Cambricon said accelerating large-model technology is driving sustained growth
in intelligent-computing demand and will continue to focus on AI chip design
innovation to bolster competitiveness, expand market share and accelerate
scenario deployment. For internet customers it is prioritizing large models,
multimodal and search/recommendation applications, and is optimizing compute
performance, communication efficiency and framework capabilities to improve
integrated performance in typical internet workloads. The company said 1H2026
hardware work includes a new-generation processor microarchitecture and
instruction set under development, with optimizations targeting large-model
training and inference and aiming to improve programmability, ease of use,
performance, power consumption and silicon area.
