Supercomputing Centers Reduce Corporate Financialization in China

Researchers from China have demonstrated that the establishment of national supercomputing centers significantly reduces corporate financialization among listed companies. Published in Financial Innovation, the study utilizes a staggered difference-in-differences model to analyze Chinese A-share listed companies from 2012 to 2023. By exploiting the staggered establishment of National Supercomputing Centers (NSCs) as quasi-natural experiments, the authors identify a causal link between computing power infrastructure and corporate asset allocation. The baseline results indicate that the deployment of computing power infrastructure reduces corporate financialization levels by approximately 1.1 percentage points. This reduction represents about 17.5% of the sample mean financialization level, suggesting an economically meaningful shift in corporate resource allocation away from financial assets.

Corporate financialization refers to the process by which nonfinancial enterprises shift funds from traditional industrial investment to financial market activities. This trend is widely recognized as a central issue affecting resource allocation efficiency in the real economy. Nonfinancial enterprises worldwide have increasingly diverted resources from productive investment, which can undermine long-term competitiveness and innovation capacity. The study addresses a gap in the literature by examining the role of technological infrastructure in shaping this trend. Previous research has focused on macroeconomic drivers such as policy uncertainty and monetary conditions, as well as micro-level factors like ownership concentration and managerial incentives. However, existing studies had not incorporated computing power infrastructure into the analytical framework of corporate financialization determinants.

The theoretical framework of the study is grounded in endogenous growth theory, which posits that sustained economic growth is driven by deliberate investment in knowledge creation. Computing power infrastructure directly affects the reallocation of resources between real and financial investment by reducing the marginal cost of knowledge production. Data centers and supercomputing facilities provide the technical foundation for large-scale data processing, storage, and analytics. This enables enterprises to convert raw operational data into commercially valuable knowledge assets. In the endogenous growth framework, data are recognized as a new factor of production that accelerates knowledge accumulation. The computing power infrastructure amplifies the wedge between the productivity of real and financial investment by reducing the fixed costs of data utilization, thereby raising the opportunity cost of holding financial assets.

The research design employs a staggered difference-in-differences model to identify the causal effect of computing power deployment on corporate financialization. The sample consists of 19,832 firm-year observations from Chinese A-share listed companies, excluding financial and insurance companies, ST and *ST companies, and observations with missing key variables. The core independent variable is a dummy variable that equals 1 for a city in a year and all subsequent years once an NSC has been established. The dependent variable, corporate financialization, is measured as the ratio of financial assets to total assets. The model includes firm and year fixed effects, as well as lagged control variables such as firm size, leverage ratio, profitability, firm growth, cash flow level, and ownership concentration.

Mechanism analysis reveals two primary transmission channels through which computing power deployment inhibits financialization. The first channel is the enhancement of data factor capitalization capability. Computing power infrastructure provides the technical conditions for large-scale data storage, processing, and analysis, reducing the barriers and marginal costs of data factor development. The study finds that NSC establishment significantly increases data-related intangible assets and digital technology patent authorizations. This capability allows firms to transform data resources into strategic assets that strengthen core business returns and competitive advantage. The second channel is the improvement of intelligent decision-making efficiency. Computing power supports the deployment of complex algorithm models and AI technology, which enhance internal information processing and shorten decision cycles. This leads to reduced investment deviations and lower management expense intensity, thereby optimizing resource allocation and suppressing managerial opportunistic behavior.

Heterogeneity analysis shows that the inhibitory effect of computing power deployment is more pronounced in specific contexts. The effect is stronger in computing-intensive industries, such as information transmission, software and IT services, and scientific research and technical services. These industries depend heavily on large-scale data processing and real-time intelligent analysis, enabling them to directly embed supercomputing resources into core production processes. The study also finds that the effect is concentrated in speculative rather than precautionary financial assets. Computing power infrastructure primarily weakens the arbitrage motive for financialization by improving core business profitability, while leaving reasonable corporate liquidity arrangements largely unaffected. Furthermore, the inhibitory effect is more significant for firms with low analyst coverage, suggesting that computing power infrastructure delivers greater incremental improvements in information transparency for firms with weaker external governance.

Extended analysis confirms that the funds released from reduced financialization flow back to the real economy. The study finds that computing power deployment promotes real investment reallocation, with significant increases in fixed asset investment intensity, R&D investment intensity, and capital expenditure intensity. Specifically, fixed asset investment intensity increases by an average of 0.8 percentage points, R&D investment intensity by 0.5 percentage points, and capital expenditure intensity by 0.7 percentage points after NSC establishment. These findings validate the capital reallocation effect, demonstrating that computing power deployment guides enterprises to shift funds from the financial sector to fixed asset purchases and R&D activities. This result aligns with resource allocation theory, which predicts that when financialization is suppressed, funds should reallocate toward real capital expenditures.

Spatial Durbin model estimation further reveals that the inhibitory effect of computing power deployment extends beyond host city boundaries to surrounding cities. The direct effect of local NSC establishment on local financialization is estimated at -0.012, while the indirect effect from neighboring cities’ NSC establishment is -0.007. The total effect of -0.019 indicates that the policy effects of supercomputing centers radiate beyond host cities to surrounding regions. This spatial spillover occurs through technology diffusion, network connectivity, and competitive demonstration channels. Modern data centers achieve interconnection with surrounding cities through high-speed fiber optic networks, enabling cross-regional access to computing power resources. Enterprises in neighboring cities may also proactively adjust resource allocation strategies in response to competitive pressure from host cities.

The study addresses potential endogeneity concerns using instrumental variable estimation. The authors select two instruments: per capita postal and telecommunications business volume in 1984 and urban terrain ruggedness. Historical communication infrastructure reflects path dependence in urban information technology development, while terrain ruggedness affects construction costs for large-scale data centers. The Kleibergen-Paap rk Wald F statistic is 48.637, far exceeding the Stock-Yogo critical value, ruling out weak identification. The 2SLS coefficient of Comput is -0.024, and the LIML estimate is -0.025, both significant at the 1% level. The close correspondence between the two estimates confirms that the results are not sensitive to the choice of estimation method and that weak instrument bias is not a concern. Robustness tests, including parallel trends tests, placebo tests, and heterogeneity-robust estimators, further confirm the credibility of the baseline findings.

The findings carry several implications for policy and practice. For corporate managers, the heterogeneity results show that returns to computing power utilization depend on existing digital readiness. Managers in less digitally mature industries should prioritize foundational investments in data governance, IT workforce development, and process automation before expecting meaningful shifts in asset allocation. For investors and financial analysts, proximity to computing power infrastructure can serve as a signal of future changes in corporate asset allocation. Firms transitioning from financial income dependence to core business strengthening may experience short-term earnings volatility but improved long-term sustainability. For industry associations, the spatial spillover evidence reveals that firms in neighboring cities also benefit from NSC establishment. Industry associations can facilitate this process by organizing cross-regional computing power sharing platforms and coordinating the joint procurement of cloud computing services.

The study has several limitations that point to specific avenues for future investigation. First, the treatment variable is a binary indicator based on NSC establishment, which effectively identifies causal effects but cannot capture continuous variation in computing power supply. Supercomputing centers differ substantially in processing capacity, service scope, and utilization rates. A binary measure treats all NSCs as equivalent shocks and may underestimate the true impact of computing power deployment. Second, the study focuses exclusively on supercomputing centers as the source of computing power infrastructure. The landscape of computing infrastructure in China has diversified substantially in recent years, with the expansion of intelligent computing centers, edge computing nodes, and the “East Data West Computing” national project. Future research could compare the financialization effects across these different computing power modalities.

Third, the empirical setting is confined to China, where the institutional environment features strong state coordination in infrastructure deployment and a financial system characterized by bank dominance and capital account restrictions. These features may amplify the inhibitory effect of computing power on financialization by limiting the range of financial investment alternatives available to firms. In market-oriented economies with deeper capital markets and fewer investment restrictions, the same computing power deployment could yield weaker financialization effects because the opportunity cost structure differs. Future cross-country comparative studies could test the boundary conditions of the findings. Suitable settings include South Korea, which has a comparable supercomputing infrastructure program, and the European Union, where the EuroHPC Joint Undertaking coordinates cross-national computing power deployment under different institutional arrangements.

Subject of Research: The impact of computing power infrastructure on corporate financialization in China

Article Title: Computing power infrastructure and corporate financialization: evidence from China’s supercomputing centers

Article References: Zhang, J., Wang, M., & Xia, X. (2026). Computing power infrastructure and corporate financialization: evidence from China’s supercomputing centers. Financial Innovation, 12(1), Article 131. https://doi.org/10.1186/s40854-026-00946-5

Image Credits: AI Generated

DOI: 10.1186/s40854-026-00946-5

Keywords: computing power infrastructure, corporate financialization, supercomputing centers, difference-in-differences, real investment, data factor capitalization, intelligent decision-making, spatial spillover, China A-share, digital economy, Computing, power