DeepSeek, Moonshot AI, and Z.ai collectively generate more inference expenditure than OpenAI on the Vercel AI Gateway. This shift in spending occurs as developers route high-volume workloads toward cheaper, open-weight alternatives rather than proprietary frontier models. The transition moves profit margins from the model layer toward the underlying infrastructure and application layers.

Open models now account for 78.4% of token volume on the gateway, nearly double the 40% share recorded in June, while closed models hold 21.6%. DeepSeek V4.1 Flash generates 59.3% of tokens but accounts for only 5.1% of spending, whereas closed models like Anthropic's Claude Opus 4.8 capture 13.7% of spend with only 1.7% of volume. This pricing disparity illustrates the aggressive competition on cost facing closed model companies.