Chinese AI firm Z.ai reported a 400% jump in first-half 2026 revenue to 953.89 million yuan (US$142 million), missing analyst estimates by 30% amid an intense domestic price war. While Z.ai's losses narrowed to 2.07 billion yuan, rising R&D costs highlight the high price of competing with Alibaba and Moonshot AI. Meanwhile, low-cost Chinese models like GLM-5.2 are forcing US rivals OpenAI and Anthropic to cut prices, even as Washington and Beijing threaten tighter regulatory restrictions on cross-border AI model access.
Z.ai first-half financial results
- ▪Z.ai's total loss for the six months ended June 30, 2026, narrowed 12.1 percent to 2.07 billion yuan, while its adjusted net loss increased 12.1 percent to 1.96 billion yuan.
- ▪Z.ai announced that its annual recurring revenue reached US$1.6 billion by the end of August 2026, compared to rival MiniMax's annual recurring revenue of US$800 million in August 2026.
- ▪Z.ai's research and development expenses rose 33.6 percent to 2.13 billion yuan for the six months ended June 30, 2026, driven by aggressive investments in computing power and base model performance.
- ▪Z.ai reported a 400 percent increase in revenue to 953.89 million yuan (US$142 million) for the six months ended June 30, 2026, which fell approximately 30 percent short of average analyst estimates.
Chinese AI model pricing advantage
- ▪Zhipu AI raised its prices several times in 2026 due to rising computing service fees, and Moonshot AI temporarily suspended new subscriptions when demand for its Kimi K3 model exceeded computing capacity.
- ▪Zhipu AI's GLM-5.2 model is priced at US$4.40 per million output tokens, compared with US$10 per million output tokens for Anthropic's Claude Sonnet 5.
- ▪OpenAI reduced the prices of its GPT-5.6 Luna and Terra models by 80 percent and 20 percent, respectively, at the end of July 2026, while Anthropic permanently retained its introductory promotional price for Sonnet 5.
Cross-border AI model usage
- ▪According to OpenRouter statistics as of August 2026, eight of the 10 large language models with the highest token processing volumes over the preceding month were from China, while two were from the US.
- ▪US models dominated spending across most of nearly 30 task categories tracked by OpenRouter, with Chinese models accounting for the largest share of spending in only seven categories.
US-China AI regulatory restrictions
- ▪China's Ministry of Industry and Information Technology warned in July 2026 that Anthropic's Claude coding software contained built-in monitoring that posed a security backdoor risk.
- ▪The US government has banned the use of DeepSeek AI models on government devices at some agencies, citing security concerns.
- ▪US Treasury Secretary Scott Bessent stated in July 2026 that the US government would investigate whether Chinese open-weight models were involved in intellectual property theft, which could lead to financial sanctions or Entity List additions.
AI model performance comparisons
- ▪Moonshot AI's Kimi K3 model features 2.8 trillion parameters, indicating that Chinese AI laboratories are actively pursuing frontier capabilities alongside cheap application diffusion.
- ▪Zhipu AI's GLM-5.2, Moonshot AI's Kimi K3, and Alibaba's Qwen3.8-Max models have approached the performance of Anthropic's Claude model on multiple tasks at lower prices.
Story comments
Loading comments…