Sage Journals: Discover world-class research

Abstract

Based on data from manufacturing companies of listed companies in China from 2011 to 2024, examine the impact of AI-driven technological innovation in manufacturing. Research has found that AI-driven technological innovation significantly improves the transformation and upgrading of manufacturing companies, and that corporate innovation inputs and innovation outputs play a mediating effect in this process. Further heterogeneity analysis found that AI-driven technological innovation has a more significant effect on improving companies and state-owned enterprises with higher levels of employee education. AI-driven technological innovation has a threshold effect on the transformation and upgrading of the manufacturing industry, which is manifested as a “inverted U-shaped” nonlinear relationship.

Keywords

AI driven manufacturing upgrading technological innovation innovation output innovation input

Introduction

As the fourth industrial revolution advances, artificial intelligence technology is reshaping the global manufacturing landscape with unprecedented speed and breadth. In 2016, known as the “first year of artificial intelligence,” the landmark event of AlphaGo’s victory over human Go champion Lee Sedol triggered widespread global attention to artificial intelligence technology and heralded the accelerated application of AI technology industrialization. Major economies have elevated artificial intelligence to a national strategy: the United States released the National Artificial Intelligence Research and Development Strategic Plan; Germany continues to advance the “Industry 4.0” Plan, and China clearly promotes artificial intelligence in the “13th Five-Year Plan” National Science and Technology Innovation Plan. Artificial intelligence is a key development direction for scientific and technological innovation. The transformation and upgrading of the manufacturing sector, as the main body of the national economy, is directly related to national competitiveness (Porter & Heppelmann, 2014). The global manufacturing industry is undergoing profound changes from traditional manufacturing to intelligent manufacturing. Artificial intelligence technology empowers key links in technological innovation such as R&D, design, and manufacturing, and is reconstructing the value creation model of the manufacturing industry (Xu et al., 2018).

Figure 1 shows the annual changes in the AI index and manufacturing industry upgrade scores of listed companies in China from 2011 to 2024. On the one hand, the AI index of Chinese listed companies, which increased year by year from 2011 to 2024, fell sharply in 2021 and picked up in 2024, rising from 0.07 in 2011 to 0.21 in 2024. The overall trend of manufacturing industry upgrade scores of Chinese listed companies from 2011 to 2024 is unclear, with constant ups and downs. The trend changes between the two are not completely consistent. The specific impact of artificial intelligence-driven technological innovation on the upgrading of manufacturing industries is complex and changeable, and further detailed investigation needs to be carried out through empirical regression.

Figure 1.

Trend chart of changes in the artificial intelligence index and manufacturing industry upgrading of China’s A-share listed companies from 2011 to 2024.

Figure 2 shows the change in the annual mean value of the two sub-dimension scores of innovation output and innovation input of listed companies from 2011 to 2024. The scoring of the innovation input dimension and innovation output dimension showed an overall upward trend from 2011 to 2024. The improvement of the scoring of the innovation input dimension and innovation output dimension shows that Chinese companies are paying more and more attention to technology research and development and innovation output, and have taken corresponding measures.

Figure 2.

Trend chart of changes in manufacturing innovation sub-dimensional scores of China’s A-share listed companies from 2011 to 2024.

Theoretical-Mechanistic Analysis

The Impact of AI-Driven Technological Innovation on Industrial Upgrading in Manufacturing

AI-driven technological innovation optimizes production processes, improves production efficiency, and significantly improves the transformation and upgrading of manufacturing companies (Jiang, 2025). Real-time data monitoring and predictive maintenance of intelligent manufacturing systems reduce equipment failure rates and reduce downtime losses for enterprises. AI-driven automated production lines reduce manual intervention, improve production accuracy and consistency, and enhance the market competitiveness of enterprises. AI technology helps companies better adapt to changes in market demand and enhance their ability to resist risks. The integration of AI and manufacturing promotes corporate innovation and drives product upgrades (Trajtenberg, 2018). The number of AI patents in the manufacturing industry is an important measure of technology integration, while companies with a large number of patents usually have stronger technical barriers and market voice, and occupy a dominant position in competition. Based on the above analysis, we propose hypothesis:

H1: AI-driven scientific and technological innovation helps accelerate the upgrading process of manufacturing and provides continuous impetus for its high-quality development.

Analysis of the Mediating Effect of AI-Driven Technological Innovation on Industrial Upgrading in Manufacturing

Artificial intelligence has improved the data mining capabilities of enterprises, allowing them to quickly collect and analyze process technology data, identify innovation needs and seek breakthroughs; artificial intelligence technology has reduced reliance on product prototypes and improved product research and development efficiency; artificial intelligence models can Predict technological energy consumption, shorten product development cycles, and efficiently carry out green product research and development. Artificial intelligence technology reduces corporate resource constraints, controls costs, and reduces risks, significantly improving the efficiency of green technology innovation, thereby promoting the upgrading of the manufacturing industry (Zhong et al., 2025). Based on this, we propose hypothesis:

H2: AI-driven technological innovation improves industrial upgrading in manufacturing by promoting green technology innovation mechanisms.

The new trade theory states that heterogeneous factors such as the innovation output of enterprises influence product quality and export decisions. Enterprises with higher innovation output have more advanced technologies and more efficient production processes to produce higher quality products (Sun et al., 2025) to improve consumer satisfaction and social trust. The promotion of innovative output simplifies production processes, reduces the demand for manual work, inhibits gender inequality at the employment level and enhances the inclusion and diversity of enterprises (Lee et al., 2025). According to Engel’s law, an increase in income motivates households to invest more in human capital, raising the level of human capital in society. Therefore, the improvement of the innovation output of enterprises can improve product quality, reduce employment gaps, and enhance the level of human capital, thereby promoting industrial upgrading in the manufacturing industry. Based on this, we propose the hypothesis:

H2a: AI-driven technological innovation improves the innovative output of enterprises and promotes the upgrading of the manufacturing industry.

AI technology enhances the information collection and processing capabilities of enterprises, translates large amounts of production and operation data into structured and standardized information, improves the internal information environment, and enhances the timeliness and accuracy of information transmission (Lu et al., 2025). At the same time, AI strengthens communication between businesses and stakeholders, suppresses distortion and interception issues in information delivery, reduces information loss, and increases innovation investment by businesses. Therefore, we propose hypothesis

H2b: AI-driven technological innovation improves manufacturing industry upgrading by increasing corporate investment in innovation.

Analysis of the Impact Mechanism of AI-Driven Technological Innovation on Industrial Upgrading in Manufacturing

Based on a theoretical analysis of technological innovation and diminishing marginal benefits. Artificial intelligence-driven improvement of innovation efficiency in the early stages of technology application significantly promotes manufacturing upgrading. However, when the level of technology application is too high, enterprises may face problems such as technology overload, resource mismatch, and increased management complexity, resulting in a decrease in the efficiency of technology application. This inhibits manufacturing upgrading. Second, the theory of organizational learning and absorptive capacity further explains this phenomenon. The technological innovation effect of an enterprise is influenced by its absorptive capacity, that is, the ability of the enterprise to identify, digest and apply external knowledge. When the level of AI-driven technological innovation exceeds the absorptive capacity of enterprises, the effectiveness of technology applications decreases significantly. Enterprises need to have the corresponding technical capabilities, management capabilities, and human resource support to effectively manage and utilize high-level AI technologies. If the level of technology application is too high and the absorptive capacity of enterprises is insufficient, it will lead to a decrease in the efficiency of technology application and even have a negative impact on manufacturing upgrading. Resource constraints and resource allocation theory shows that: an enterprise’s resources are limited, and the effect of technological innovation depends on the rational allocation of resources. When the level of AI-driven technological innovation is too high, companies may focus too much resources on technology research and development and application, while ignoring other key areas, such as market expansion and product optimization, resulting in unbalanced resource allocation, which in turn inhibits manufacturing upgrades. In summary, hypothesis H3 of this thesis is presented:

H3: There is a threshold effect on the impact of AI-driven STI on manufacturing upgrading.

Models and Data

Basis for Sample Data Selection

The data analysis in this paper primarily relies on data from publicly listed manufacturing companies in China (see Table 1). China’s manufacturing value added accounts for over 30% of the global total, maintaining the top position globally for 13 consecutive years. By selecting Chinese samples, the study directly investigates the technological transformation patterns of the world's largest manufacturing system. The “Made in China 2025” strategy clearly identifies “intelligent manufacturing” as the primary focus, while policies such as the “New Generation Artificial Intelligence Development Plan” position AI as the core engine for manufacturing upgrades, creating the most comprehensive policy-industry linkage experimental field in the world. The “Information Disclosure Management Measures for Listed Companies” in China mandates the disclosure of technology strategies in the MD&A section, making data from listed companies more accessible than that of non-listed firms or SMEs. Furthermore, the R&D investment intensity of China’s A-share manufacturing companies (2.89%) significantly exceeds the industry-wide average (1.84%), making it an ideal window for observing the application of AI technologies.

Table 1.

Descriptive Statistical Characteristics of the Sample.

Variable	Quantity	Mean	Median	Standard deviation	Maximum value	Min
Y	36,214	3.99	4.38	0.99	6.41	2.25
AI	36,214	0.24	0.11	0.34	1.08	0.00
Board	36,214	2.23	2.07	0.2	2.69	1.87
ROA	36,214	0.04	0.04	0.08	0.18	0.16
Size	36,214	20.72	20.66	1.15	22.56	18.37
Leverage	36,214	0.41	0.41	0.21	0.87	0.06
TOP1	36,214	0.31	0.32	0.16	0.62	0.05
Age	36,214	3.12	2.94	0.28	3.57	2.48
Growth	36,214	0.17	0.15	0.43	1.03	0.69
Independent	36,214	0.42	0.37	0.04	0.46	0.34
Cash	36,214	0.05	0.05	0.06	0.19	0.20

Interpreted Variables

There are many measurement methods for manufacturing upgrading level (Upgrade). They can be mainly divided into the following three categories: First, structural indicators, which use the vector angle method to calculate the technical complexity of manufacturing to measure the level of manufacturing upgrading, the second is efficiency indicators. The level of upgrading of the manufacturing sector is measured by measuring total factor productivity and labor productivity (Han et al., 2025); the third is comprehensive indicators, which measure the level of upgrading of the manufacturing sector by constructing a system of evaluation indicators (Jani & Khan, 2025). Comparing the three existing measurement methods above, it can be found that no matter what the means of promoting the transformation of the manufacturing industry, it is inseparable from the central role of technological progress as the basic driving force. For this purpose, this thesis characterizes the level of manufacturing upgrading with the indicator of technological complexity of manufacturing (Rong et al., 2025), which characterizes the manufacturing upgrading path from the perspective of technological progress. The technical complexity of manufacturing is calculated by

$y_{ct} = \frac{\sum_{i} Outpu t_{ict} \times prod y_{i, 1997}}{\sum_{i} Outpu t_{ict}}$ (1)

Among them, prody_i, 1997 represents the technical complexity of the manufacturing industry segment i in 1997. This value is obtained by simply averaging the technical complexity of each HS6-digit product in the segment i. The original data comes from the BACI database and the world macroeconomic database; Output represents the output of regional c segment industry i in year t. It should be noted that the technical complexity of the industry was fixed in 1997 in order to illustrate that the annual changes in the technical complexity of the regional manufacturing industry do not stem from the natural changes in the technical complexity of the various sub-sectors at world level, but from the impact of the transformation and upgrading of the regional manufacturing industry. Only in this way can we identify the endogenous mechanisms by which the optimization of resource allocation brought about by the market-oriented allocation of factors promotes the upgrading of regional manufacturing industries.

Core Explanatory Variable: Artificial Intelligence Technology (AI)

The measurement design for the AI-driven technological innovation variable (AI) is based on the essential characteristics of the technology and the feasibility of obtaining real-world data. The core principle is that AI, as a general-purpose technology, has pervasive application capabilities and can deeply integrate into all aspects of a company’s value chain, including R&D, manufacturing, supply chain management, and customer service. Traditional methods using industrial robot quantities as proxy variables have significant limitations. These methods mainly reflect the degree of automation in the production process and fail to cover the key roles of AI in management decision optimization, business model innovation, and other aspects. More importantly, province-level robot data cannot be accurately matched with micro-level firm behaviors and is also unable to distinguish between cutting-edge intelligent technologies and traditional automation.

Therefore, this study constructs a measurement indicator based on information in the company’s annual report. This choice is grounded in two key logical reasons: On the one hand, the “Management Discussion and Analysis” (MD&A) section of the annual report provides an official and systematic disclosure of a company’s operational status, core strategies, and technological investments. As a statutory information disclosure medium, its content must be highly authentic and comprehensive (He et al., 2025). When management discusses future development directions and core competitiveness, they inevitably highlight the deployment of advanced technologies, including AI. This textual expression directly reflects the company’s actual attention to and commitment to AI technology. On the other hand, the pervasive nature of technology application requires the measurement tool to cover multiple dimensions. By precisely identifying 15 key technical terms, such as “machine learning,”“natural language processing,” and “intelligent decision-making,” which encompass algorithms, data, and application scenarios, we can systematically capture AI’s presence across multiple business functions and avoid the one-sidedness of a single indicator (e.g., the number of patents; D. Li & Wang, 2024).

From a technical implementation perspective, the design of the indicator fully considers data comparability and signal-to-noise separation. The standardized process of dividing the total frequency of keywords by the total length of the MD&A section is crucial. This step effectively eliminates frequency interference caused by differences in report length or narrative style between companies, making core word density, rather than absolute numbers, a proxy for substantial capability. Scaling the value into a percentage form enhances the intuitive economic interpretability of the coefficients in subsequent econometric analyses. The text processing method (Python + Jieba word segmentation) ensures the standardization of data scraping and recognition, ensuring that the indicator construction process is objective and replicable (Kalidas, 2025).

This article the accounting approach was inspired by the work of Wu and Xu (2022). Takes a specific set of keywords related to AI, including AI, business intelligence, image understanding, investment decision-making assistance systems, intelligent data analysis, intelligent robotics, machine learning, deep learning, semantic search, biometrics, face recognition, speech recognition, authentication, autonomous driving, and natural language processing. Secondly, Python’s crawler technology was used to compile the annual reports of all A-share listed companies from 2011 to 2024, and the “jieba” word segmentation function was used to search and match the above keywords, and the total word frequency of these keywords in the annual report was counted (Aghion et al., 2005). Finally, the frequency of AI-related words in the annual report is calculated by dividing the length of the MD&A (Management Discussion and Analysis) segment to obtain an indicator (AI) that measures the level of AI-driven technological innovation. The larger the value of this indicator, the higher the level of AI-driven technological innovation in a business. To facilitate observation and analysis, this paper magnifies this indicator by a factor of 100.The construction of the AI variable in this paper is based on solid theoretical rationale and practical feasibility. It overcomes the limitations of traditional proxy indicators, which are often coarse-grained and unidimensional. By leveraging the source information from corporate strategy disclosures and using text analysis, this approach captures the diffuse penetration characteristics of technology within complex organizational structures. Ultimately, the AI variable serves as a micro-level proxy that is both vertically comparable (across periods) and horizontally comparable (across firms), reflecting the breadth of technology application.

Enterprise Control Variables

Based on existing relevant studies, this thesis includes the following control variables in the empirical regression analysis: size of the enterprise (Size), age of the enterprise (Age), growth level of the enterprise (Growth), return on total assets (ROA), asset-liability ratio (Leverage), cash-flow ratio (Cash), board size (Board), equity concentration (TOP1), Proportion of independent directors (Independent). Table 2 presents detailed definitions and construction instructions for the control variables used herein.

Table 2.

Control Variable Definitions and Construction Instructions.

Variable definition	Variable	Construction instructions
Enterprise size	Size	Logarithm of total assets of enterprises
Enterprise age	Age	The logarithm is taken after the year in which it is located, minus the time of establishment of the enterprise
Level of enterprise growth	Growth	Growth rate of operating income (%)
Return on total assets	ROA	The ratio of the enterprise’s investment remuneration to the total amount invested
Asset-liability ratio	Leverage	Ratio of total liabilities to total assets of the enterprise
Cash-flow ratio	Cash	Net operating cash flow to total assets
Board size	Board	Logarithmic value of the number of members of the Board
Equity concentration	TOP1	Shareholding ratio of the largest shareholder of the company
Proportion of independent directors	Independent	Share of independent directors of enterprises in the board of directors

Model Construction

To accurately examine the impact of AI-driven technological innovation on manufacturing industry upgrading (Furman & Seamans, 2019), this paper sets the following measurement equation:

$Y_{it} = β_{0} + β_{1} A I_{it} + β_{2} Z_{it} + μ_{pt} + ν_{jt} + φ_{i} + ε_{it}$ (2)

Among them, the subscripts i, p, j, and t represent enterprises, provinces, industries, and years respectively, Y is a measure of industrial upgrading in manufacturing, AI is artificial intelligence technology, and Z represents a series of enterprise-level control variables (Chen et al., 2022). The coefficient “β”₁measuring the impact of artificial intelligence technology on manufacturing industry upgrading. If β₁ > 0, it indicates that artificial intelligence-driven technological innovation can improve manufacturing industry upgrading. In order to control the unobserved factors that may affect the upgrading of the manufacturing industry due to changes in local policies, economic environment, etc. in a specific province over time, this article controls the fixed effect μpt of the province over the year; in order to control the technological progress, industry specifications, etc. of a specific industry over time changes in unobserved factors that may affect the upgrading of the manufacturing industry, this article controls the fixed effect vjt of the industry over the year; in addition, This article also controls the enterprise fixation effect φ_i, with ε_it standing for the random interference term.

Basic Estimates Results and Analysis

Basic Estimates Results

Table 3 reports the results of the baseline estimates of the impact of AI-driven technological innovation on industrial upgrading in the manufacturing industry, validating hypothesis H1. Column (1) does not include control variables, only the impact of the core explanatory variable AI on manufacturing upgrades is considered. The results show that the estimated coefficient of AI-driven technological innovation on industrial upgrading in manufacturing industry is significantly positive at the 1% level. Column (2) incorporates control variables at the enterprise level based on the results of column (1), and the core explanatory variable estimation coefficient is still significantly positive at the 1% level. Column (2) shows that for every 1% increase in the level of technological innovation driven by artificial intelligence, the level of industrial upgrading in the manufacturing industry can increase by 0.144%. The results of the control variables showed that the larger the enterprise, the larger the board of directors, the higher the concentration of equity, the higher the proportion of independent directors, the higher the total asset return, the better the upgrading of the manufacturing industry, and the worse the upgrading of the manufacturing industry, the higher the asset-liability ratio, the older the enterprise, the faster the enterprise grows, and the higher the cash flow ratio.

Table 3.

Basic Estimates Results.

Variable	(1) y	(2) y
AI	0.182*** (0.032)	0.144*** (0.030)
Board		0.136** (0.065)
ROA		0.914*** (0.124)
Size		0.241*** (0.016)
Leverage		−0.871*** (0.062)
TOP1		0.471*** (0.106)
Age		−0.291** (0.127)
Growth		−0.045*** (0.012)
Independent		1.461*** (0.201)
Cash		−0.331*** (0.085)
Constants	4.061*** (0.007)	−1.321** (0.540)
Enterprise fixed effect	Yes	Yes
Industry 1 year fixed effect	Yes	Yes
Province fixed effect for 1 year	Yes	Yes
Sample size	36,018	36,015
R² value	.563	.578

Note. *, **, and *** represent the significance levels of 10%, 5%, and 1% respectively: the standard errors of cluster robustness are in brackets; all regressions in this article cluster the standard errors at the enterprise level. Same apply to all tables.

The results in Table 3 not only align with the inherent logic of economics but also profoundly reveal the structural contradictions in manufacturing transformation.

The results stem from the uniqueness of industrial upgrading, which fundamentally conflicts with the short-term growth logic:

The core of manufacturing upgrading lies in a technology-intensive transformation, not merely in scale expansion. A high firm growth rate implies a risk of extensive, inefficient expansion. Abundant cash flow may reflect innovation inertia—high cash flow ratios are often found in mature market firms. These firms, due to technological path dependency, are more likely to distribute dividends or invest in financial management (according to annual report notes, the average R&D expenditure of the top 10 firms by cash flow in the sample is only 2.1%), rather than invest in high-risk AI research. In contrast, leading firms in upgrading often show negative operating cash flow (during the period of technological investment) but sustain long-term innovation through equity financing.

2. The consistency of other control variables strengthens the credibility of the conclusion:

The control variable system in Table 3 is highly logically consistent and collectively points to the deeper rules of transformation and upgrading: industrial upgrading requires breaking away from the traditional economic growth paradigm.

The core mechanism behind the negative effect of firm growth and cash flow ratio should be viewed within the internal contradictions of industrial upgrading. High-speed expansion in firms is often accompanied by the scale replication of traditional capacity, which essentially consolidates the existing production paradigm with capital investment, creating path dependency in technological evolution. The capacity expansion reflected by the growth rate indicator directly conflicts with the technological leap driven by AI: when firm resources are excessively allocated to low-end capacity expansion, it inevitably crowds out resources for disruptive innovations like machine learning and intelligent decision-making, thus trapping firms in a “growth trap” in the process of industrial value chain reconstruction. The paradox of scale expansion and stagnation or even decline in value chain position emerges.

The negative impact of abundant cash flow reveals the misallocation of innovation resources. Free cash flow could support technological R&D, but in transitioning economies with imperfect institutional environments, it is more likely to devolve into a protective layer of organizational inertia. In firms with imperfect governance structures, management tends to avoid high-risk disruptive innovation investments, opting instead to use abundant cash for low-risk arbitrage or homogeneous capacity expansion. While this behavior can maintain short-term financial stability, it essentially delays the generational replacement of core technologies, causing firms to miss the strategic window in the paradigm shift to smart manufacturing.

A deeper theoretical contradiction is that industrial upgrading is essentially a systemic reconstruction of the traditional production function, not just a marginal improvement of efficiency parameters. When the core standard of industrial upgrading shifts from the “scale-cost” dimension to the “intelligent-value” dimension, expansion models based on low-end factor aggregation and financial strategies based on path dependency become institutional barriers to technological breakthroughs. High cash flow and high growth may be indicators of competitive advantage in the old paradigm, but in the AI-driven industrial revolution, they may evolve into signs of organizational inertia. This distortion of value indicators in the transformation process is a micro projection of the structural dilemma in industrial upgrading faced by late-developing countries.

Endogenous Problems

Considering that there may be potential endogenous problems of reverse causality and missing variables at the company level. This paper takes the instrumental variable approach to the underlying endogeneity problem. First, this paper refers to the design method of tool variables, and selects the industry mean as the tool variable (IV1). Second, this thesis refers to the idea of construction of tool variables based on heteroskedasticity, using the third power of the difference between the AI-driven STI level and the AI-driven STI level mean by industry and province as the tool variable (IV2). By the above treatment of AI-driven STI levels, this paper is able to guarantee that the instrumental variables, after controlling for fixed effects, are not related to the residual terms of industrial upgrading of individual manufacturing industries and are highly related to the actual AI-driven STI levels of this classification.

Columns (1) and (2) of Table 4 show the regression results based on the first instrumental variable method, and columns (3) and (4) show the regression results based on the second instrumental variable method. The first-stage regression results show that the industry mean instrumental variable (IV1) has a coefficient of 0.790***, indicating that the overall technological environment of the industry significantly drives firms’ AI investments. For every 1 unit increase in AI levels at the firm level, a 1 unit increase in the industry mean leads to a 0.790 unit increase in the firm’s AI level. These first-stage results validate the scale effect of industry-wide technology diffusion. Additionally, the first-stage F-statistic (142.34) is well above the critical value of 10, proving a strong correlation between IV1 and AI, thus satisfying the basic conditions for instrument validity.

Table 4.

Dealing With Endogenous Problems.

Variable	(1) AI	(2) Y	(3) AI	(4) Y
IV1	0.790*** (0.031)
IV2			0.355*** (0.014)
AI		0.155** (0.063)		0.055*** (0.016)
Constants	−0.495*** (0.154)	−1.545*** (0.534)	−0.416*** (0.122)	−1.480*** (0.540)
Firststage F-statistic	142.34		140.48
Kleibergen-Paap rk LM value		372.65***		85.78***
Control variables	Yes	Yes	Yes	Yes
Enterprise fixed effect	Yes	Yes	Yes	Yes
Industry 1 year fixed effect	Yes	Yes	Yes	Yes
Province fixed effect for 1 year	Yes	Yes	Yes	Yes
Observations	36,187	36,187	35,006	35,006
R² value	.805	.583	.855	.574

Meanwhile, the heteroscedasticity-based instrumental variable (IV2) has a coefficient of 0.355***, indicating that firms are highly sensitive to the industry-province technology gradient, and they tend to fill local technology gaps. The technological input of firms shows a significant positive correlation with the local technological gap. The statistical significance of IV2 is also supported by the first-stage F-statistic (140.48), further confirming the validity of the instrument.

The second-stage regression results further confirm the positive impact of AI on industrial upgrading. Based on IV1, the estimated coefficient for AI is 0.155**, showing that after controlling for endogeneity, for every 1 unit increase in AI levels, the degree of industrial upgrading increases by an average of 0.155 units. This result reflects the systemic driving effect of overall industry technology improvements on industrial upgrading. For IV2, the estimated coefficient for AI is 0.055***, which is lower than the IV1 estimate, suggesting that the marginal contribution of firms’ technological investments based on the local technology gradient to industrial upgrading is relatively small. This discrepancy may arise from the limitations of local technological catch-up, where firms’ technological inputs in the process of filling local technology gaps are constrained by the local technological environment.

Nonetheless, the first- and second-stage F-statistics are both above the critical value of 10, indicating a strong correlation between the chosen instrumental variables and the core explanatory variable. The Kleibergen-Paap rk LM statistic significantly rejects the null hypothesis at the 1% level, confirming the instruments’ identifiability. The estimated results from both instrumental variables show a significant positive effect of AI on industrial upgrading, with the coefficients being statistically significant, further validating the robustness of the core conclusion.

Overall, the results in Table 4 systematically reveal the promoting effect of AI on industrial upgrading from two dimensions: overall industry technology diffusion and local technology gradients. The industry mean instrumental variable (IV1) captures the scale effect of the industry technology environment, while the heteroscedasticity-based instrumental variable (IV2) focuses on the local effect of firms filling local technology gaps. Despite the numerical differences in the estimates from the two instruments, their directions are consistent, and the statistical significance is well supported, jointly proving the positive impact of AI-driven technological innovation on industrial upgrading. This result not only provides empirical evidence for policy-making but also offers theoretical reference for firms’ strategic decisions on technological investment.

Robustness Test

Corrected Sample Selection Bias

Whether an enterprise’s AI technology is affected by corporate characteristics such as its financial situation, boardroom situation, and external environment, may have sample self-selection issues in this article. To overcome the sample selectivity bias described above, the data were preprocessed using the propensity score matching method (PSM), creating groups of introduced AI enterprises (processing groups) and groups of non-introduced AI enterprises (control groups) that are similar in key characteristics. Further regression models estimated the impact of AI-driven STI on industrial upgrading in manufacturing. The test results of the matched samples are shown in column (1) in Table 5, which shows that after correction of the self-selection bias by the PSM, the estimation coefficient of the AI is still significantly positive, supporting the conclusions in the benchmark estimation.

Table 5.

Robustness Test.

Variable	(1) y	(2) y	(3) YR	(4) YBG	(5) f. Y	(6) y	(7) Y
AI	0.165*** (0.076)		1.025*** (0.243)	2.555*** (0.405)	0.115** (0.035)		0.095** (0.054)
AI_Sub		0.040** (0.009)
L_AI					0.103*** (0.054)
Constants	−3.235*** (0.882)	−1.352** (0.542)	17.52*** (4.232)	−9.281 (0.275)	0.541 (0.586)	−1.463** (0.619)	−3.451*** (1.014)
Control variables	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Enterprise fixed effect	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Industry 1 year fixed effect	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Province fixed effect for 1 year	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Observations	29,540	35,069	35,069	15,648	31,627	31,658	21,834
R² value	.723	.577	.714	.869	.605	.596	.677

Replacing Core Variables

In order to ensure that the rating and measurement of manufacturing industry upgrades are robust, and taking into account the time span, sample coverage and domestic and foreign differences of each institution’s ratings, this article selects the manufacturing industry upgrade rating (YR) of the China Research Data Service Platform (CNRDS) and the manufacturing industry upgrade rating (YBG) to replace the explained variables. The estimation results are shown in columns (3) and (4) of Table 5, and the estimation results are still significantly positive at the 1% level after replacing the explained variables.

Lag Term and Front-End Term

Column (5) in Table 5 for the explained variables (FY) is processed in the first phase, and column (6) is processed in the first phase of the core explanatory variables (LAI). Estimation results indicate that AI-driven technological innovation significantly contributes to industrial upgrading in the manufacturing industry, both in terms of lagging the core explanatory variables and pre-processing the explained variables.

Different Stage Samples Were Used

The time interval of the sample selected for the benchmark regression in this paper is 2011 to 2024, and considering that AI innovation policies may have new impact on estimates, this paper retains data from 2017 and later for re-estimation. The results are shown in column (7) of Table 5. After switching to samples at different stages, the estimation coefficients of AI-driven technological innovation for industrial upgrading of the manufacturing industry are all significantly positive at the 1% level, once again indicating that the basic conclusions of this paper are robust.

Empirical Analysis

Analysis of the Mechanism of Action

Benchmark regression has verified the core hypothesis H1 of this paper, that is, artificial intelligence-driven technological innovation is conducive to improving the manufacturing industry upgrading of enterprises. According to the theoretical analysis above, AI-driven scientific and technological innovation improves innovation output and green innovation mechanisms increase enterprises’ innovation output input to improve the level of manufacturing industry upgrading of enterprises. Next, this thesis performs a mechanistic empirical test, setting the following regression model:

$l n G r e e n_{i i} = β_{0} + β_{1} A I_{i i} + β_{2} Z_{i i} + μ_{p i} + v_{j i} + φ_{i} + ε_{i t}$ (3)

$Y_{it} = α_{0} + α_{1} lnGree n_{it} + α_{2} Z_{it} + μ_{pt} + v_{jt} + φ_{i} + δ_{it}$ (4)

In formula (2), lnGreen represents green technology innovation and consists of two variables: enterprise innovation output (TFP_LP) and enterprise innovation input (lnC). $ε_{it}$ and $δ_{it}$ are the random interference term of formula (3–4). The meanings of other variables are the same as those of the benchmark regression model. If the coefficient β₁ in Equation 2 is significant, it means that the mediating variable has a significant impact on the upgrading of the manufacturing industry through technological innovation driven by artificial intelligence; if the coefficient α₁ in Equation 3 is significant, it means that the mediating variable has a significant impact on the upgrading of the manufacturing industry. Manufacturing industry upgrading has a significant impact; if both coefficients are significant, it can be proved that artificial intelligence can have a significant impact on the upgrading of the manufacturing industry through the above-mentioned intermediary variables.

The results of the intermediary test for green technology innovation are shown in columns (1) and (2) in Table 6. Among them, the regression coefficient of AI on lnGreen in column (1) is significantly positive, indicating that technological innovation driven by artificial intelligence can significantly increase the number of corporate green patent applications and promote corporate green technology innovation; the regression coefficient of lnGreen on manufacturing industry upgrading in column (2) The regression coefficient is also significantly positive, indicating that green technology innovation can significantly improve the level of manufacturing industry upgrading. Therefore, the empirical test verifies the hypothesis H2 that AI-driven technological innovation improves the level of industrial upgrading in manufacturing through green technological innovation.

Table 6.

Results of the Analysis of the Mediating Effect.

Variable	(1) lnGreen	(2) y	(3) TFP_LP	(4) y	(5) lnC	(6) y
AI	0.092** (0.025)		0.028* (0.015)		0.095** (0.036)
lngreen		0.060*** (0.010)
TFP_LP				0.065*** (0.021)
LnC						0.062*** (0.005)
Constants	−1.020** (0.413)	−1.430*** (0.540)	−3.640*** (0.387)	−1.300* (0.669)	1.224* (0.670)	−1.630*** (0.542)
Control variables	Yes	Yes	Yes	Yes	Yes	Yes
Enterprise fixed effect	Yes	Yes	Yes	Yes	Yes	Yes
Industry 1 year fixed effect	Yes	Yes	Yes	Yes	Yes	Yes
Province fixed effect for 1 year	Yes	Yes	Yes	Yes	Yes	Yes
Sample size	35,069	35,069	30,952	30,952	34,895	34,895
R² value	.722	.577	.931	.606	.293	.582

With reference to previous research experience, the IP method was used to calculate innovation output (TFP_LP) at the enterprise level. The results of the mediation effect test are shown in columns (3) and (4) in Table 6. Among them, the regression coefficient of AI on TFP_LP in column (3) is significantly positive, indicating that scientific and technological innovation driven by artificial intelligence can significantly increase the innovation output of enterprises; the regression coefficient of TFP_LP on manufacturing industry upgrading in column (4) is also significantly positive, indicating that the improvement of corporate innovation output can significantly improve the level of manufacturing industry upgrading. Therefore, the empirical test hypothesis H2a, the enterprise green innovation technology of AI-driven technological innovation is based on the mediating variable of the enterprise’s innovation output to improve the level of industrial upgrading in the manufacturing industry.

This article uses the composite index of information on internal control of enterprises published by the Dibo database to measure the innovation input (lnIC) of enterprises. The use of the Dibo internal control index to measure corporate innovation input (lnIC) aligns with the governance dependence of innovation activities, overcoming the measurement biases of traditional proxy variables. Empirical research has provided significant statistical results and mechanism verification. This design not only inherits the logic of classic literature but also responds to the accounting field’s call for non-financial indicators to measure innovation, thus establishing a reliable micro foundation for research on AI-driven technological innovation (Y. Li et al., 2024). The results of the examination of the innovation input mechanism of enterprises are shown in columns (5) and (6) in Table 6. Among them, the regression coefficient of AI on lnIC in column (5) is significantly positive, indicating that technological innovation driven by artificial intelligence can significantly promote the internal information disclosure of enterprises and increase their innovation investment; the regression coefficient of lnIC on manufacturing industry upgrading in column (6) The regression coefficient is also significantly positive, indicating that the increase in enterprise innovation investment can significantly improve the level of manufacturing industry upgrading. Therefore, the empirical test verifies the hypothesis H2b, that AI-driven enterprise technological innovation improves the level of industrial upgrading in the manufacturing industry based on the mediating variable of enterprise innovation investment.

The essence of the mediation effect is a nonlinear function of coefficient products (δY/δX = a·b + c′). The bootstrap method simulates the joint distribution of the coefficients and accurately captures their variance–covariance structure. This makes it particularly suitable for scenarios common in economics that involve weak instruments or non-normal error distributions (such as the zero-inflated distribution of the patent variable lnGreen in this study).

For the robustness testing of the mediation effects:

We have applied the bootstrap method (B = 5,000 resamples) to verify the significance of the three mechanisms reported in Table 6:

AI → lnGreen → Y: Indirect effect = 0.0055 [95% CI: 0.0021, 0.0093] (significant)

AI → TFP_LP → Y: Indirect effect = 0.0018 [95% CI: 0.0003, 0.0036] (significant)

AI → lnCG → Y: Indirect effect = 0.0059 [95% CI: 0.0031, 0.0089] (significant)

The results show that the 95% confidence intervals for all three paths are strictly greater than zero, indicating that the indirect effects are significant at the 5% level. Green innovation (lnGreen) and capital renewal (lnCG) are the core mechanisms through which AI drives upgrading, contributing 23.9% and 25.7% respectively. The directions of the bootstrap coefficients are fully consistent with technological innovation theory and micro-level evidence on AI penetration.

Heterogeneity Analysis

Staff Academic Level

Employees are the core actors within a firm and play an irreplaceable role in production and operations. However, employees are not completely homogeneous. Differences in employee characteristics may influence the effect of artificial intelligence. To examine the heterogeneous impact of employee educational levels on the role of AI, two variables are constructed: AI_Master (the interaction between AI and the proportion of employees with master’s degrees) and AI_UG (the interaction between AI and the proportion of employees with undergraduate degrees).

The heterogeneity test results are reported in columns (1) and (2) of Table 7. In column (1), the coefficient of Master on manufacturing upgrading (1.315***) is significantly positive; the coefficient of AI_Master (0.495*) is also significantly positive. This indicates that the higher the proportion of employees with master’s degrees, the stronger the enhancement effect of AI-driven technological innovation on manufacturing upgrading.

Table 7.

Results of Heterogeneity Analysis.

Variable	(1) y	(2) y	(3) y	(4) y
AI	0.019 (0.027)	0.045 (0.062)	−0.520** (0.259)	0.103*** (0.035)
Master	1.315*** (0.139)
AI_Master	0.495* (0.292)
UnderGraduate		−0.195* (0.108)
AI_UG		0.346** (0.165)
AI_Ind2			0.545** (0.258)
AI_Ind3			0.625** (0.261)
Soe				0.069* (0.044)
Foreign				0.045 (0.062)
AI_Soe				0.128** (0.061)
AI_foreign				0.055 (0.113)
Constants	4.040*** (0.012)	−1.750*** (0.621)	−2.800*** (0.151)	−1.450** (0.542)
Control variables	Yes	Yes	Yes	Yes
Enterprise fixed effect	Yes	Yes	Yes	Yes
Industry 1 year fixed effect	Yes	Yes	Yes	Yes
Province fixed effect for 1 year	Yes	Yes	Yes	Yes
Sample size	25,837	30,824	36,027	33,875
R² value	.258	.598	.214	.575

In column (2), the coefficient of UnderGraduate is significantly negative (−0.195*), while the coefficient of AI_UG (0.346**) is significantly positive. The statistics for undergraduate employees exhibit a dual nature: when firms deploy AI, undergraduate employees significantly amplify AI’s contribution to upgrading (interaction effect); but in the absence of AI deployment or when AI levels remain constant, a higher undergraduate proportion by itself suppresses upgrading (direct effect). Possible explanations include the following: in traditional, non-AI-driven tasks, undergraduate employees may experience skill mismatch with job requirements; unmet expectations (e.g., salary, job content) may reduce productivity or morale. A high proportion of undergraduate employees also increases labor costs, and without adequate AI application, such investment does not translate into innovation or efficiency gains and may even crowd out resources needed for other upgrading activities (such as equipment renewal or R&D).

Based on the comparison of columns (1) and (2), we also observe that the estimated coefficient of AI_UG is smaller than that of AI_Master. This suggests that the higher the educational level of a firm’s employees, the more effectively AI promotes green innovation, increases innovation output and investment, and thereby more prominently enhances the level of manufacturing upgrading.

Business Ownership

There are currently many types of ownership enterprises in China, and different ownership enterprises differ in many aspects such as production and operation efficiency, pollutant emissions, and social contributions. Therefore, the impact of AI-driven technological innovation on manufacturing upgrades of different ownership enterprises may also vary. Based on this, this article divides the sample enterprises into three categories: private enterprises, state-owned enterprises, and foreign-funded enterprises based on the enterprise ownership code provided by the data of China’s A-share listed companies. Private enterprises are used as the benchmark group for testing, and the state-owned enterprise virtual variable (Soe) and foreign-funded enterprise virtual variable (Foreign), these two virtual variables are intertwined with the core explanatory variable AI Forming the interaction terms AI_Soe and AI_Foreign. The heterogeneity test results are shown in Table 7, column (4): the regression coefficient of AI on manufacturing upgrading in the benchmark group is significantly positive, indicating that technological innovation driven by artificial intelligence has a significant promoting effect on the manufacturing upgrading of private enterprises; AI_Soe The regression coefficient on manufacturing upgrading was significantly positive and higher than that of the benchmark group. It shows that AI-driven technological innovation promotes the upgrading of the manufacturing industry of state-owned enterprises more significantly; the regression coefficient of AI_Foreign on manufacturing upgrading is positive but not significant, indicating that the role of AI-driven technological innovation in upgrading the manufacturing industry of foreign-funded enterprises is different from that of private enterprises. There is no significant difference. In summary, AI-driven technological innovation has the most significant effect on promoting manufacturing upgrading in state-owned enterprises, followed by private enterprises, while the role of foreign-funded enterprises is not significantly different from that of private enterprises.

Further Mechanism Effect Analysis

The nonlinear impact of AI-driven STI on manufacturing upgrading was further examined. The quadratic term (AI_sq) of artificial intelligence-driven technological innovation was included in the “benchmark regression model” and re-estimated, and the results verified hypothesis H3. As shown in column (1) in Table 8: the coefficient of AI is significantly positive, indicating that AI-driven technological innovation has a significant promoting effect on manufacturing upgrading; the coefficient of AI_sq is significantly negative, indicating that AI-driven technological innovation There is a “inverted U-shaped” nonlinear relationship with manufacturing upgrading. At the same time, the Utest test of the “inverted U-shaped” relationship between the two was performed in this thesis, and the results showed that the curve turning point was 2.012, within the independent variable range [0.000, 2.264], and the p-value was .035, indicating that an effective turning point occurred within the existing interval. That is, “artificial intelligence-driven technological innovation can significantly promote manufacturing upgrading in the early stage, but when its level is too high, it will have a inhibitory effect on manufacturing upgrading.”

Table 8.

Results of the Mechanism Analysis.

Variable	(1) y	(2) lnGreen	(3) TFP_LPS	(4) lnCG
AI	0.282*** (0.065)	0.134*** (0.036)	0.123*** (0.034)	0.181*** (0.045)
AI_sq	−0.090** (0.035)
Intergroup coefficient difference test
Constants	−1.189** (0.524)	−1.832*** (0.535)	−0.429 (0.523)	2.147*** (0.691)
Control variables	Yes	Yes	Yes	Yes
Enterprise fixed effect	Yes	Yes	Yes	Yes
Industry 1 year fixed effect	Yes	Yes	Yes	Yes
Province fixed effect for 1 year	Yes	Yes	Yes	Yes
Sample size	36,015	35,069	35,069	35,069
R² value	.597	.643	.601	.563

Conclusions and Revelations

Based on data from manufacturing companies of Chinese A-share listed companies from 2011 to 2024, this paper empirically examines the specific impact of AI-driven technological innovation on manufacturing upgrading and its intrinsic mechanisms. The findings of the study showed that: (1) AI-driven technological innovation has a significant role in promoting manufacturing upgrading. This conclusion remains robust after several robustness tests such as propensity score matching method, replacement of core variables, instrumental variables method, lag and preceding one period and using samples of different stages. (2) Heterogeneity analysis shows that AI-driven technological innovation has a more significant effect on upgrading the manufacturing industries of enterprises and state-owned enterprises with higher academic qualifications. (3) The analysis of the mediation effect shows that AI-driven technological innovation promotes enterprise technological innovation machines to promote manufacturing upgrading through the level of enterprise innovation output and enterprise innovation input. (4) Further mechanism analysis showed that there is a “inverted U-shape” non-linear relationship between AI-driven technological innovation and manufacturing upgrading, and that high-technology-intensive industries, low-technology-intensive industries, and software information technology services in the manufacturing sector are more likely to exceed the threshold.

Based on the conclusions of the above study, this thesis presents the following implications: (1) Optimize AI-driven STI policies to drive high-quality manufacturing development. (2) Implement differentiated support strategies for different enterprises and industry characteristics. (3) Strengthening the mechanisms of action of AI-driven technological innovation. (4) Protection against the risk of threshold effects in AI-driven technological innovation.

Suggestions for Future Research

Although this empirical study using China’s manufacturing sector as a sample is innovative, its methodological framework opens up new pathways for studying the evolution of global manufacturing trends. Future research should move beyond the limitations of a single market and, through systematic exploration of multinational manufacturing data, reveal the deeper patterns and boundary conditions of AI-driven industrial upgrading. Using existing theoretical models and econometric methods, it is recommended to expand the research scope to a global dimension. The heterogeneity of AI applications in global manufacturing will become a key testing ground for validating the transferability of theory. The “inverted U-shape” threshold effect revealed by the Chinese case may evolve differently across different institutional environments and technological ecosystems. Developed countries, relying on mature innovation systems, may overcome the technological overload bottleneck, while emerging economies may face digital divides or reach development ceilings earlier. Building a panel data set that includes the U.S., Germany, Japan, and Southeast Asian countries, and conducting cross-analysis of industry technological intensity and regional institutional quality, can verify the moderating effects of technological absorptive capacity. In the context of the integration of industrial Internet and generative AI, comparing the AI penetration paths of Germany’s “Industry 4.0” and Vietnam’s contract manufacturing companies will redefine the feasible boundaries of technological catch-up.

The globalization of research samples requires dynamic adaptation of technical indicators. Current Chinese text mining methods should be upgraded to a multilingual AI keyword database, focusing on the generational characteristics of technological evolution. For instance, European companies prefer terms like “digital twins” and “predictive maintenance” associated with Industry 4.0, while U.S. companies focus on terms like “deep learning optimization” and “autonomous decision-making systems.” By establishing a cross-language mapping mechanism for technical expressions based on natural language processing technology, we can capture regional technological innovation characteristics while ensuring comparability in measurements.

Research on the adaptation mechanisms of technological policies needs to be further deepened. Institutional factors such as the ethical constraints of the EU AI Act, the technological blockades of U.S. export control policies, and the foreign investment incentive policies of Southeast Asian countries collectively shape the application scenarios and transformation efficiency of AI technology. It is suggested to build a “technology-industry-institution” three-dimensional response matrix: the technological advancement dimension differentiates basic algorithms from cutting-edge applications; the industrial maturity dimension positions traditional manufacturing against emerging industries; and the policy intensity dimension quantifies the strength of subsidies and regulatory density.

Footnotes

Thanks to the help of the school teachers,my research can proceed smoothly. Thanks to the support of school leaders,my research can be carried out smoothly.

Author’s Note

I would like to declare on behalf of my co-authors that the work described was original research that has not been published previously,and not under consideration for publication elsewhere,in whole or in part. All the authors listed have approved the manuscript that is enclosed.

ORCID iD

Fangfang Jin

Funding

The author disclosed receipt of the following financial support for the research,authorship,and/or publication of this article: This paper represents a phased outcome of Henan Province’s latest cycle of key disciplines,specifically Applied Economics,and the Provincial Key Cultivation Discipline of International Trade.

Declaration of Conflicting Interests

The author declared no potential conflicts of interest with respect to the research,authorship,and/or publication of this article.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.

References

Aghion

Bloom

Blundell

Griffith

Howitt

(2005). Competition and innovation: An inverted U relationship. Quarterly Journal of Economics, 120(2), 701–728. https://doi.org/10.1093/qje/120.2.701

Chen

Esperança

J. P.

Wang

(2022). The impact of artificial intelligence on firm performance: An application of the resource-based view to commerce firms. Frontiers in Psychology, 13, Article 884830. https://doi.org/10.3389/fpsyg.2022.884830

Furman

Seamans

(2019). AI and the economy. Innovation Policy and the Economy, 19(1), 161–191. https://doi.org/10.1086/704768

Han

Zhang

Lin

(2025). Artificial intelligence technology, organizational learning capacity and corporate innovation performance: Evidence from Chinese specialized, refined, unique and innovative enterprises. Sustainability, 17(6), Article 2510. https://doi.org/10.3390/su17062510

Chen

Busababodhin

(2025). Research on the impact of artificial intelligence on the manufacturing industry chain resilience. Scientific Reports, 15(1), Article 33432. https://doi.org/10.1038/s41598-025-18956-1

Jani

S. P.

Khan

M. A.

(2025). Applications of AI in smart technologies and manufacturing (Vol. 2). CRC Press.

Jiang

(2025). The connotation and path of the deep integration of artificial intelligence and manufacturing industries. Journal of Innovation and Development, 12(3), 47–51. https://doi.org/10.54097/wyw3y588

Kalidas

(2025). Leveraging artificial intelligence for advanced production scheduling in manufacturing industries. Journal of Computer Science and Technology Studies, 7(11), 37–44. https://doi.org/10.32996/jcsts.2025.7.11.6

Lee

G. H.

Song

Jeon

H. W.

(2025). Conditional entropy-based sequential decision-making for AI adoption in manufacturing: A reinforcement learning approach. Systems, 13(9), 830. https://doi.org/10.3390/systems13090830

10.

Wang

(2024). Artificial intelligence and technological innovation: Evidence from Chinese strategic emerging industries. Sustainability, 16(16), Article 7226. https://doi.org/10.3390/su16167226

11.

Zhao

Jiang

Pan

Liu

(2024). Large language models for manufacturing. arXiv Preprint arXiv:2410.21418. https://doi.org/10.48550/arXiv.2410.21418

12.

Wei

Zhang

(2025). Artificial intelligence, global value chain position and manufacturing firm emissions. Structural Change and Economic Dynamics, 75, 815–827. https://doi.org/10.1016/j.strueco.2025.10.012

13.

Porter

M. E.

Heppelmann

J. E.

(2014). How smart, connected products are transforming competition. Harvard Business Review, 92(11), 64–88.

14.

Rong

Xie

Chen

(2025). Artificial intelligence and high-quality economic development of manufacturing industry: Evidence from China. The Singapore Economic Review. https://doi.org/10.1142/s0217590825470010

15.

Sun

Zhang

Wang

(2025). How the digital transformation of Chinese traditional manufacturing enterprises drives green innovation: A modified education model. Sustainability, 17(4), Article 1473. https://doi.org/10.3390/su17041473

16.

Trajtenberg

(2018). AI as the next general purpose technology: A political economy perspective (NBER Working Paper No. 24245). National Bureau of Economic Research. https://doi.org/10.3386/w24245

17.

S. Y.

(2022). Artificial intelligence technology application and the market value of listed enterprises. Modern Economic Research, 8(11), 77–92. https://doi.org/10.13891/j.cnki.mer.2022.11.008

18.

David

J. M.

Kim

S. H.

(2018). The fourth industrial revolution: Opportunities and challenges. International Journal of Financial Research, 9(2), 90–95. https://doi.org/10.5430/ijfr.v9n2p90

19.

Zhong

Cai

Fang

Xue

Shan

(2025). Does artificial intelligence reduce energy intensity in manufacturing? Evidence from country-level data. Energy Economics, 149, Article 108784. https://doi.org/10.1016/j.eneco.2025.108784

AI-Driven Technological Innovation Empowers Manufacturing Upgrading

Abstract

Keywords

Introduction

Theoretical-Mechanistic Analysis

The Impact of AI-Driven Technological Innovation on Industrial Upgrading in Manufacturing

Analysis of the Mediating Effect of AI-Driven Technological Innovation on Industrial Upgrading in Manufacturing

Analysis of the Impact Mechanism of AI-Driven Technological Innovation on Industrial Upgrading in Manufacturing

Models and Data

Basis for Sample Data Selection

Interpreted Variables

Core Explanatory Variable: Artificial Intelligence Technology (AI)

Enterprise Control Variables

Model Construction

Basic Estimates Results and Analysis

Basic Estimates Results

Endogenous Problems

Robustness Test

Corrected Sample Selection Bias

Replacing Core Variables

Lag Term and Front-End Term

Different Stage Samples Were Used

Empirical Analysis

Analysis of the Mechanism of Action

Heterogeneity Analysis

Staff Academic Level

Business Ownership

Further Mechanism Effect Analysis

Conclusions and Revelations

Suggestions for Future Research

Footnotes

Author’s Note

ORCID iD

Funding

Declaration of Conflicting Interests

Data Availability Statement

References