Sixth-generation AI processor under development! Cambricon officially announces: adaptation for five major domestic large models completed.

At the 2026 semi-annual results briefing, Cambricon Chairman Chen Tianshi recently released a significant signal, revealing that the company's sixth-generation AI processor microarchitecture and instruction set are still in the research and development stage.

Chen Tianshi disclosed that Cambricon has successfully completed inference adaptation for five major domestic mainstream large models: GLM, DeepSeek, Qwen, Kimi, and MiniMax, corresponding to model developers Zhipu, DeepSeek, Alibaba, Moonshot AI, and MiniMax respectively.

On the training side, Cambricon is continuously advancing the adaptation and optimization of models in the DeepSeek, Qwen, and Hunyuan series.

According to Chen Tianshi, the sixth-generation AI processor will focus on deep optimization for large model training and inference scenarios. The new generation of products will comprehensively enhance competitiveness across multiple dimensions, including programming flexibility, ease of use, performance, power consumption control, and area optimization.

As a leading enterprise in the domestic AI chip sector, Cambricon has previously launched multiple product series, including Siyuan and Xuansi. Currently, the company is continuously advancing its "chip-model synergy" strategy to strengthen the ecosystem collaboration between AI chips and large models.

In terms of financial performance, Cambricon delivered an impressive report card. In the first half of 2026, the company achieved operating revenue of 5.996 billion yuan, a year-on-year increase of 108.13%; net profit attributable to shareholders of the listed company reached 2.311 billion yuan, up 122.61% year-on-year. Non-GAAP net profit attributable to shareholders was 2.166 billion yuan, up 137.30% year-on-year.