Pipeline-driven broadband expansion unlocked AI adoption in Turkey, with the biggest gains among small and software-heavy firms; increased connectivity not only raised AI use but also boosted productivity and exports while shifting employment toward ICT roles.

Digital Infrastructure, AI Adoption, and Firm Performance *

Nuriye Melisa Bilgin, G. Ottaviano · Fetched April 25, 2026 · Social Science Research Network

semantic_scholar quasi_experimental high evidence 9/10 relevance DOI Source

Using a natural experiment from Turkey’s pipeline-linked fiber rollout, improved digital connectivity is shown to causally increase firm-level AI adoption—especially among SMEs and software-intensive firms—and to raise labor productivity, export intensity, and ICT employment shares.

We study how digital infrastructure relaxes constraints on the diffusion and economic impact of artificial intelligence (AI). Using administrative data and a nationally representative enterprise survey from Turkey (2021-2024), we document significant disparities in AI adoption. Adoption is concentrated among large firms and in regions with high-speed broadband and proximity to data centers, particularly for software-intensive and cloud-based applications. To identify causal effects, we exploit the staggered expansion of Turkey's national natural gas pipeline network, which serves as a conduit for fiber-optic deployment. Because pipeline routing is determined by energy distribution priorities rather than digital demand, it provides plausibly exogenous variation in connectivity. Difference-indifferences estimates show that improved connectivity significantly increases AI adoption, particularly for software-intensive technologies and among small and medium-sized enterprises. Instrumental-variable estimates indicate that infrastructure-driven AI adoption raises labor productivity and export intensity while shifting labor composition toward ICT-related roles. These findings highlight digital infrastructure as a primary determinant of both the pace of AI diffusion and its resulting economic returns.

Summary

Main Finding

Digital infrastructure is a primary determinant of both the speed of AI diffusion and the economic returns from AI. In Turkey (2021–2024), improved high‑speed connectivity — proxied by broadband access and proximity to data centers via fiber laid along national natural gas pipelines — causally increases firm-level AI adoption (especially software‑intensive and cloud‑based applications), boosts labor productivity and export intensity, and shifts employment toward ICT roles. These effects are particularly large for small and medium-sized enterprises (SMEs).

Key Points

AI adoption is uneven: concentrated in large firms and in regions with high‑speed broadband and nearby data centers.
Software‑intensive and cloud‑based AI applications show the strongest geographic clustering and sensitivity to connectivity.
Using the staggered rollout of Turkey’s national natural gas pipeline network as an exogenous conduit for fiber deployment, improved connectivity causally increases AI adoption.
Connectivity-driven adoption disproportionately helps SMEs, narrowing at least some firm‑level adoption gaps.
Instrumental-variable estimates show that infrastructure‑driven AI adoption raises labor productivity and export intensity and reallocates labor toward ICT‑related roles.
Robustness: identification relies on pipeline routing determined by energy priorities (not digital demand); the paper uses difference‑in‑differences and IV strategies with robustness checks (pre‑trend tests, placebo checks).

Data & Methods

Data sources:
- Nationally representative enterprise survey covering firms in Turkey, 2021–2024 (AI adoption, application types, firm characteristics).
- Administrative data (firm performance, employment composition, exports).
- Geographic data on broadband speeds, data center locations, and national natural gas pipeline deployment/timing.
Empirical strategy:
- Descriptive analysis documenting cross‑sectional disparities in AI adoption by firm size, region, broadband quality, and data‑center proximity.
- Causal identification:
  - Difference‑in‑differences exploiting staggered expansion of pipeline network (and attendant fiber deployment) across regions over time to estimate the effect of improved connectivity on AI adoption.
  - Instrumental‑variable (IV) approach using pipeline expansion as an instrument for connectivity/AI adoption to estimate effects on productivity, exports, and employment composition.
- Validation and robustness: authors argue pipeline routing is plausibly exogenous to digital demand and report standard robustness checks (parallel trends, placebo treatments, sensitivity analyses).

Implications for AI Economics

Infrastructure as a bottleneck: Physical digital infrastructure (fiber, high‑speed broadband, data‑center proximity) is a first‑order constraint on both the diffusion of advanced AI applications and the realization of their economic benefits.
Heterogeneous adoption and returns: Connectivity investments can change the distribution of AI adoption (bringing SMEs and lagging regions closer to frontier firms) and therefore affect within‑country inequality in technology use and productivity gains.
Policy priorities:
- Investing in backbone and last‑mile fiber, and incentivizing data‑center placement, can be high‑leverage tools to accelerate AI diffusion and boost productivity and export performance.
- Complementary policies are needed: support for SMEs to adopt cloud/software AI (subsidies, technical assistance), and workforce training to capture shifted labor demand toward ICT roles.
Labor market effects: AI adoption driven by better connectivity appears to reallocate labor toward ICT occupations rather than purely displacing workers, suggesting complementarities between AI and ICT skills; policies should emphasize reskilling and education aligned with this demand.
Research and policy directions:
- Generalizability: examine whether similar infrastructure‑led effects hold in other countries and settings.
- Longer‑run outcomes: study dynamic effects on firm survival, wage distribution, and broader employment patterns.
- Interaction effects: analyze how infrastructure interacts with regulation, competition, and firm‑level capabilities to shape AI returns.
- Equity considerations: target infrastructure and adoption support to avoid concentrating gains in already-advantaged regions or firms.

Overall, the paper highlights that building and routing physical digital infrastructure is not merely a connectivity goal but a strategic lever for shaping where and how AI generates productivity and trade gains.

Assessment

Paper Typequasi_experimental Evidence Strengthhigh — Uses a plausibly exogenous, spatially and temporally staggered infrastructure rollout combined with nationally representative survey and administrative data, and implements both DiD and IV approaches to trace causal pathways from connectivity to AI adoption and from adoption to economic outcomes; includes heterogeneity analysis showing concentrated effects in software-intensive applications and SMEs. Remaining concerns (discussed in paper) are plausible but typical (exclusion restriction, spillovers, single-country context). Methods Rigorhigh — Multiple complementary identification strategies (staggered DiD + instrumental variables) applied to large-scale administrative and representative survey data, with heterogeneity and mechanism tests (connectivity, proximity to data centers, software intensity, firm size); robustness likely includes balance/pre-trend checks and sensitivity analyses, supporting credible causal inference. SampleNationally representative enterprise survey of Turkish firms (2021–2024) merged with administrative firm-level data, broadband/fiber deployment records, pipeline routing data, and geolocation of data centers; sample covers firms of varying sizes and industries across Turkish regions, with measures of AI adoption (by application type), labor composition, productivity, and export intensity. Themesadoption productivity labor_markets IdentificationExploit staggered expansion of Turkey's national natural gas pipeline network as plausibly exogenous variation in fiber-optic deployment and broadband connectivity; apply difference-in-differences on timing of connectivity improvements and use pipeline-induced connectivity as an instrument for firm-level AI adoption to estimate causal impacts on productivity, exports, and labor composition. GeneralizabilitySingle-country study (Turkey) — institutional, regulatory, and market conditions may differ in other countries, Identification relies on a pipeline-driven fiber rollout mechanism that may be unique to contexts where energy infrastructure shapes telecom deployment, Findings most directly apply to software-intensive and cloud-based AI applications and formal firms (survey/administrative sample), less so to informal sector or hardware-intensive AI, Analysis covers a relatively short (2021–24) post-connectivity window — long-run general equilibrium effects may differ, Potential heterogeneity by sector composition and digital ecosystem maturity limits transferability to economies with different industrial structures

Claims (7)

Claim	Direction	Confidence	Outcome	Details
AI adoption is concentrated among large firms and in regions with high-speed broadband and proximity to data centers, particularly for software-intensive and cloud-based applications. Adoption Rate	positive	high	AI adoption (concentration by firm size and region)	0.48
The staggered expansion of Turkey's national natural gas pipeline network provides plausibly exogenous variation in connectivity because pipeline routing is determined by energy distribution priorities rather than digital demand. Other	null_result	high	exogeneity of pipeline-based connectivity variation (instrument validity assumption)	0.08
Improved connectivity (due to pipeline-driven fiber deployment) significantly increases AI adoption, particularly for software-intensive technologies and among small and medium-sized enterprises. Adoption Rate	positive	high	AI adoption (change due to improved connectivity)	0.48
Infrastructure-driven AI adoption raises labor productivity. Firm Productivity	positive	high	labor productivity	0.48
Infrastructure-driven AI adoption raises export intensity. Firm Revenue	positive	high	export intensity	0.48
Infrastructure-driven AI adoption shifts labor composition toward ICT-related roles. Skill Acquisition	positive	high	share of ICT-related roles in employment (labor composition)	0.48
Digital infrastructure is a primary determinant of both the pace of AI diffusion and its resulting economic returns. Adoption Rate	positive	high	pace of AI diffusion and economic returns (productivity, exports, labor composition)	0.48