According to Amazon’s data, data centers are now essentially structured with 5–6-year long-term agreements and a 3-year payback period. To be honest, this return sounds worse than Bitcoin mining. Mining carries the risk of Bitcoin price volatility; data centers carry the risk of demand fluctuations. Even though they’re called long-term agreements, when major shifts occur, these contracts can’t hold up. Since Anthropic identified coding as a key use case at the end of 2025, AI hasn’t uncovered any significant new applications in over half a year. The long-term outlook is likely bright, but short-term demand growth has been highly volatile. I’ve heard that A’s ARR growth has already slowed. Moreover, coding has already been commoditized by models like Bharat’s DS and Kimi, which have drastically driven down prices. Why haven’t we found new high-value use cases? Because the models are still clumsy and evolving too slowly. For example, why do human doctors still exist? Even though large models possess vast medical knowledge, in practice, their diagnostics still can’t match those of real doctors. Personally, I’ve been disappointed by the pace of intelligent progress in large models over the past six months—it’s clearly fallen short of my expectations. The future is undoubtedly bright.