China's national supercomputing centers nudge firms away from financial speculation and back into the real economy: treated firms cut financial asset holdings by roughly 1.1 percentage points and raise fixed investment, R&D and capex intensity, with strongest effects in computing‑intensive industries and spillovers to neighbouring cities.

Main Finding

Deployment of city-level computing power (via China’s National Supercomputing Centers) causally reduces corporate financialization among Chinese A‑share firms (2012–2023). Baseline estimation using a staggered difference‑in‑differences design finds an average reduction in financialization of about 1.1 percentage points. The effect operates through two firm‑level channels — increased data factor capitalization and improved intelligent decision‑making — which raise core business returns and lower the incentives to hold financial assets. Effects are stronger in computing‑intensive industries and firms with low analyst coverage, concentrated in speculative (not precautionary) financial asset holdings, and extend to surrounding cities via spatial spillovers. Computing power deployment also reallocates corporate funds toward the real economy (higher fixed assets, R&D, and capex intensity).

Key Points

• Research question: Does city‑level computing power infrastructure affect firms’ propensity to hold financial assets (corporate financialization), and how?
• Identification: Uses staggered establishment timing of National Supercomputing Centers (NSCs) across Chinese cities as quasi‑natural, multiperiod shocks; employs a staggered DID framework.
• Main quantitative result: NSC deployment reduces corporate financialization by ≈1.1 percentage points (statistically significant).
• Mechanisms:
  • Data factor capitalization: easier/cheaper processing → transform operational data into valuable data assets → higher returns to core business → less incentive to hold financial assets.
  • Intelligent decision‑making efficiency: AI/analytics enabled by computing power improve forecasts, resource allocation, risk warning → reduce precautionary demand and agency‑driven speculation.
• Heterogeneity:
  • Stronger inhibitory effect in computing‑intensive industries.
  • Stronger for firms with low analyst coverage (weaker information environments).
  • Concentrated reduction in speculative financial assets rather than precautionary holdings.
• Extended outcomes: Increases in fixed asset investment, R&D spending, and capital expenditure intensity following computing power deployment.
• Spatial effects: Spatial Durbin model shows inhibitory effects spill over from host cities to neighboring cities.
• Theoretical framing: Endogenous growth / knowledge‑capital formation — computing power lowers the cost of converting data into knowledge, shifting the relative returns toward real (knowledge‑creating) investment.
• Policy takeaway highlighted by authors: Digital infrastructure (computing power) can be a policy tool to counter excessive corporate financialization and redirect resources to the real economy; targeted/tiered subsidies and intercity computing networks can amplify effects.

Data & Methods

• Sample: Chinese A‑share listed firms, 2012–2023.
• Data sources: CSMAR and Wind (corporate financials and firm controls); NSC establishment dates and city locations from policy/administrative records.
• Empirical strategy:
  • Staggered difference‑in‑differences exploiting variation in the timing of NSC openings across cities.
  • Mechanism tests linking NSC deployment to firm outcomes consistent with data capitalization and decision‑making efficiency (proxied by measures of firm profitability, investment behavior, and information environment).
  • Heterogeneity analysis by industry computing intensity and analyst coverage.
  • Extended outcome analysis on real investment (fixed assets, R&D, capex intensity).
  • Spatial Durbin model to assess spillover effects to neighboring cities.
• Identification assumptions: NSC site selection driven by macro science/industrial criteria (Ministry decisions) and plausibly exogenous to individual firms’ financialization trends; parallel trends assumption tested via event‑study style variation (paper states staggered DID for causal identification).

Implications for AI Economics

• Infrastructure matters for firm‑level capital allocation: Access to high‑performance computing is not only an R&D/innovation enabler but also reshapes corporate portfolio decisions, reducing speculative financialization and reallocating funds to productive investment.
• Data as an economic factor: Empirical support that lowering the cost of data processing increases firms’ ability to capitalize data as a productive factor — reinforcing theoretical treatments of data in endogenous growth models and motivating richer micro measures of "data capital" in empirical work.
• Policy design for digitalization: Blanket digital subsidies may be suboptimal; targeted infrastructure investment or tiered subsidies toward computing‑intensive sectors and firms with weak information environments can yield larger real‑economy benefits (less financialization, more R&D and capex).
• Spatial and network considerations: Intercity computing networks and deliberate placement of shared computing infrastructure can produce amplification and spillover effects—important for regional development and coordination policy.
• Research directions: Quantify long‑run dynamics (persistence of reallocation), disentangle precautionary vs speculative motives with richer micro data, assess interplay between computing power deployment and financial market structure/regulation, and evaluate external validity in other institutional contexts (e.g., market openness, different financial systems).