Chinese AI Models Gain Ground in U.S., Capturing 30-46% of Corporate Tokens Since February as Costs Surge

According to CNBC, Chinese-built AI models including DeepSeek and Z.ai's GLM 5.2 are gaining adoption among U.S. companies as they narrow the performance gap with OpenAI and Anthropic while operating at significantly lower costs. Data from OpenRouter, a platform enabling developers to access multiple AI models, shows Chinese models accounted for over 30% of tokens used by U.S. companies each week since February, peaking at 46%, compared to just an 11% average over the prior 12 months. Open-source Chinese models cost 60-90% less than leading American rivals, with companies increasingly routing non-critical tasks to cheaper alternatives that meet performance thresholds.

GLM 5.2, released in June, achieved the fastest adoption rate tracked by Vercel in 2026, with daily token volume growing approximately 27 times in its first full week. Performance has also improved, with Chinese models now estimated to be six to nine months behind top U.S. frontier models while operating near their capability level. AI startup Lindy shifted 100% of its traffic from Anthropic's Claude to DeepSeek, with CEO Flo Crivello stating the switch will save millions within months while maintaining or improving performance on core use cases.