Brian’s latest move at Coinbase is a game-changer—this can’t be overstated. Essentially, it signals that enterprise AI adoption has entered the cost-engineering phase, which will have a structural impact on the monetization path of large models (especially cutting-edge closed-source models). The core logic is shifting from the old approach of 'buy the strongest model + encourage employees to use it more → token usage explodes → high revenue' to 'use engineering methods to drive more token usage but at a lower unit cost.' Previously, large model companies relied on high unit prices × high usage to achieve rapid revenue growth. Now, enterprises are leveraging LLM Gateway + smart routing + caching + default low-cost models to offload a significant portion of routine tasks from frontier models (like Claude, GPT, Gemini, etc.). The result? Total token usage continues to grow exponentially, but the share flowing to the most expensive APIs is decreasing. This directly compresses the monetization efficiency of high-end models. The cost-engineering phase will actually accelerate AI adoption across more enterprises, with some token consumption shifting to lower-cost models. This raises an important question for the market to consider: 'As more enterprises adopt AI and cost-engineering improves, high-end models may lose some market share, but could the absolute value of their commercial share still grow rapidly? After all, the overall market is expanding.' Of course, cost-effective models will see faster growth, partly because they’re starting from a smaller base. This trend is also a big win for CSP providers, as it can significantly improve the profit-cost structure for cloud vendors.