At WAIC 2026, SenseTime officially launched the 'Galaxy Plan' to accelerate the large-scale commercialization of domestic AI infrastructure. By partnering with nearly 20 industry players, the company aims to construct five 10,000-GPU domestic compute clusters, bridging the gap from pilot projects to industrial-scale production.
To overcome scalability hurdles, SenseTime's AI infrastructure platform has optimized performance through three pillars: cost-efficiency, adaptability, and power efficiency. By utilizing heterogeneous hybrid inference technology, the company has boosted MFU (Model FLOPs Utilization) for domestic chips, achieving a price-performance ratio 1.25 times that of NVIDIA's H-series. Reports indicate a 2.5x increase in Token output at the same cost level.
Technically, the introduction of the compute-electricity coordination Agent and the new TPW (Tokens Per Watt) metric has improved token output per unit of electricity by 80%. Furthermore, #SenseTime is pioneering space-based computing via a strategic partnership with Star.vision to deploy a satellite constellation, ensuring global AI accessibility.
[AgentUpdate Depth Analysis] The 'Galaxy Plan' represents a critical strategic pivot for the Chinese AI Agent ecosystem. As Agentic workflows evolve from simple task completion to complex multi-step reasoning, they demand superior inference infrastructure. By unifying diverse domestic silicon (e.g., Cambricon, Ascend) through heterogeneous inference software, SenseTime is successfully mitigating the historically fragmented hardware landscape. This modular, heterogeneous approach is highly advantageous for AI Agents requiring scalable, specialized inference deployments. Unlike the monolithic architectural path often taken by Western giants, SenseTime’s strategy prioritizes localized ecosystem flexibility, which is essential for sustaining long-term commercial viability in enterprise AI environments. Ultimately, this initiative transitions domestic compute from mere 'usability' to 'high-efficiency profitability,' setting a necessary foundation for the next wave of autonomous agents that require sustained, cost-effective infrastructure support to flourish on a global scale.
