Coinbase CEO Brian Armstrong announced that the company has successfully reduced its AI spending by nearly half through strategic infrastructure optimizations. These optimizations include improved default settings, routing, and caching mechanisms, which have been implemented amid a surge in token usage. Key measures include adopting open-source and cost-effective models like GLM 5.2 and Kimi 2.7, which have proven sufficient for 91% of employees who never reached usage limits. Additionally, intelligent routing now automatically directs tasks to the most appropriate models, optimizing for cost and efficiency. The company also enhanced its caching strategy, increasing the cache hit rate significantly, exemplified by LibreChat's improvement from 5% to 60%. These efforts are part of a broader initiative to streamline operations and increase visibility into AI spending.