Global ICT investment has increasingly relocated carbon: from 2000 to 2020 ICT-related FDI created a denser, reciprocal web of cross-border embodied CO2 flows whose destinations evolved from resource-rich to low-cost to skill-rich regions. Countries with tighter environmental rules and higher corporate taxes are more often net exporters of these embodied emissions, underscoring the need for cross-border carbon accounting and coordinated policy.

Main Finding

The global CO2 transfer network driven by ICT investments (GCNI) became increasingly interconnected from 2000 to 2020. CO2 transfer links show strong reciprocity and triadic closure, and their geographic pattern is tightly associated with foreign direct investment (FDI) motivation factors. Market size and openness, similarity between partners, and evolving vertical FDI motives (from natural-resource seeking → labour-cost seeking → skill-seeking) predict which regions receive embodied CO2 from ICT investment. Regions with stricter environmental regulation and higher corporate tax rates are more likely to be sources (initiators) of these transfer relationships.

Key Points

• Timeframe: 2000–2020.
• Network structure:
  • Increasing interconnectedness and density over time.
  • Significant reciprocity: CO2 flows often go both directions between pairs of regions.
  • Triadic closure: if A↔B and B↔C, A↔C links are more likely (clustered triplets).
• Sender/receiver patterns:
  • Larger markets and more open economies are more likely to be receivers of CO2 transfers tied to ICT investment.
  • Countries/regions with similar attributes are more likely to form CO2-transfer links (homophily).
• Evolution of vertical FDI motives for receivers:
  • Early period: natural-resource abundance mattered more.
  • Middle period: low-cost labour became a stronger predictor.
  • Later period: highly skilled labour became the dominant predictor.
• Regulatory/tax drivers:
  • Regions with stricter environmental regulation and higher corporate taxes tend to initiate (export) ICT-driven CO2 transfers.
• Policy relevance: these FDI-related drivers shape where ICT investment–related emissions are embodied and thus affect attribution of carbon responsibilities.

Data & Methods

• Data:
  • Multi-regional input–output (MRIO) data covering global economic linkages and sectoral flows, applied to ICT investment–related transactions to quantify embodied CO2 transfers across regions for 2000–2020.
  • Institutional/economic covariates capturing FDI motivations (market size, openness, resource endowments, labour cost/skill, environmental regulation stringency, corporate tax rates, similarity measures).
• Analytical methods:
  • MRIO accounting to compute embodied CO2 transfers associated with ICT investments between regions.
  • Complex network analysis to characterize the GCNI’s topology and its evolution (density, reciprocity, clustering/triadic closure).
  • Exponential random graph models (ERGMs) to statistically model tie formation in the GCNI as a function of node-level attributes (FDI motivation variables) and local network structures (reciprocity, triangle closure), and to trace how predictors change over time.

Implications for AI Economics

• Carbon attribution for digital/AI value chains: ICT and AI investments create embodied CO2 flows that cross borders; models of AI’s environmental footprint must account for these transferred emissions rather than attributing emissions only to production or consumption within single jurisdictions.
• Investment disclosure and carbon accounting: investors and firms in the AI/ICT sector should disclose cross-border embodied emissions tied to FDI and project-level investments to improve transparency and enable aligned mitigation strategies.
• Policy design and regulation:
  • Host and home country policies (environmental regulation, tax) influence where emissions are effectively shifted via ICT investment—raising issues of emission leakage and jurisdictional responsibility.
  • International coordination may be needed to prevent carbon relocation driven by tax/regulatory differentials, and to incorporate ICT-driven embodied emissions into climate agreements or border-adjustment mechanisms.
• Location choices for AI/ICT firms: shifting FDI motives (toward skill-seeking) imply that as AI and advanced ICT mature, high-skill regions may increasingly attract ICT investment—and the associated embodied emissions—altering the geography of AI production and its environmental externalities.
• Modeling economic impacts of AI: macro and trade models that evaluate AI adoption, digital infrastructure deployment, or ICT capital flows should incorporate MRIO-style embodied-emission transfers and network formation effects to capture environmental and distributional consequences.
• Research and governance priorities: improved data on investment-driven embodied emissions, standardized disclosure protocols, and network-aware policy tools (targeting clusters or triangular relationships) could better manage ICT/AI-related carbon responsibilities and support sustainable digital development.