U.S. companies are increasingly routing workloads to Chinese-built AI models as steep pricing from OpenAI and Anthropic pushes developers to cheaper alternatives. Platforms such as OpenRouter report Chinese models accounting for more than 30% of U.S. token usage weekly since early February, peaking at 46%, up from an 11% average over the prior year. Startups like Lindy say moving traffic from Anthropic’s Claude to DeepSeek cut costs by millions while maintaining or improving core use cases. Zhipu AI’s GLM 5.2 has seen rapid uptake on Vercel, with daily tokens up 27x and customer counts up 80x in its first full week post-launch, buoyed by pricing that can be a fraction of top-tier U.S. models. Analysts estimate leading Chinese systems now trail American frontier models by roughly six to nine months, narrowing the performance gap. The shift comes amid heightened U.S. scrutiny: OpenAI limited new model rollouts at the government’s request, and export controls on some Anthropic models were recently lifted. With developers “routing to the cheapest model that’s good enough,” companies face a strategic trade-off between price volatility in proprietary U.S. systems and growing reliance on competitive Chinese alternatives.
Related articles:
— Introducing Meta Llama 3
— AI Risk Management Framework
— Amazon Bedrock Pricing
