Data Localization as Tax Nexus: Server PEs for AI Providers in Asia

Mandatory data localization in Asia can compel foreign artificial intelligence (AI) providers to place or control infrastructure in a given country, raising the question of whether this creates a server permanent establishment (PE) under article 5 of the OECD and UN Models. Drawing on a doctrinal study of Vietnam, Indonesia, Myanmar, Cambodia and Lao PDR, this article shows how localization and presence requirements can satisfy the fixed place and “at disposal” tests in the case that providers deploy dedicated inference environments that they control, even without personnel on site. It proposes an operational “control and co-location” test to distinguish compliant hosting from PE-relevant architectures. It then outlines a profit attribution approach under the Authorized OECD Approach, adapting DEMPE analysis to algorithmic functions: routine returns for local hosting and orchestration, residual profits for offshore model development and data curation. Finally, it identifies double taxation risks and recommends targeted safe harbours and stronger dispute resolution mechanisms.