To be the Global Gateways: An Empirical Study on the Infrastructuralization of Large Language Models from the Access Perspective

Abstract

This paper empirically examines global access to large language models (LLMs) in order to assess their ongoing infrastructuralization and the challenges they face in moving toward general-purpose artificial intelligence. Drawing on the concept of connectivity in communication and media studies and adopting the “gateway” as a core infrastructural lens, the study employs computer network experiments to measure global access conditions to LLMs. Using network-level indicators—including packet loss, latency, and jitter—it evaluates accessibility, speed, and stability, as well as the inequalities embedded in LLM connectivity. Based on nearly 200,000 network probes conducted across 62 global network nodes, the study finds that city nodes in the Global South generally experience disadvantages in accessibility compared to those in the Global North, particularly when accessing LLMs developed in Western countries. The results further show that some LLMs outperform earlier information infrastructures, such as search engines and databases, in terms of accessibility and speed, with Global South nodes accessing LLMs notably faster than databases. However, LLMs have not yet demonstrated clear advantages in connection stability over previous-generation information infrastructures. While the emergence of LLMs signals an early stage of global information exchange and human-machine interaction, persistent geopolitical constraints on connectivity remain unresolved. Addressing issues of access, usability, stability, and societal value is therefore crucial if LLMs are to realize their infrastructural potential and evolve into genuinely global gateways.