Leaky Rents at the Frontier

This paper examines how China's governance system shapes competition among AI labs. Firms with stronger political ties before breakthrough often receive better access to state-mediated inputs, yet those ties do not consistently translate into stronger efficiency-adjusted frontier performance. Any regulatory edge also proves limited: advantages fade, while outsiders that succeed are often brought into the system afterward.

The analysis compares leading Chinese AI labs (Baidu, SenseTime, iFlytek, Alibaba/Qwen, ByteDance/Doubao, DeepSeek, and others) along a gradient of pre-breakthrough political embeddedness, from highly embedded national-team members to self-funded outsiders like DeepSeek. Frontier performance is measured through benchmark competitiveness, speed of frontier catch-up, architectural contribution, and developer adoption.

Key findings: highly embedded firms benefit from access but less embedded firms have frequently delivered stronger efficiency-adjusted frontier advances. The system exerts its strongest political influence not by shutting outsiders out in advance, but by absorbing them once they have succeeded. These temporary state-conferred advantages are termed "leaky rents," benefits that matter initially but erode under competitive pressure.

Written for Econ 15 (Institutions and Economic Growth) at Dartmouth College.