Chinese Open-Weight AI Models Capture 29% Market Share on Cost

Chinese AI models gain ground as US pricing pressures mount

Enterprises are rapidly migrating from premium American AI services to Chinese open-weight models that deliver comparable performance at a fraction of the cost, according to new data from cloud platform provider Vercel.

Open-weight models—which companies can download and run on their own infrastructure—now represent 29 percent of token volume on Vercel's AI Gateway platform, nearly triple their share from April. The shift reflects growing price sensitivity as organizations reassess AI spending against tightening budgets.

Zhipu's GLM-5.2 model has seen particularly dramatic adoption. Since mid-June, daily token volume for the model surged 50-fold on the San Francisco-based platform, Vercel reported Tuesday. The model operates at roughly one-fifth the cost of Anthropic's Claude Opus 4.8, making it an attractive alternative for cost-conscious developers.

DeepSeek's V4 Flash has emerged as the single highest-volume model on the gateway, capturing more than 20 percent of platform traffic Wednesday—up from approximately 15 percent one month prior. The streamlined version of DeepSeek's flagship V4 Pro demonstrates how Chinese AI labs are successfully targeting the performance-per-dollar segment that matters most to production workloads.

Why it matters

This migration signals a fundamental shift in enterprise AI economics. For the first time, businesses can access near-frontier AI capabilities without paying token-by-token subscription fees to American providers. The performance gap that once justified premium pricing has narrowed to the point where cost considerations now dominate procurement decisions. Chinese labs have effectively commoditized advanced language model capabilities, forcing US providers to compete on price rather than capability alone.

The economics driving the switch

The distinction between closed-source and open-weight models centers on deployment flexibility and cost structure. Proprietary services from OpenAI and Anthropic require ongoing cloud subscriptions with per-token charges that scale with usage. Open-weight models eliminate recurring fees—companies download the code once and run inference on their own hardware.

Until recently, enterprises accepted premium pricing because open alternatives lagged significantly in capability. That calculus has changed as Chinese labs have closed the performance gap while maintaining their cost advantage. Organizations can now achieve production-grade results without the margin pressure of per-token billing.

The trend reflects broader dynamics in the AI industry, where model capability improvements have outpaced price adjustments from established providers. As Chinese competitors demonstrate that advanced performance need not command premium pricing, pressure will intensify on US companies to justify their cost structures or risk further market share erosion.

The data was first reported by the South China Morning Post, which noted the traffic patterns on Vercel's platform through Wednesday.