Chinese AI models gain traction in US as costs drop and performance improves

US companies are increasingly turning to Chinese AI models to cut costs. Since February, weekly token usage of Chinese models by US firms has exceeded 30%, peaking at 46%, up from an average of 11% in the previous 12 months. Models from DeepSeek and Zhipu AI now rival top US products in performance but cost 60% to 90% less.

Several US businesses have already made the switch. AI startup Lindy moved all its traffic from Anthropic's Claude to DeepSeek, with its CEO saying costs "plummeted," saving millions in months. On Vercel, Zhipu's GLM 5.2 saw a 27x surge in daily token usage and an 80x increase in customers in its first full week.

Performance-wise, GLM 5.2 scored just 1% below Anthropic's Opus 4.8 in a key benchmark, at a fifth of the cost. A Brookings scholar noted Chinese models trail US leaders by 6-9 months, but their cost advantage is reshaping enterprise choices. Open-source models, which allow customization, are becoming a go-to for firms managing AI budgets.