This article first appeared on GuruFocus .
Silicon Valley's artificial intelligence race appears to be moving beyond pure model capability, as investors may increasingly watch whether cheaper Chinese models can gain ground through developer adoption. Hangzhou DeepSeek Artificial Intelligence, the Hangzhou-based AI model developer better known as DeepSeek, Alibaba Group Holding ( NYSE:BABA ), a Chinese company with AI models referenced by startups, and Moonshot AI, an AI company, are drawing more attention from American startups seeking lower-cost alternatives as AI coding and heavy token usage become more expensive.
DeepSeek's latest flagship model costs about $0.87 per million output tokens, compared with roughly $30 for OpenAI, an AI company developing frontier models, and $25 for Anthropic, the AI company behind Claude. Vercel, a cloud platform used to deploy websites and apps, found DeepSeek's share of AI traffic rose to 17% from under 1% in May, while OpenRouter, a platform that tracks AI model usage, said DeepSeek usage doubled between January and June 2026 and became the most popular model among its users.
The shift may matter because leading AI labs are still subsidizing heavy usage while moving toward potential IPOs at valuations of around $1 trillion each. Lindy, a San Francisco-based AI assistant startup, said it switched to DeepSeek from Anthropic's Claude after AI costs exceeded personnel costs, while Airbnb, a home-rental platform, and Anysphere, the company behind coding platform Cursor that is being acquired by SpaceX ( NASDAQ:SPCX ), have added Chinese models to their AI mix rather than fully replacing U.S. providers. Larger corporate customers do not appear to be moving away from U.S. labs yet, but rising usage among startups and developers suggests Chinese AI companies could be gaining influence through adoption, not just raw capability.
