On October 28, SmartSens Technology (hereinafter referred to as "SmartSens"), a leading domestic CMOS Image Sensor (CIS) manufacturer, announced that the official website of its wholly-owned subsidiary, Weiyun Creation (Shanghai) Artificial Intelligence Technology Co., Ltd. (hereinafter referred to as "Weiyun Creation ") (https://weiyuncreation.com/), has officially launched. The company released two AI training-inference appliances, the Weiyun™ 1000Pro and Weiyun™ 1000Ultra, as well as the Weiyun™ Station Physical AI Agent Workstation, forming a product matrix covering enterprise-level local AI computing, high-performance large model inference, and professional desktop AI applications.
This marks a substantive step in SmartSens's layout in the field of physical AI, indicating that this chip company, renowned for image sensors, is attempting to extend its capability boundaries from the chip level to the system and solution levels.
Considering its technical architecture and SmartSens's strategic trajectory over the past two years, the debut of Weiyun Creation completes a key link in SmartSens's transition from CIS sensing chips to a full-stack physical AI layout. SmartSens is building a complete computing and sensing foundation for physical AI along the chain of "sensing—edge computing—cluster training/inference—high-speed interconnect."
According to business registration data, Weiyun Creation was established on September 3, 2026, with a registered capital of CNY 30 million. It is 100% owned by SmartSens, with Zhao Songen as the legal representative. Its business scope covers integrated circuit chip design and services, artificial intelligence industry application system integration services, artificial intelligence application software development, cloud computing equipment manufacturing and sales, etc.
Information on the official website indicates that Weiyun Creation positions itself as a technology innovation enterprise focused on artificial intelligence R&D and application, concentrating on physical AI and intelligent computing fields, providing integrated hardware-software products and solutions. The company will rely on core technologies such as self-developed innovative physical AI agents and AI-ASIC cluster interconnection to continuously lay out AI training-inference appliances, AI video creation tools, customized AI agents, and vertical domain applications, gradually covering an AI product ecosystem from local computing infrastructure to software products and industry applications.
Currently, Weiyun Creation has announced three products: Weiyun 1000Pro, Weiyun 1000Ultra, and Weiyun Station.
Among them, the Weiyun 1000Pro is an enterprise-grade flagship AI training-inference appliance, with computing power reaching up to 13 PFLOPS, maximum video memory of 1.28 TB, and inference power consumption as low as 1 kW. Adopting a 2U rack-mounted design, it supports single-machine full-parameter local deployment of mainstream large models such as DeepSeek-V4.1-Flash, GLM-5.3, and MiniMax-H3. Through horizontal scaling, it can achieve multi-machine distributed full-parameter deployment of larger-scale models such as Kimi K3 and Qwen3.8-2.4T, catering to enterprise local AI data center needs and providing integrated support from local computing power to business applications.
The Weiyun 1000Ultra is a flagship high-performance AI training-inference appliance, featuring up to 47.5 PFLOPS FP4 computing power and up to 4.66 TB of video memory, designed for full-parameter deployment of ultra-large-scale models and complex computational tasks. This product supports single-machine local full-parameter deployment of large models such as Kimi K3, Qwen3.8-2.4T, DeepSeek-V4-Pro, GLM-5.3, and MiniMax-H3. Adopting 8U/12U chassis solutions, it supports configuration upgrades and multi-machine interconnection, with optional GPU and FPGA acceleration cards, meeting the high-performance computing needs of large enterprises and industry AI platforms.
The Weiyun Station is a physical AI agent workstation, with computing power reaching up to 7.7 PFLOPS and video memory of 680 GB. Targeting developers, industry users, professional creators, and OPCs (One-Person Companies), it integrates local model execution, AI application development, and assisted creation into the desktop working environment. This product supports single-machine full-parameter local deployment of models such as DeepSeek-V4.1-Flash, GLM-5.3-Flash, MiniMax-M3.1-Flash, Qwen 3.5-397B-A17B, and MiniMax-H3. Through horizontal scaling, it can achieve multi-machine distributed deployment of larger-scale models. The Weiyun Station supports optional GPUs to enhance token throughput and video processing performance, optional high-speed network cards supporting up to 12 high-performance, low-latency environmental perception cameras, and combines self-developed spatial perception agent technology to enable AI to perceive, understand, and interact with real environments.
The official website also disclosed the WAAG™ (Weiyun AI-ASIC Grid) heterogeneous AI-ASIC cluster interconnection technology. This architecture densely integrates computing power, storage capacity, and transport capacity into a single server, achieving multi-cluster computing synergy and low-latency interconnection through high-speed optical interconnects. Combined with KV Cache dynamic offloading technology to reduce storage load, it provides efficient inference and localized deployment specifically for full-parameter large models.
In addition, Weiyun Creation provides ready-to-use innovative AI applications centered around content creation, enterprise production, and AI digital humans, including the self-developed professional AI video creation tool "Zhiying," AI digital humans, AI-assisted programming, and customized industry agents.
From "Integrated Sensing and Computing" to "Cluster Training and Inference"
Prior to this, market perception of SmartSens's AI layout largely remained at the level of "CIS chip supplier." In the approximately CNY 3.2 billion private placement completed in September this year, the third fundraising investment direction, "CIS and Edge AI ASIC Solutions for Visual AI," is jointly implemented by SmartSens and its wholly-owned subsidiary Feiling Micro, aiming to build an "AI ASIC+Sensor" integrated sensing and computing architecture. Feiling Micro has previously launched M2 and A2 chips, respectively covering edge AI processing needs for automotive vision and robotics/industrial vision. This constitutes the edge computing foundation of SmartSens's AI strategy.
However, edge computing power is naturally constrained by power consumption, cost, and scenarios. In physical AI scenarios, whether for embodied intelligent robots, advanced intelligent driving, or industrial inspection, edge-side chips alone cannot complete large model training and full-parameter inference.
Weiyun Creation's WAAG™ heterogeneous AI-ASIC cluster interconnection technology precisely targets this gap: achieving multi-cluster computing synergy through high-speed optical interconnects, combined with KV Cache dynamic offloading to reduce storage load, supporting full-parameter local deployment of the highest-level large models. Edge-side integrated sensing and computing handles "local perception and real-time response," while cluster training and inference handle "heavy-load training and complex reasoning." Only when these two are connected via high-speed interconnects does a truly deployable computing closed loop for physical AI emerge.
SmartSens's technical reserves along this chain are not hastily assembled. Its HOC High-Speed Optical Interconnect Business Group has built full-stack R&D capabilities covering optical devices, high-speed circuits, signal processing, and system integration, with core members possessing over ten years of relevant experience. The previous private placement prospectus explicitly proposed building a "Visual AI—AI Interconnect—Edge AI ASIC" full-chain ecosystem, listing MicroLED optical interconnect as a continuous layout direction. Weiyun Creation's WAAG™ architecture is the first public implementation of this technical accumulation in system-level products.
CIS Remains the Primary Entry Point for Physical AI Perception
It must be emphasized that SmartSens's extension into AI computing and system layers does not mean the CIS main business is marginalized. On the contrary, perception is SmartSens's unique entry advantage into physical AI.
The core of physical AI lies in enabling agents to understand and interact with the real world, with visual perception being the starting point of this interaction. SmartSens's technical accumulation in global shutter CIS—the performance of the SmartGS series in robot obstacle avoidance, spatial positioning, and industrial inspection—makes it a key supplier in the physical AI data collection phase. Strategic cooperation with Orbbec Inc on embodied intelligence 3D vision further strengthens "Deep Engine+CIS" dual-chip synergy, covering a multimodal data platform from EGO first-person perspective data collection to WristCam wrist interaction.
In the first half of 2026, SmartSens's AI Smart Vision Ecology segment contributed revenue of CNY 495 million. Machine vision demands in emerging scenarios such as drones, robots, and AI glasses are becoming clear sources of incremental growth. These scenarios are typical landing grounds for physical AI. The customer relationships, scenario understanding, and data interface experience accumulated in perception chips constitute a differentiated barrier for SmartSens when extending to system-level solutions, distinguishing it from pure computing vendors.
Beyond CIS: SmartSens's Full-Stack Layout for Physical AI
SmartSens is laying out a grand chessboard for physical AI. On this board, CIS perception chips, edge ASICs, MicroLED optical interconnects, and the newly released AI computing products from Weiyun Creation together form a complete foundation puzzle. Of course, moving from single-point "chip-level capabilities" to integrated "system-level solution competitiveness" still involves non-negligible gaps in organization and capabilities. System integration and AI software capabilities require a talent structure and operational logic different from chip R&D. The WAAG™ architecture focuses on dense integration of computing, storage, and transport power, providing a differentiated technical path for edge physical AI scenarios.
However, SmartSens's strategic main line is already very clear: As the CIS industry bids farewell to simple pixel races and enters an era of comprehensive competition integrating performance, scenarios, and computing power, SmartSens has not limited itself to the sensor chip track but extends upward based on visual perception, connecting interconnects, edge computing, and cluster training/inference chains to capture full-chain added value.
The official release of Weiyun Creation's three AI computing products this time is a landmark step in SmartSens's physical AI perception-computing foundation chessboard, moving from forward-looking layout at the chip level to system-level product implementation verification. The success or failure of this game hinges on whether Weiyun Creation can transform SmartSens's long-term accumulated chip hardware engineering advantages into stable and reliable delivery capabilities for system products in real-world scenarios.
We await the outcome.
Editor: ChipInsight - Langkejian