China's AI industry shifts toward hardware to overcome compute gaps and user retention issues

Chinese large language model (LLM) developers are facing a dual challenge: a significant gap in training computing power compared to US counterparts and low user loyalty in the software market. While the performance gap between leading Chinese and US models has narrowed from over 30% in 2023 to approximately 2.7% by early 2026, Chinese firms remain constrained by the massive scale of compute required for frontier models. US models like GPT-5 and Grok 4 utilize an order of magnitude more floating-point operations than leading Chinese models.

To combat high user churn rates—where 30-day retention for major AI applications sits at only 12.8%—Chinese AI companies are aggressively expanding into consumer hardware. By developing AI-native devices such as smartphones, smart glasses, and earbuds, firms aim to transform AI from a transient software service into a permanent hardware-integrated ecosystem. Recent moves include StepX releasing the STEPX Neo smartphone, Alibaba launching Qwen AI glasses and earphones, and ByteDance collaborating on AI-integrated mobile devices. This shift seeks to create higher switching costs for users and establish proprietary ecosystems centered around personal AI agents.