Goldman Sachs' Privorotsky points out that model economics are still improving at a very rapid pace, with smaller, cheaper models providing increasingly useful intelligence, leading to a rapid increase in "computing power that can be bought for every dollar," while the number of model competitors continues to rise. The Silicon-Data large model token spending index has fallen by approximately 38% since the end of June.
This trend is very beneficial for AI adoption and the overall market, as lower AI usage costs make it easier for companies to scale deployments; however, the impact is much more complex for companies selling tokens and model access—lower prices make it harder for differences between models to translate into long-term pricing power, and moats are harder to maintain. Earnings expectations are still being revised upwards, but valuation multiples may continue to be pressured downwards.
In other words, AI demand remains strong, but the market's valuation requirements for companies in the industry chain are beginning to rise.
