Telecom infrastructure firms that build sensing, seizing and reconfiguring capabilities perform better — much of that edge runs through stronger governance-compliance frameworks and AI adoption, which together account for 68.3% of variation in performance across 87 firms.

Main Finding

Dynamic capabilities (sensing, seizing, transforming) significantly improve telecommunications infrastructure organizations’ performance both directly (β = 0.284, p < 0.01) and indirectly by enabling stronger GRC (Governance, Risk, Compliance) implementation (indirect effect β ≈ 0.156, p < 0.01; ~19.9% of total effect) and deeper AI adoption (indirect effect β ≈ 0.198, p < 0.01; ~26.6% of total effect). Together the model explains 68.3% of variance in organizational performance (R2 = 0.683).

Key Points

• Theoretical framing: integrates Dynamic Capabilities Theory (Teece) with Resource-Based View; addresses the “black box” by testing GRC and AI adoption as mediators.
• Sample & scope: 87 telecommunications infrastructure organizations across 23 countries, 2019–2023 (multi-source secondary data).
• Main effects:
  • Direct: Dynamic Capabilities → Performance (β = 0.284, p < 0.01).
  • Mediated: DC → GRC → Performance (DC→GRC β = 0.521; GRC→Performance β = 0.243).
  • Mediated: DC → AI Adoption → Performance (DC→AI β = 0.547; AI→Performance β = 0.311).
• Complementarity: GRC and AI work as complementary pathways; both materially account for the capability→performance link.
• Measurement highlights:
  • Dynamic Capabilities index composed of sensing (R&D, partnerships, patents), seizing (CAPEX intensity, deployment speed, acquisitions), and transforming (restructuring, exec turnover, NFV/SDN adoption).
  • GRC maturity index: board independence, specialized committees, disclosure quality.
  • AI adoption maturity: breadth/depth of AI use cases (mean ≈ 8.7 use cases).
  • Performance: balanced scorecard (EBITDA margin 29.4%, ROA 8.7%, network availability 99.89%, revenue/employee $486k).
• Psychometrics & model fit:
  • CFA fit: χ2/df = 1.87, CFI = 0.941, TLI = 0.932, RMSEA = 0.052, SRMR = 0.047.
  • Reliability: Cronbach αs 0.84–0.88; AVE > 0.50.
  • Structural model fit similarly strong; mediation tested with 5,000 bootstrap resamples.
• Limitations noted by design: secondary cross-sectional aggregation across 2019–2023 (averaged), possible selection/endogeneity concerns; sample N = 87 (industry-focused).

Data & Methods

• Data sources: firm annual reports/10-K/20-F, ITU World Telecommunication/ICT Indicators, GSMA Intelligence, regulatory filings, corporate governance documents, vendor/analyst AI reports.
• Sample: 87 firms, 23 countries; region breakdown: NA (31%), Europe (34.5%), Asia-Pacific (25.3%), Latin America (5.7%), MEA (3.4%).
• Time window: 2019–2023 (averaged measures to reduce short-term volatility).
• Variable construction:
  • Dynamic Capabilities: composite index of sensing, seizing, transforming sub-measures (R&D intensity, CAPEX intensity, deployment speed, corporate actions, NFV/SDN adoption, etc.).
  • GRC: board independence ratio, specialized committees, disclosure quality, GRC spending context.
  • AI adoption: maturity index (number & scope of use cases across operations, customer service, security, analytics).
  • Performance: multi-dimensional balanced scorecard (financial, operational, strategic metrics).
• Analysis: Confirmatory Factor Analysis + Structural Equation Modeling (AMOS 26.0); model fit indices reported; mediation tested via bootstrapping (5,000 resamples).
• Key quantitative results: correlations DC–GRC r = 0.487, DC–AI r = 0.521, DC–Performance r = 0.543 (all p < 0.001).

Implications for AI Economics

• Returns to AI are capability-dependent: AI adoption yields stronger performance when firms possess dynamic capabilities (high absorptive/transformative capacity). Models of AI investment returns should endogenize firm capabilities rather than assume uniform productivity gains.
• Complementarity between governance and AI: GRC implementation is not merely a compliance cost but a value-enabler—improved governance reduces risk, lowers adoption frictions, and amplifies the productive impact of AI. Economic models of AI diffusion should include governance quality as a mediator/moderator.
• Heterogeneous diffusion and productivity: the large explained variance (R2 = 0.683) and significant mediation shares imply substantial cross-firm heterogeneity. Aggregate welfare or productivity estimates of AI must account for uneven capability endowments; simple aggregate extrapolations risk over- or under-estimating effects.
• Investment prioritization in capital-intensive sectors: given capital constraints (CAPEX intensity high), reallocating resources toward developing dynamic capabilities and governance frameworks can increase the marginal returns of subsequent AI investments. Cost–benefit frameworks for infrastructure firms should compare spending on AI tools versus spending on capability and governance building.
• Policy and regulation levers: regulators aiming to speed productive AI deployment in critical infrastructure could focus on raising minimum governance standards, disclosure/transparency, and capability-building incentives (training grants, shared R&D consortia), which may be more effective than direct subsidies for AI tools alone.
• Risk-adjusted valuation and financing: financiers and economists valuing infrastructure firms should price AI-related upside conditional on measures of dynamic capability and GRC maturity; risk premia should reflect adoption/compliance readiness, not just planned AI budgets.
• Directions for empirical AI economics:
  • Incorporate firm-level capability indices and governance measures into causal identification strategies for AI impacts (prefer panel data, instrumental variables, or natural experiments).
  • Quantify interaction elasticities (how much capability increases the productivity elasticity of AI).
  • Study public-good aspects of governance and potential coordination failures in capital-intensive sectors (e.g., standard-setting, shared risk pools).
• Labor and market structure: AI-driven productivity gains conditional on capabilities may yield uneven labor impacts across firms and regions—policy responses should consider retraining and redistribution in contexts where only capability-rich firms realize gains.

Suggested next steps for researchers/policymakers: validate causal pathways with panel or quasi-experimental designs, disaggregate AI uses for differential returns, and study how public policy can cost-effectively build the capabilities and governance that unlock AI value in infrastructure sectors.