AT&T, a major US telecom operator, stated that open models currently account for approximately 25% of its AI usage, and the company expects this proportion to rise to 70%-80% in the future. In some application scenarios, switching from closed proprietary models to open models has already reduced costs by 80%-90%.
Goldman Sachs indicated that a clear model stratification is emerging in the enterprise market: tasks with less demanding inference requirements can switch to lower-cost open models; while the most complex tasks continue to use more powerful proprietary models. As domestically developed open models gradually become alternatives to US models, the cost of AI tokens tracked by Goldman Sachs has also decreased by 49% since May. The increasing cost of model usage is beneficial for enterprises to expand their AI usage, but it will also put greater pressure on the unit price and profit margins of basic model vendors. In the next stage of AI commercialization competition, cost efficiency, in addition to model capabilities, is rapidly becoming more important.
