On July 21, Zhipu AI embarked on the construction of a cutting-edge, 1GW-scale AI computing power data center within China, harnessing solely domestic AI chips. Concurrently, Zhipu AI proudly announced the official closure of its acquisition of CASIAH, a leading Chinese AI heterogeneous computing power software firm, on the same day. CASIAH, a spin-off from the Compiler Lab at the Institute of Computing Technology, Chinese Academy of Sciences, has been at the forefront of developing heterogeneous computing power software stacks and compiler optimization technologies, earning its reputation as one of China's premier AI infrastructure teams.

Insider sources reveal that these two strategic moves by Zhipu AI are designed to bolster two critical facets: computing power provision and computing power utilization. The establishment of the data center is poised to supply robust computing resources essential for large-scale model training endeavors. Meanwhile, the integration of CASIAH's expertise is expected to significantly enhance the utilization efficiency of heterogeneous chips, drive down inference costs, and streamline model deployment processes through its advanced foundational software capabilities, including compilers, Runtime systems, and inference engines.