Open-source models + smart routing is a ruthless move: it halves AI costs while enabling exponential growth in usage. Brian Armstrong’s approach is worth copying.

Wu Says learned that Coinbase CEO Brian Armstrong stated that the company has cut its enterprise AI spending by nearly half through infrastructure optimization, while AI token usage continues to grow exponentially. Key cost reduction measures include: adopting open-source models such as GLM 5.2 and Kimi 2.7 as the default options for the internal LLM gateway, intelligently routing tasks to the most cost-effective model, and significantly increasing the cache hit rate of tools like LibreChat from 5% to 60%. Armstrong emphasized that the goal of managing AI costs is not to limit usage, but to make exponential growth sustainable by reducing waste.