Author: Jiu Lin
This year, price hikes have become the main theme of the electronics industry, and industrial automation products are also being repriced amid this round of cost pressures.
Major industrial giants have successively issued price increase notices. Schneider Electric's entire Acti9/Acti9-S series of miniature circuit breakers has seen a price hike of approximately 6%. During the same period, the ATS22 series soft starters, the entire STB Advantys series of I/O and Momentum series controllers, Ethernet gateway modules, Ethernet communication modules, and industrial PCs (IPCs) have also been priced higher.
Omron has released a price adjustment notice, increasing the prices of industrial automation products including PLCs, HMIs, robots, relays, sensors, switches, and temperature controllers. The minimum price increase is 5%, while some parallel robot products have seen a price hike of up to 50%.
Driven by the continuous guidance of manufacturing revitalization policies, the release of demand in strategic industries such as new energy and semiconductor manufacturing, and the emergence of new application scenarios like embodied robots, the industrial computer market is poised for a new cycle of growth. IDC forecasts that shipments are expected to reach 6.93 million units by 2029.
01. What the Industrial Field Really Needs Is Not "Cloud AI"
As AI enters the industrial field, the first thing it changes is the location of data processing.
On an electronics manufacturing production line, industrial cameras generate a massive amount of images per second. If all images are uploaded to the factory server and the system waits for the cloud or central node to return results, the system must not only bear the network transmission pressure but also face issues of latency, jitter, and data security. For high-speed chip mounting, solder joint inspection, or wafer defect detection, the judgment results must be returned within the equipment's takt time; otherwise, no matter how accurate the AI model is, it cannot be integrated into the production process.
Edge computing has therefore become the practical choice for industrial AI. Industrial PCs connect directly to cameras, PLCs, and robots: cameras capture images, and the IPCs handle decoding, preprocessing, and inference. The results are returned to the control system via digital I/O, serial ports, EtherCAT, or industrial Ethernet. Only filtered results, defect samples, and necessary operational data are uploaded to the MES, data middle platform, or cloud platform. The cloud is responsible for training, analysis, and cross-factory comparisons, while on-site equipment makes decisions within a time window ranging from milliseconds to seconds.
Advantech MIC-733: Turning the Jetson AGX Orin into an Industrial Node
The Advantech MIC-733 represents a relatively mature route in industrial AI: turning the NVIDIA Jetson AGX Orin module into a directly deployable industrial computing node. The advantage of the Jetson route is its high level of integration, with the CPU, GPU, AI accelerators, and memory integrated into a single module, resulting in a compact device size, relatively controllable power consumption, and a mature software ecosystem. The trade-off is that its expandability and memory upgrade space are not as good as those of x86 platforms, and model deployment relies more heavily on the NVIDIA software stack.
The MIC-733 is available in 32GB and 64GB memory versions. The 64GB version features a 12-core Arm Cortex-A78AE CPU, 2048 CUDA cores, and 64 Tensor Cores, delivering AI performance of up to 275 TOPS; the 32GB version offers AI performance of up to approximately 200 TOPS. It adopts a fanless design and provides four Gigabit Ethernet ports, with PoE available as an option. It also features USB 3.2 Gen2, digital I/O, RS-232/422/485, GMSL2 interfaces, and M.2 NVMe storage. With a 9 to 36V DC input and dual SIM card design, it can be installed in AMRs (Autonomous Mobile Robots), in-vehicle equipment, or outdoor edge terminals. Optional PCIe expansion modules also leave room for high-speed acquisition cards.
These interfaces determine its application directions. On AMRs, the MIC-733 can locally fuse camera, LiDAR, and positioning data to perform obstacle recognition, localization, and path planning without transmitting all raw data back to the server. In multi-camera vision scenarios, PoE and USB interfaces can directly connect to workstation cameras for workpiece positioning, robot grasping, and process monitoring. In industrial parks, warehousing, and transportation scenarios, complete videos can be kept locally, with only alerts and structured results uploaded. The key to the Jetson route is not to turn the device into a "mini GPU workstation," but to provide a complete software ecosystem in places where power consumption and physical size are constrained.
ADLINK DLAP-411-Orin: An Edge Vision Platform for AMRs
If the MIC-733 demonstrates the general form factor of an industrial node, the ADLINK DLAP-411-Orin is closer to a complete mobile robot. This platform uses the NVIDIA Jetson AGX Orin processing engine, configured with an 8-core Carmel Arm 64-bit CPU, 1792 Ampere GPU cores, 32GB of LPDDR5, and 64GB of eMMC, and provides Gigabit Ethernet, USB, and M.2 interfaces. Its compact size, low power consumption, and camera-direct interfaces make it suitable for embedding inside robots and mobile devices.
The ADLINK DLAP-211-Orin series further covers scenarios such as smart cities, machine vision inspection, autonomous vehicles, and robotics. Compared to traditional IPCs, the advantage of the Orin platform lies in the high integration of the GPU, deep learning accelerators, and memory, enabling visual inference within a smaller form factor. The limitation is that the expansion slots and storage upgrade space are limited, requiring the number of cameras, model size, and system image to be determined during the product design phase.
Previously, ADLINK collaborated with the Vietnamese robotics company Phenikaa-X to develop an AMR pallet truck, moving edge AI IPCs from a "general development platform" into complete robot systems. This poses three requirements for IPCs: first, the device must be small enough to be embedded inside the robot; second, the interfaces must directly face the cameras and controllers to reduce additional conversion equipment; third, the system must support remote updates and fault diagnosis, as the cost of maintaining robots one by one will rise rapidly as their numbers increase.
Neousys Nuvo-10109GC: 600W Blackwell Pushes IPCs Toward Edge Servers
When the industrial field requires larger models, higher throughput, and multi-channel vision, the Jetson route is no longer the only answer. The Neousys Nuvo-10109GC targets a single 600W GPU, supporting the NVIDIA RTX PRO 6000 Blackwell and selected 600W graphics cards. Its product form factor is already close to an edge server capable of working on the shop floor.
The Nuvo-10109GC uses 12th, 13th, and 14th Gen Intel Core processors, supporting 35W or 65W LGA1700 models, and up to 128GB of ECC or non-ECC DDR5-4800 memory. In addition to the PCIe x16 Gen4 slot used by the GPU, it also provides three PCIe Gen3 x4 signal expansion slots for installing camera acquisition cards, in-vehicle Ethernet cards, or high-speed network adapters. Official performance specifications include: up to 4000 TOPS for large language model inference and generative AI at FP4 precision, and up to 125 TFLOPS for autonomous driving and AI vision applications.
The device provides two 2.5GbE ports, one Gigabit Ethernet port, and an optional 10GbE port, configured with six USB 3.2 interfaces. Storage utilizes a pluggable 2.5-inch SATA bracket, while also supporting an NVMe bracket and one M.2 Gen4 x4 NVMe slot. The 9 to 32V DC input is compatible with 12V and 24V in-vehicle systems, and the operating temperature ranges from -25°C to 60°C. To address the transient power consumption and transportation vibration of the 600W GPU, the product is also equipped with dedicated power supply, cooling, and graphics card locking structures.
In wafer, packaging, and high-end electronics inspection, computing power is just the starting point. Camera acquisition cards must stably feed multiple images into memory, PCIe channels must avoid contention between different functional cards, NVMe must withstand continuous writes, and inference results must be correlated with wafer coordinates, batches, and equipment parameters. The multi-channel PCIe expansion of the Nuvo-10109GC is precisely designed to fit these data entry points and acceleration cards into a single machine. It can be used for on-site inference of complex vision models, as well as for VIL/HIL testing, playing back real sensor data to the controller.
AAEON BOXER-6839-RPL: Built for Automation
AI will not turn all industrial equipment into GPU servers. A large number of material feeding machines, production line workstation equipment, SCADA systems, and PLC networks first require stable x86 compatibility, rich I/O, and sustainable data logging. The AAEON BOXER-6839-RPL represents this type of demand.
This product features a modular design for the entire machine, aiming to meet the diverse project needs of current industrial system integrators. The whole machine supports multiple generations of Core processors and is equipped with a rich and complete set of I/O interfaces. It can serve as a highly adaptable hardware platform for material feeding machines, production line workstation equipment, and various scenarios such as high-performance machine vision.
The BOXER-6839-RPL is equipped with the full range of 12th, 13th, and 14th Gen Intel Core processors, supporting up to 24 cores and 32 threads for high-performance computing power. Aligning with the diversified needs of modern industry, integrators can freely configure solutions based on project budgets and performance requirements. In terms of storage, the BOXER-6839-RPL supports both SATA and NVMe storage media. The device is equipped with two 2.5-inch drive bays, plus one M.2 2280 M-key slot, balancing high-speed read/write performance with long-term stable data logging needs, making it suitable for SCADA (Supervisory Control and Data Acquisition) and PLC (Programmable Logic Controller) network scenarios.
Whether it is the cost-effective computing power required by material feeding equipment and production line nodes, or high-concurrency, real-time parallel computing scenarios such as machine vision inspection, the BOXER-6839-RPL can stably handle them all.
NORCO BIS-6390ARA-D10: 6 TOPS Also Has Its Own Market
Compared to Advantech, ADLINK, and Neousys, NORCO's route is closer to domestic embedded computing. The BIS-6390ARA-D10 uses the Rockchip RK3588 processor, equipped with four Cortex-A76 and four Cortex-A55 cores, with the A76 reaching a maximum clock speed of 2.4GHz. It integrates a Mali-G610 MP4 GPU and a 6 TOPS NPU, supporting mixed-precision operations such as INT4, INT8, INT16, and FP16.
In terms of storage and memory, the product supports 4GB to 16GB of LPDDR4/LPDDR4X, providing up to 128GB of eMMC, TF card, and M.2 NVMe SSD expansion. Interfaces include dual Gigabit Ethernet ports, four USB 3.0 ports, serial ports, and various display interfaces, with support for Wi-Fi 6, Bluetooth 5.0, and 4G/5G module expansion.
The 6 TOPS NPU can meet the needs of lightweight object detection, personnel recognition, equipment status recognition, and video analysis. NORCO also emphasizes the fanless design, industrial-grade wide temperature range, shock resistance, dustproof capabilities, and operating system adaptation. For equipment manufacturers that need to deploy in large quantities, low power consumption, low cost, and long-term supply are often more important than peak computing power.
02. The Next Competition in Industrial Edge Computing
Currently, the industrial computer market maintains a relatively stable competitive landscape: relying on a market share of over 36%, Advantech remains the most influential vendor in China's IPC market, while Evoc, NORCO, NODKA, and ADLINK also hold their own places.
From a platform perspective, x86 remains the mainstream in industrial computing. It boasts a mature Windows and Linux ecosystem, capable of being compatible with a vast amount of industrial software, acquisition cards, motion control cards, and databases, and it also allows customers to retain their existing engineering and maintenance systems. The emergence of Intel Core Ultra and AMD Ryzen AI has endowed x86 platforms with low-power NPU capabilities, enabling "general control" and "lightweight inference" to be completed within the same processor.
The Arm route continues to expand its influence in power-sensitive and size-constrained scenarios. NVIDIA Jetson relies on the GPU, CUDA, and JetPack ecosystems, making it suitable for robotic vision, mobile devices, and multi-camera edge inference. Domestic SoCs (System on Chip) such as Rockchip provide an alternative choice in industrial IoT (Internet of Things), smart energy, and low-to-medium cost vision equipment. Their advantages lie in integration and power consumption, while their limitations are that memory expansion, industrial software compatibility, and long-term supply require project-level validation. RISC-V remains in specific vertical fields and the ecosystem building stage; whether it can scale up in the future depends on the maturity of heterogeneous computing, operating systems, and industrial software.