Geolocation and Starlink Geoff Huston, September 2025 --- Challenges of IP Geolocation Unlike telephone numbers, IP addresses have no embedded geographic location. Mapping IP addresses to locations (country, city) requires additional data sources. Representing location can be done via latitude/longitude or country codes. Country definitions and naming conventions are complex; standards like ISO 3166 help provide uniform country codes. Geolocation databases are essential for cybersecurity, intellectual property enforcement, and statistical analysis (e.g., counting Internet users per country). Many IP-to-location databases are private or subscription-based; APNIC Labs relies on Maxmind and ipinfo.io for research. Limitations of Regional Internet Registries (RIR) Data RIRs record the country of the entity assigned an IP resource, not the actual use location. IP addresses can be subdivided and deployed across multiple countries without RIR tracking. So, RIR country codes have limited value for precise geolocation. Measuring ISP Market Share APNIC Labs attempts to estimate ISP user populations within countries using: Country population estimates from United Nations Population Division (mid-year 2024 + growth rate). Internet user penetration data, preferentially from ITU, cross-checked with World Bank and CIA World Factbook. Google ad impression data (~35M/day), mapped via geolocation and BGP routing tables to origin Autonomous Systems (ISPs). Major Assumptions Google's ad placement is uniform (random sample) across all users in a country. Each user uses only one ISP (ignoring multiple ISPs or devices). Due to assumptions, estimates have ~15% uncertainty. Problematic where ad volumes are low (e.g., Russia due to sanctions reducing ad placements). Assumes all users physically reside in the mapped country, ignoring international travelers, ships, aircraft. Starlink and Geolocation Anomalies Starlink operates globally with satellite Internet, complicating geolocation. Example: Yemen shows an implausible 6 million Starlink users (~60% of users), exceeding this country's reasonable estimates. Contributing factors: Heavy maritime Starlink use on ships in the Red Sea near Yemen. Starlink’s IP geolocation assigns addresses to a single country even for ships at sea. Regulatory approval is patchy; some users purchase Starlink service registered in neighboring countries. Similar anomalies identified in 21 countries where Starlink 'market share' exceeds 10%, often involving small island nations or regions with high maritime/roaming activity (e.g., Tuvalu, Kiribati, Cook Islands, Sudan, Myanmar). These inflated Starlink figures distort national ISP market share estimates. Broader Geolocation Questions Raised by Satellite and Mobile Use Defining geolocation for mobile, moving users (ships at sea, aircraft in flight) is difficult. Questions include: Should IP location be the earth station location or legal jurisdiction of ship/aircraft? Territory claims vs. international waters or airspace complicate definitions. Statically assigning location to IPs on moving platforms may be conceptually flawed. APNIC Labs’ Approach To mitigate distortions from Starlink geolocation data, APNIC Labs overrides Starlink geolocation data for the 20 most affected countries. These IPs are marked as "unclassified" to avoid skewing national market share statistics. While imperfect, this approach limits misleading results due to satellite-based roaming. --- Disclaimer Views are personal to the author and do not necessarily represent APNIC. --- About the Author Geoff Huston AM, M.Sc., is Chief Scientist at APNIC, the Regional Internet Registry for the Asia Pacific. Website: www.potaroo.net