SenseTime is deploying five AI clusters: betting on national chips and the ecosystem
Chinese AI giant SenseTime has officially launched a large-scale project called "Galaxy." This initiative brings together nearly two dozen leading chip and infrastructure equipment manufacturers. The project plans to build five computing clusters, each equipped with 10,000 Chinese AI accelerators.
In addition to the clusters themselves, the project includes the construction of a token processing center and a support program for 200 AI startups. This is not just about expanding computing power — it is a strategic move to build a fully sovereign technology chain independent of Western suppliers such as NVIDIA.
Notably, SenseTime has not disclosed either the construction timeline or the investment amount. Such secrecy may indicate that the project is in an early stage of approval, or that data on government support is strategically sensitive. However, the fact that nearly 20 partners have been brought together speaks to a high degree of coordination within China's semiconductor ecosystem.
Currently, Chinese AI accelerators lag behind their Western counterparts in performance, especially in training large models. Therefore, creating clusters of 10,000 chips is an attempt to compensate for the gap in individual performance through massive parallelism. If the project proves successful, it could become a prototype for future national computing networks.
My comment: SenseTime's initiative is a vivid example of how geopolitical constraints accelerate technological sovereignty. Even if the current clusters do not match the efficiency of the NVIDIA H100, the very creation of such infrastructure will give Chinese developers invaluable experience in optimizing software for local hardware. In the long term, this could shift the balance of power in the global AI market.