV firms in the Yangtze River Delta, the paper finds that firm-level AI adoption significantly raises firms' 'new quality productive forces', with larger effects in Jiangsu and Zhejiang and among small firms, operating through R&D personnel optimization and higher innovation output while R&D spending itself is not a positive mediator.
Artificial intelligence is increasingly recognized as an important driver of firms’ new quality productive forces. Based on panel data of new energy vehicle firms in the Yangtze River Delta from 2001 to 2023, this paper constructs firm-level indicators of artificial intelligence and new quality productive forces, and examines the impact of artificial intelligence on the new quality productive forces of new energy vehicle firms and its underlying mechanisms. The results show that artificial intelligence significantly promotes the growth of new quality productive forces in new energy vehicle firms, and this conclusion remains robust after addressing endogeneity concerns and conducting robustness checks. In addition, R&D expenditure does not constitute a significant mediating channel, although its negative path coefficient suggests a possible short-term adjustment effect. This indicates that the expansion of R&D funds does not necessarily translate immediately into improvements in firms’ new quality productive forces. Heterogeneity tests show that the promoting effect of artificial intelligence is more pronounced in Jiangsu and Zhejiang and among small-sized enterprises. Mechanism tests provide empirical evidence that artificial intelligence affects firms’ new quality productive forces through the optimization of R&D personnel structure and the improvement of innovation output.
Summary
Main Finding
Artificial intelligence (AI) adoption significantly and robustly increases firm-level new quality productive forces (NTFP) in new energy vehicle firms in the Yangtze River Delta (2001–2023). The effect operates primarily via optimization of R&D personnel structure and improved innovation output; R&D expenditure does not mediate the effect and shows a short-term negative path consistent with staged resource reallocation.
Key Points
- Positive effect: Higher firm-level AI is associated with greater NTFP (high-technology, high-efficiency, high-quality productive forces). Results hold after robustness checks and attempts to address endogeneity.
- Mechanisms:
- R&D personnel structure optimization and skill upgrading (increased proportion and improved composition of R&D staff) help convert AI adoption into NTFP gains.
- Improved innovation output (greater/increasing effective invention patenting) raises conversion of innovation into productive forces.
- R&D expenditure shows a negative short-run coefficient → evidence of staged resource reallocation (investment in platforms/systems/organizational change) rather than a lasting negative effect.
- Heterogeneity: AI’s positive effect on NTFP is stronger in Jiangsu and Zhejiang provinces and among small-sized firms.
- Measurement contributions: Constructs firm-level indices for NTFP (entropy-based across new-type workers, means of labor, objects of labor) and a firm-level AI measure based on log(frequency) of 80 AI-related keywords in annual reports.
Data & Methods
- Sample: Panel of new energy vehicle firms in the Yangtze River Delta, 2001–2023.
- Dependent variable: NTFP — composite index via entropy method using indicators in three dimensions:
- New-type workers: number and share of R&D personnel.
- New-type means of labor: net fixed assets, capitalized R&D expenditure, R&D capitalization intensity.
- New-type objects of labor: gross operating margin, operating revenue, net intangible assets, counts of independently applied invention and utility model patents, R&D expenditure and its share of revenue.
- Key explanatory variable: AI level (lnwords) = ln(1 + total frequency of 80 AI-related keywords in firms’ annual reports).
- Mediators: proportion of R&D personnel (staff), log of invention patent applications (lnpat), log of R&D expenditure (lnexp).
- Controls: return on assets (roa), ln(total assets), ln(employees), ln(government subsidies).
- Econometric approach: two-way fixed effects panel regressions (firm and year fixed effects). Authors report robustness checks and addressing endogeneity (details provided in full paper).
- Mechanism tests: mediation analysis on R&D personnel structure, patenting, and R&D expenditure to establish channels.
Implications for AI Economics
- Micro-level effects: AI can materially raise firms’ “new quality productive forces,” highlighting the importance of firm-level adoption and organizational change (not just aggregate TFP measures).
- Human capital & organization matter: The dominant channels are changes in R&D team composition/skills and improved innovation efficiency — models and empirical work on AI’s economic effects should include human-capital and organizational allocation channels, not only capital-labor substitution.
- R&D spending interpretation: Aggregate R&D expenditure increases may not immediately translate into higher productive quality; short-run negative or non-significant effects can reflect reallocation to digital platforms, training, and systems. Dynamic/nonlinear specifications are important.
- Heterogeneity & policy targeting: Regional and firm-size differences matter — policy and firm strategies should be tailored (e.g., stronger support in areas/firms where AI adoption yields larger returns or to help lagging provinces/firms catch up).
- Measurement suggestions: Text-based AI adoption indicators (keyword frequency) and firm-level NTFP indices are feasible ways to study AI impacts empirically; however, researchers should be mindful that word-frequency proxies capture attention/communication as well as deployment.
- Future research: extend to other industries/regions, refine causal identification of AI adoption (e.g., instrumenting AI deployment or using quasi-experiments), obtain direct measures of AI capital and usage intensity, and explore medium-/long-run dynamics of R&D reallocation and productivity conversion.
Assessment
Paper Typequasi_experimental
Evidence Strengthmedium — Uses long firm-level panel data and tests mechanisms, and reports robustness checks that address endogeneity concerns, which supports a credible link between AI and firm-level productive forces; however the summary does not provide details on instrument validity, exogeneity assumptions, or potential remaining confounders, and the AI indicator and outcome are constructed measures that may suffer measurement error—so causal claims are plausible but not ironclad.
Methods Rigormedium — Methodologically sound elements (panel data, robustness checks, heterogeneity and mediation analyses) are reported, but the summary lacks key methodological details (exact econometric identification, strength and validity of instruments, controls used, sample size) needed to judge rigor as high; possible issues include measurement of AI, remaining endogeneity, and omitted variable bias.
SampleFirm-level panel of new energy vehicle (NEV) firms located in the Yangtze River Delta region of China (including Jiangsu and Zhejiang) covering 2001–2023; authors construct firm-level indices for 'artificial intelligence' and 'new quality productive forces' and use firm financials, R&D expenditure, personnel structure and innovation output as covariates/mediators; sample size and exact firm selection criteria are not reported in the summary.
Themesproductivity innovation human_ai_collab adoption
IdentificationFirm-level panel regressions exploiting within-firm variation over 2001–2023 with firm and year controls; authors report addressing endogeneity via robustness checks and alternative specifications (including instrumental-variables and/or dynamic-panel approaches as reported), and test mediating channels (R&D personnel structure and innovation output). Exact identification details (instruments, timing assumptions) are not specified in the summary.
GeneralizabilityIndustry-specific: focused on new energy vehicle (NEV) firms — may not generalize to other sectors., Region-specific: Yangtze River Delta (China) — China-specific institutional and market conditions limit external validity., Time frame: 2001–2023 spans periods with very different AI technologies — early years may not reflect modern generative AI effects., Measurement: constructed AI and 'new quality productive forces' indices may be context-specific and not comparable across settings., Firm heterogeneity: stronger effects found in certain provinces and small firms, suggesting limited applicability to large firms or other regions.
Claims (10)
| Claim | Direction | Confidence | Outcome | Details |
|---|---|---|---|---|
| Artificial intelligence significantly promotes the growth of new quality productive forces in new energy vehicle firms. Firm Productivity | positive | high | new quality productive forces |
0.48
|
| The positive effect of artificial intelligence on firms' new quality productive forces remains robust after addressing endogeneity concerns and conducting robustness checks. Firm Productivity | positive | high | new quality productive forces |
0.48
|
| R&D expenditure does not constitute a significant mediating channel between artificial intelligence and firms' new quality productive forces. Firm Productivity | null_result | high | new quality productive forces (mediating role of R&D expenditure) |
0.48
|
| The path coefficient for R&D expenditure is negative, suggesting a possible short-term adjustment effect (even though the mediation is not significant). Firm Productivity | negative | high | R&D expenditure path to new quality productive forces |
0.08
|
| The promoting effect of artificial intelligence on new quality productive forces is more pronounced in Jiangsu and Zhejiang provinces. Firm Productivity | positive | high | new quality productive forces (regional heterogeneity) |
0.48
|
| The promoting effect of artificial intelligence on new quality productive forces is more pronounced among small-sized enterprises. Firm Productivity | positive | high | new quality productive forces (firm-size heterogeneity) |
0.48
|
| Artificial intelligence affects firms' new quality productive forces through optimization of R&D personnel structure. Firm Productivity | positive | high | new quality productive forces (mediated by R&D personnel structure) |
0.48
|
| Artificial intelligence affects firms' new quality productive forces through improvement of innovation output. Firm Productivity | positive | high | new quality productive forces (mediated by innovation output) |
0.48
|
| The paper constructs firm-level indicators of artificial intelligence and new quality productive forces for new energy vehicle firms. Other | positive | high | construction of measurement indicators (methodological contribution) |
0.8
|
| The empirical analysis is based on panel data of new energy vehicle firms in the Yangtze River Delta from 2001 to 2023. Other | null_result | high | dataset/time coverage |
0.8
|