Goldman Sachs' Privorotsky pointed out that model efficiency continues to improve at a very rapid pace, with smaller and more affordable models offering increasingly useful intelligence, leading to a rapid increase in 'computing power per dollar.' At the same time, the number of model competitors is steadily growing. The Silicon-Data Large Model Token Spending Index has dropped by approximately 38% since the end of June. This trend is highly favorable for AI adoption and the overall market, as lower AI usage costs make it easier for enterprises to scale deployments. However, the impact is much more complex for companies selling tokens and model access—price declines make it harder to translate differences between models into long-term pricing power, and defending competitive moats becomes more challenging. Earnings expectations are still being revised upward, but valuation multiples may continue to face downward pressure. In other words, while AI demand remains strong, the market is beginning to raise its valuation requirements for companies in the industry chain.