At the 26th China International Industry Fair (CIIF), AI chip company Axera Intelligence officially launched the AX8860 high-compute edge AI processor series. Purpose-built for deploying large models at the edge and on-device, the release signals edge AI’s transition from technology validation to a productivity-focused, end-to-end deployment cycle.
The AX8860 series addresses real-world AI compute demands in physical environments. Built on 6nm process technology, it delivers 256 TOPS of mixed-precision compute with native FP4 support, 273 GB/s memory bandwidth, and up to 128 GB of LPDDR5X memory. The chip supports 64-channel video decoding and 16-channel video encoding, with multi-chip scaling via AXLink cascade technology and PCIe 5.0—two chips deliver approximately 512 TOPS, while four chips reach approximately 1024 TOPS.
At the exhibition, Axera presented four thematic zones covering core compute platforms, industry solutions, software toolchains, and ecosystem collaboration. Centered around the AX8860, the company simultaneously launched the Yuanxi A256S module and showcased Yuanxi series inference cards and servers, forming a complete product layout from chip to system level.
From an industry perspective, edge AI chips are evolving from “model-capable” to “production-ready.” The AX8860 launch demonstrates that domestic edge AI processors have achieved comprehensive competitiveness across compute performance, energy efficiency, and software stacks. For industrial vision, intelligent security, and robotics applications, the maturation of high-compute edge platforms will substantially lower deployment barriers and accelerate the engineering cycle from algorithm to product.
Junyi Microelectronics continues to expand its edge AI core board portfolio while closely tracking advancements in edge AI silicon. The emergence of high-compute platforms like the AX8860 provides downstream customers with richer compute options and opens new possibilities for industrial-grade edge computing solutions built on domestic chips.