AI chipmaker Axera (Aixin Yuanzhi) launched its AX8860 high-compute edge AI chip series at the China International Industry Fair, themed "Unbounded smart manufacturing, computing that arrives". Designed specifically for deploying large models at the edge and on devices, the chip packs 256 TOPS of mixed-precision compute with native FP4 support, 273 GB/s of memory bandwidth and 128 GB of LPDDR5X memory, built on a 6 nm process.
For scaling, the chip supports AXLink cascade technology and PCIe 5.0 high-speed interconnection, enabling multi-chip coordination and elastic compute expansion. It also integrates 64-channel video decoding and 16-channel encoding for multi-camera vision workloads, targeting smart manufacturing, warehousing and logistics, and robotics.
Alongside the chip, Axera released the Yuanxi A256S module and showcased Yuanxi inference cards and inference servers. The product line supports mainstream frameworks including vLLM, SGLang and PyTorch, offers OpenAI-API-compatible interfaces, and ships with a complete toolchain covering quantization, conversion, compilation, profiling and debugging — letting customers scale compute along a single technology route without repeated re-selection.
Source: Sina Finance; DRAMeXchange
