Home
Zhipu Acquires Zhongke Jiahe in Hundreds-of-Millions-Yuan Deal to Boost AI Infrastructure and Domestic Computing Compatibility

Zhipu has reportedly invested hundreds of millions of RMB to acquire AI infrastructure firm XCore Sigma, bolstering its foundational model engineering, domestic computing power compatibility, and inference optimization.
Specializing in AI heterogeneous computing software, XCore Sigma focuses on compilation technology rooted in the Chinese Academy of Sciences’ Institute of Computing. Its team has contributed to compilers for Loongson, Shenwei, Cambricon, and Huawei Ascend chips, offering full-chain capabilities from virtual instruction set design to optimized deployment for domestic AI models.
Zhipu previously backed AI Infra firms like Jilu Technology, Wuwen Xinqiong, and Silicon Flow, addressing computing clusters and inference optimization. However, surging model usage post-GLM-5 release—reaching hundreds of millions of daily Coding Agent calls—exposed engineering challenges in high-concurrency inference architecture and KV Cache management.
This acquisition aims to enhance Zhipu’s multi-chip adaptation. XCore Sigma’s compiler bridges diverse domestic chip ecosystems, boosting utilization, while its SigInfer engine promises reduced latency and higher throughput. Successful production deployment could significantly lower model service costs.
As AI competition shifts toward computing power, software infrastructure, and ecosystems, underlying technologies like compilers are becoming crucial for large model companies’ long-term competitiveness.
Related article
OpenAI Launches GPT-5.3 Instant, Cutting Hallucinations by 27 Percent
Merely two hours after Google DeepMind unveiled Gemini 3.1 Flash-Lite, OpenAI moved with unprecedented velocity, officially releasing its most refined conversational model to date: GPT-5.3Instant.Unlike previous iterations that prioritized achieving
DeepSeek Faces 3-Day Service Outage, Lasting Over 10 Hours; Official Confirms Full Repairs
China’s top AI firm DeepSeek has restored full service stability following three consecutive days of unusual disruptions. Monitoring data confirms that between March 29 and 31, users experienced connection drops and response delays across the web cha
OpenAI buys NextSlide, team to work on ChatGPT
OpenAI has acquired NextSlide, a presentation startup, with its team joining OpenAI to contribute to ChatGPT-related product development. The financial details of the deal remain undisclosed. NextSlide founder Ahmed Beshry revealed that the acquisiti
Related Special Topic Recommendations
Comments (0)
0/500

Zhipu has reportedly invested hundreds of millions of RMB to acquire AI infrastructure firm XCore Sigma, bolstering its foundational model engineering, domestic computing power compatibility, and inference optimization.
Specializing in AI heterogeneous computing software, XCore Sigma focuses on compilation technology rooted in the Chinese Academy of Sciences’ Institute of Computing. Its team has contributed to compilers for Loongson, Shenwei, Cambricon, and Huawei Ascend chips, offering full-chain capabilities from virtual instruction set design to optimized deployment for domestic AI models.
Zhipu previously backed AI Infra firms like Jilu Technology, Wuwen Xinqiong, and Silicon Flow, addressing computing clusters and inference optimization. However, surging model usage post-GLM-5 release—reaching hundreds of millions of daily Coding Agent calls—exposed engineering challenges in high-concurrency inference architecture and KV Cache management.
This acquisition aims to enhance Zhipu’s multi-chip adaptation. XCore Sigma’s compiler bridges diverse domestic chip ecosystems, boosting utilization, while its SigInfer engine promises reduced latency and higher throughput. Successful production deployment could significantly lower model service costs.
As AI competition shifts toward computing power, software infrastructure, and ecosystems, underlying technologies like compilers are becoming crucial for large model companies’ long-term competitiveness.
OpenAI Launches GPT-5.3 Instant, Cutting Hallucinations by 27 Percent
Merely two hours after Google DeepMind unveiled Gemini 3.1 Flash-Lite, OpenAI moved with unprecedented velocity, officially releasing its most refined conversational model to date: GPT-5.3Instant.Unlike previous iterations that prioritized achieving
DeepSeek Faces 3-Day Service Outage, Lasting Over 10 Hours; Official Confirms Full Repairs
China’s top AI firm DeepSeek has restored full service stability following three consecutive days of unusual disruptions. Monitoring data confirms that between March 29 and 31, users experienced connection drops and response delays across the web cha
OpenAI buys NextSlide, team to work on ChatGPT
OpenAI has acquired NextSlide, a presentation startup, with its team joining OpenAI to contribute to ChatGPT-related product development. The financial details of the deal remain undisclosed. NextSlide founder Ahmed Beshry revealed that the acquisiti











