SenseTime Launches Galaxy Project to Scale Domestic AI Infrastructure

The project arrives as SenseTime pivots to address three shifting market trends: rising enterprise token demand, the maturation of industrial AI, and the increased feasibility of building computing centers on Chinese silicon. To support this, the company has assembled a roster of partners including Cambricon, Huawei Ascend, and Biren Technology, with plans to construct a 'token factory' and five large-scale computing clusters.

Performance Claims and Market Reality

SenseTime’s growth projections hinge on ambitious internal metrics. The company reports processing 2.42 trillion tokens daily, forecasting a 25-fold increase to 10 trillion by late 2026. Furthermore, it claims its hybrid inference technology offers cost-effectiveness 1.25 times that of Nvidia’s H-series hardware. However, these figures lack third-party verification. In the transition from sandbox optimization to real-world production environments—marked by volatile data pipelines and firmware hurdles—such performance gains often face significant degradation.

Beyond infrastructure, the company is diversifying its portfolio with long-term bets on optical and quantum computing. A notable expansion includes a space computing partnership with Guoxing Aerospace, aiming to launch a satellite constellation by 2026 to extend AI services into maritime and disaster-response zones. With new data centers launching in Shanghai, Yancheng, and Hong Kong, plus plans for an overseas base in Saudi Arabia, SenseTime is positioning its domestic stack as a viable alternative for both local and international markets. Whether these high-performance claims translate into reliable production utility remains to be seen once the company reports verified results in coming quarters.