AI算力版图持续向轨延伸，太空算力产业化将迎关键里程碑

Smart Things Author | Liu Yu Editor | Chen Junda At the 2026 World Artificial Intelligence Conference (WAIC), the evolution of large model architecture and computing infrastructure has once again become the focus of the industry. As the scale of large model parameters continues to expand and agent applications continue to penetrate into the industry, the AI ​​industry's demand for global deployment, coverage in remote areas, and computing power in weak or no-network environments continues to increase. For a long time, AI computing power has mainly relied on ground data centers to provide services. However, in scenarios such as high latitudes, deep seas, deserts, and oceans, traditional ground infrastructure is unable to meet some real-time computing needs due to network coverage, construction costs, and transmission delays. At the same time, in some cross-regional business scenarios, how to improve the efficiency of computing resource scheduling has also become a direction that the industry continues to explore. Against this background, computing infrastructure is gradually evolving from a single ground-based intelligent computing center to a space-ground collaborative architecture. By building a collaborative computing system of "satellite nodes + ground intelligent computing centers", deploying part of the computing power to orbit and uniformly scheduling it with ground computing resources has become a path to expand the boundaries of global AI computing services. At this WAIC, China's space AI leader Chengdu NationStar Aerospace Technology Co., Ltd. (hereinafter referred to as "Guoxing Aerospace") and Shanghai SenseTime Technology Development Co., Ltd. (hereinafter referred to as "SenseTime") announced the joint construction of the "SenseTime Computing Constellation". The two parties plan to follow the development path of "single satellite verification - platform construction - constellation networking" to promote the construction of space-ground integrated AI infrastructure, and propose to form a space intelligent computing network composed of thousands of computing power satellites by 2030, with a total computing power of over 10,000 pixels. This has also made how space computing power forms a large-scale commercial closed loop once again become the focus of the industry. 1. Planning for ultra-large-scale networking, space computing power has entered a new stage of industrialization. In the past few years, space computing has remained more in the concept exploration and technology verification stage. As the demand for AI computing power continues to grow, and technologies such as satellite manufacturing, on-board computing platforms, and inter-satellite communications continue to mature, the global space computing power field has formed a multi-party competition pattern. SpaceX has mentioned