AXERA (00600) appeared at the Industrial Fair and officially released the AX8860 high-computing-power edge AI chip series.

Leading AI chip company AXERA (00600) made a major appearance at CIIF 2026 (China International Industry Fair) under the theme "Intelligent Manufacturing Without Boundaries, Computing Power Reaches Everywhere," comprehensively showcasing its AI computing power foundation and product matrix for the real physical world, and officially releasing the AX8860 high-computing-power new product series. As a high-computing-power AI chip purpose-built for deploying large models at the edge and device side, the release of the AX8860 series marks edge AI's transition from the technology validation stage into a new cycle of business-value-oriented end-to-end productivity deployment. Four Major Exhibition Zones, Panoramic Display of AI Computing Power Deployment Capabilities At this year's Industrial Fair, AXERA set up four major exhibition zones: Core Computing Power and Technology Foundation, Industry Solutions, Software Architecture and Toolchain, and Ecosystem and Value Co-creation, collectively presenting full-scenario computing power solutions for physical AI. With AX8860 at the core, the company showcased a series of product forms ranging from chips and core modules to inference cards and inference servers, providing strong support for intelligent manufacturing from computing power foundation to application deployment. In the Ecosystem and Value Co-creation zone, AXERA partner Yuanli Lingji presented its self-developed DM0.5 general-purpose embodied brain, showcasing three tasks: connector plug-in quality inspection, Insert Anything general industrial insertion, and dexterous hand flexible cutting, vividly demonstrating the application potential of embodied intelligence in electronic assembly, precision manufacturing, industrial quality inspection, and flexible processing scenarios. As the intelligent core of this demonstration, DM0.5 relies on AXERA's high-computing-power platform to complete model inference, providing unified intelligent support for all three types of tasks. As large models, intelligent agents, and embodied intelligence continue to develop, AI is moving from the information world to the physical world, and edge and device-side computing power has become an important foundation for real-time response in intelligent manufacturing. For demanding industrial sites, AXERA believes that a complete closed loop of on-site intelligence must simultaneously satisfy four requirements: "see clearly, compute accurately, respond quickly, and execute stably." Only when on-site devices achieve closed-loop processing of "perceptionjudgmentexecution" can intelligence respond immediately at the forefront of events, driving edge AI to truly transform into new quality productive forces that drive industrial upgrading. AX8860 Series Officially Released: An Edge-Side Large Model Computing Power Foundation Born for On-Site Inference As a high-computing-power AI chip purpose-built for deploying large models at the edge and device side, AX8860 does not merely target peak specifications, but focuses on practical needs such as model deployment, on-site inference, and system collaboration, providing a more complete computing power foundation for physical AI applications. AX8860 combines 256 TOPS mixed-precision computing power (with native FP4 support) and 273 GB/s memory bandwidth, equipped with 128 GB LPDDR5X large-capacity memory, providing abundant computing power foundation for large models and multimodal applications. In terms of interconnect and expansion, the chip supports AXLink cascade technology and PCIe 5.0 high-speed interconnect, laying the foundation for multi-chip collaboration and elastic computing power expansion. Meanwhile, 64-channel video decoding and 16-channel video encoding capabilities can easily adapt to multi-channel visual perception and video analysis tasks, while the 6 nm advanced process balances ultimate computing power, energy efficiency ratio, and product integration, fully meeting the stringent demands for edge-side high computing power in scenarios such as intelligent manufacturing, warehousing and logistics, and Siasun Robot&Automation. Around AX8860, AXERA simultaneously released the Yuanxi Module A256S, and unveiled the Yuanxi series inference cards and inference servers. A256S adopts a modular design, featuring small size, strong computing power, and fast deployment, supporting multiple types of model applications including LLM, VLM, and Omni, providing more flexible combinations between core computing power and application hardware, helping customers accelerate product development and deployment. Dr. Qiu Xiaoxin, Founder and Chairman of AXERA, said at the release event: "We want to be our customers' long-term stable computing power foundation, so that enterprises can focus on business innovation without being distracted by computing power migration. AXERA is committed to bringing intelligence to every site, and the release of the AX8860 series high-computing-power chips is the fulfillment of this promise." Full-Spectrum Product Matrix Unveiled, Letting Customers Say Goodbye to the "Computing Power Multiple-Choice Question" With the official release of the AX8860 series, AXERA's AI chip landscape has once again expanded. Targeting differentiated computing power needs, AXERA has built a complete product matrix covering from small computing power to ultra-large computing power. From chips and core modules to inference cards and inference servers, AXERA is using a continuously supplied, non-disruptive product portfolio to help customers expand computing power along a unified technology roadmap, reducing repeated selection and system migration costs. As AXERA emphasized at the release event: customers can choose computing power according to their business stage and continuously expand along the same technology roadmap, without repeatedly facing the "computing power multiple-choice question." To provide customers with a more user-friendly deployment experience, the AX8860 series offers a complete toolchain and open-source software for developers and industry users, supporting mainstream inference frameworks and development environments such as vLLM, SGLang, and PyTorch, and providing service interfaces compatible with OpenAI API. Meanwhile, the product comes with a complete toolchain covering quantization, conversion, compilation, profiling, and debugging, helping users lower the threshold for model migration and application deployment, achieving "out-of-the-box" usability. From the Cloud to the Site, Bringing Intelligence Back to Where Things Happen The value of edge AI for intelligent manufacturing lies not in merely pursuing on-paper peak performance, but in whether it can stably produce results in real environments. Whether it is steel plate surface defect identification, cargo box and production line scheduling in warehousing systems, or companionship, rehabilitation, and care in Siasun Robot&Automation collaboration, AI needs to complete perception, inference, and response within the window of event occurrence. The AXERA AI computing power foundation, represented by the AX8860 series, is serving as infrastructure for edge-side scenarios, providing solid computing power support for the downward deployment of model capabilities: computing happens wherever data is generated; judgment is completed on-site in real time without waiting for remote round trips; instructions directly reach devices, driving AI to deliver value at the "last mile." For intelligent manufacturing, this means the chain from "seeing a problem" to "judging a problem" to "taking action" is greatly compressed, enabling on-site systems to move from passive response toward a more real-time, more collaborative, and more autonomous production paradigm. Looking ahead, AXERA will continue to improve full-scenario computing power supply and a full set of easy-to-use toolchains, working with industry partners to go deep into real business scenarios, driving the combination of AI capabilities with industry know-how, business innovation, and delivery services, so that underlying computing power better serves industrial upgrading. Let computing power reach the site, let intelligence return to where things happen, and let edge and device-side AI become perceivable, deployable, and executable productivity. As Dr. Qiu Xiaoxin said: "Computing power need not be far away in the clouds; intelligence should be right before our eyes."