China’s lower-cost artificial intelligence models are gaining traction with startups, creating a new adoption threat for Anthropic and OpenAI.
Key Points:
- Startups are testing Chinese AI models as token costs rise for coding and workflow automation.
- DeepSeek’s model traffic share on Vercel reportedly rose to 17% from under 1% in May.
- US labs still lead on capability, but cheaper models could shape developer habits.
China AI
Parmy Olson, a Bloomberg Opinion columnist, wrote that the AI race may not be decided only by the most powerful model, because adoption by developers can create its own strategic lead.
She pointed to growing use of cheaper models from Hangzhou DeepSeek Artificial Intelligence Co., better known as DeepSeek, as well as Alibaba Group Holding Ltd. and Moonshot AI Pte Ltd., as startups try to control AI spending.
One London startup founder using Anthropic’s $200-a-month Claude Max plan told Olson he feared the company could raise frontier model prices by five times or more.
That concern reflects a wider cost problem for teams that rely on AI coding and heavy token use, since engineers can burn through thousands of dollars in compute each week.
Anthropic and OpenAI have reasons to charge more, as both companies subsidize heavy usage while moving toward pay-as-you-go pricing and possible IPOs at valuations around $1 trillion each.
Also Read: Palantir Stock Loses Steam As Democrats Eye Government Contract Exposure
DeepSeek Adoption
DeepSeek’s latest flagship model costs about $0.87 per million output tokens, compared with about $30 for OpenAI and $25 for Anthropic, according to Olson’s figures.
That price gap matters because many routine business tasks do not require the most advanced model available.
Smaller systems can sort email, draft simple responses, summarize documents, answer customer-service questions and clean data, which makes lower-cost Chinese models useful for everyday software work.
Vercel Inc., a cloud platform that tracks model use among programmers, found DeepSeek’s share of AI traffic rose to 17% from under 1% in May.
OpenRouter Inc. also reported that DeepSeek usage doubled between January and June 2026, while Chinese open-source models from Xiaomi Corporation, MiniMax Group Inc. and Tencent Holdings Limited gained token share at the expense of Google and OpenAI.
The shift is still limited, since OpenRouter represents about 3% of global AI traffic and skews toward startups and independent developers rather than large enterprise buyers.
Still, Lindy, a San Francisco AI assistant startup, said it moved to DeepSeek from Anthropic’s Claude after AI costs for its 25-person team exceeded personnel costs, with CEO Flo Crivello saying the change saved “millions.”
Larger companies, including Airbnb Inc. and Anysphere Inc., the company behind Cursor, have added Chinese models to their mix rather than fully replacing US providers. The retrospective lesson is that US labs still build the strongest frontier models, but software markets often reward tools that developers adopt early, use often and build around for years.
Read Next: Bitget rToken Rockets To $114M In Just Five Weeks After June Launch