"The intelligent upgrade of the industry is forcing a comprehensive reconstruction of the underlying infrastructure. The infrastructure upgrade of industrial communication must precede AI deployment. In other words, without network upgrades, everything else is empty talk." In an exclusive interview with GongKong China (media), Zhang Heng, Product Director of Moxa China, pointed out a commonly overlooked issue in the current intelligent transformation of the manufacturing industry.
Today, AI is accelerating its shift from the cloud to the production floor, driving industrial upgrades from the mere stacking of single-point applications to comprehensive upgrades across the entire chain of perception, analysis, decision-making, and execution. As an expert in industrial network communication, Moxa has always been at the forefront of technology implementation, sensing every iteration of market demands. In this interview, Zhang Heng shared observations and insights from the front lines of practice, focusing on core topics such as the changing demands for industrial networks in the AI era, scenario implementation, and security governance.
From Connectivity Pipelines to Intelligent Foundations: AI Reshaping the Positioning of Industrial Networks
In the past, the primary task of industrial networks was to achieve data connectivity and single-point control. From the early days of completing device connections based on industrial switches, to the continuous advancement of IT and OT integration and cloud-edge collaboration, and then to the collaborative hosting of AI, robotics, edge computing, and data, ultimately moving towards a network foundation for AI factories, the development trajectory of demands in China's industrial communication market is clearly visible.
As the manufacturing industry enters a new stage of accelerated AI implementation, manufacturing customers are no longer satisfied with mere device connectivity. Instead, they require networks to support the efficient collaboration of multiple types of business, enabling various complex and precise control data to be provided with a stable, synchronized, and deterministic operating environment through an "end-to-end unified network." "Industrial networks are no longer just communication devices in a factory; they have become the foundational network infrastructure for industrial AI," Zhang Heng stated.
The fundamental driving force behind these changing demands lies in the fact that AI is reconstructing the operational logic of production systems. When perception, analysis, decision-making, and execution form a complete intelligent closed loop, AI transitions from a tool for auxiliary analysis to a core component of the production system, and the role of the network is upgraded accordingly. Zhang Heng summarized the irreplaceability of industrial networks in the AI production closed loop into four core dimensions:
01. First, transforming on-site perception into usable real-time data. Industrial networks connect sensors, PLCs, robots, AI vision, and equipment, ensuring that on-site data is delivered to AI analysis systems in a stable and low-latency manner;
02. Second, bringing AI decisions back to the production floor. The judgments generated by AI ultimately need to be translated into equipment actions, such as adjusting robot paths, changing process parameters, or triggering equipment control. The industrial network serves as the critical bridge between AI and OT control systems;
03. Third, ensuring real-time performance and determinism. No matter how accurate AI decisions are, if there is communication latency, packet loss, or network interruption, the decisions cannot be reliably executed;
04. Fourth, keeping the entire network continuously visible, manageable, and protected. In the context of the high integration of AI, OT, and IT, the network must provide traffic visibility, equipment status monitoring, fault diagnosis, and cybersecurity protection to ensure the long-term stable operation of equipment.
This is precisely the aspect that most manufacturing enterprises currently overlook in AI implementation. Without a solid underlying network foundation, AI decisions are out of the question. Network latency, data packet loss, and time asynchronization among equipment—every tiny error can lead to disastrous consequences in production decision-making.
"The value of industrial networks lies not only in connecting AI and equipment but also in ensuring that AI judgments can be reliably transmitted with the correct data at the right time, forming a sustainable closed-loop process of perception, analysis, decision-making, and execution," Zhang Heng summarized. "In other words, AI decides what to do, while the industrial network determines whether it can be done reliably in real time."
From Reliable Connectivity to Deterministic Communication: Upgrading Industrial Network Technologies
As the core foundation for AI implementation scenarios, the capability boundaries of industrial networks affect whether AI decisions can be reliably executed. This also drives industrial networks towards predictable and controllable deterministic transmission. Preventing latency jitter, supporting the coexistence of multiple types of traffic, and ensuring reliability in mobile scenarios have become imperative issues to solve. This change directly redefines the capability benchmarks for OT networks. Traditional production lines require reliable control, with stable connectivity as the core, whereas after AI enters the production system, the core requirement is upgraded to predictable data and control closed loops.
Upgraded Requirements for Latency and Jitter
In high-precision scenarios such as AI vision servo control, AI models can complete hundreds of inferences per second. A latency jitter of just a few milliseconds can lead to trajectory calculation deviations, ultimately resulting in product scrapping or even equipment collisions. Therefore, the importance of low latency, low jitter, and time synchronization has been elevated to an unprecedented level in the AI era.
Deterministic Transmission of Multiple Types of Traffic
AI vision generates massive amounts of image data, and inference models require large-scale data exchange. Meanwhile, critical control data such as PLC and motion control are also transmitted over the same network, and various types of traffic must not interfere with each other. This requires the network to possess multiple capabilities such as traffic isolation, QoS, TSN, redundancy, and priority management to ensure the deterministic transmission of critical services.
For traditional production lines, network stability is sufficient. However, for applications in AI scenarios, every millisecond of network latency must be predictable and controllable. This is the most profound change that AI has brought to industrial networks.
On the wireless side, as applications such as generative AI, digital twins, AI vision inspection, and autonomous mobile robots rapidly penetrate the manufacturing floor, factories are evolving into AI-driven physical intelligence. Wireless technologies such as Industrial Wi-Fi and Industrial 5G have taken center stage due to their flexible deployment and ability to adapt to the needs of flexible manufacturing and dynamic expansion. They are currently being deployed at scale in key scenarios such as AGVs, overhead cranes, and industrial robots.
However, the specific characteristics of wireless networks also bring new challenges. Wireless networks are invisible and intangible, highly susceptible to various environmental factors, making anomalies difficult to detect in a timely manner; they are often only noticed when operations are hindered. Meanwhile, communication interruptions during device roaming across AP zones in mobile scenarios can directly cause robot standstills and inspection frame drops, thereby affecting production capacity and yield rates. Therefore, roaming quality in mobile scenarios has become the most critical pain point for enterprises deploying industrial wireless networks today.
To address these challenges, Zhang Heng introduced two core capabilities of Moxa in wireless technology. The first is Turbo Roaming technology, which provides high-speed roaming switching in less than 50 milliseconds to help mobile devices such as AMRs maintain connection stability when switching between different APs. The second is AeroMesh technology, featuring a multi-path redundancy architecture that supports self-organizing networks and automatic healing. When a communication path or network node fails, the system can find an available path and restore communication within 2.5 seconds.
Currently, these two technologies have been implemented in multiple AI scenarios. For example, in the scenario of autonomous mobile robots equipped with AI vision, the robots need to continuously transmit high-definition images, sensor data, and control commands when moving across workshops. The introduction of Turbo Roaming effectively meets the needs for stable transmission of visual data and real-time control for autonomous mobile devices. In the intelligent inspection scenario of high-tech factories, Turbo Roaming technology enables high-speed roaming switching in less than 50 milliseconds, supporting the real-time and stable transmission of images and sensing data during long-term autonomous inspections.
During the interview, Zhang Heng specifically mentioned that industrial wireless networks should not only focus on transmission performance but also drive the transition of operation and maintenance (O&M) models from post-fault handling to proactive O&M. Relying on Moxa's MXview One global visualization management platform, enterprises can grasp the status of all network equipment, security conditions, and abnormal alerts in real time, shifting troubleshooting forward to prevent faults from expanding and affecting production. The addition of this proactive O&M capability ensures that the value of industrial networks is no longer limited to communication connectivity but extends to full-lifecycle stability assurance.
From Security Foundation to Global Expansion Empowerment: Cybersecurity Advancing Towards Full-Lifecycle Governance
In the AI era, the number of connected devices continues to grow, and software dependency is deepening, expanding the attack surface of industrial networks accordingly. Meanwhile, as the EU Cyber Resilience Act (CRA) is progressively advancing, market requirements for equipment suppliers have extended from product security to continuous governance capabilities throughout the entire product lifecycle. These dual pressures are profoundly changing manufacturing enterprises' cognition and selection criteria for OT security.
"OT security is not about purchasing the most secure equipment available today, but about choosing a supplier that can continuously discover, respond to, patch, and prove security throughout the equipment's lifecycle of over a decade. This is a long-term commitment from the equipment supplier and a highly critical core capability," Zhang Heng revealed the essence of OT security.
Security is not a one-time product delivery but a long-term systematic capability. Therefore, Moxa not only provides industrial security-related products such as Firewalls, VPNs, IPS (Intrusion Prevention Systems), and IDS (Intrusion Detection Systems) but also makes a long-term commitment to full-lifecycle security governance capabilities.
Currently, many enterprises' understanding of OT security remains at the level of deploying security devices such as firewalls and security gateways, which proves inadequate in the face of increasingly complex attack postures in the AI era. Full-lifecycle security governance is becoming the core standard for industrial network security.
Currently, Moxa has obtained the dual IEC 62443-4-1 Maturity Level 3 (ML3) certifications (IECEE and ISCI/ISASecure). This certification covers product security development lifecycle governance, threat modeling, standardized operation of the Product Security Incident Response Team (PSIRT), vulnerability and patch management, and supply chain security control.
Regarding how manufacturing enterprises should build the security capabilities of their network foundations, Zhang Heng provided a clear practical path:
01. First, establish the concept of "security by design" and integrate cybersecurity risks into the product design, development, testing, and verification processes;
02. Second, establish a continuous vulnerability management and response mechanism, balancing patch updates with production continuity;
03. Third, pay attention to the supplier's software supply chain management capabilities to ensure that the affected scope can be quickly located when vulnerabilities occur in third-party components;
04. Finally, establish a security governance system that covers the entire product lifecycle.
For enterprises advancing AI-empowered production line upgrades, Zhang Heng also offered practical advice: "First, build a solid overall industrial backbone network and establish a reliable, stable, deterministic, and resilient industrial communication foundation. Only with a good network foundation can the effectiveness of AI be sustainable and continuously advanced."
Notably, Moxa will soon make an appearance at the 2026 CIIF (China International Industry Fair) as a "Reliable Communication Expert and Professional Global Expansion Partner," showcasing its technical capabilities in high-end manufacturing and highly reliable deterministic backbone networks, as well as full-scenario real-time edge connectivity.
It is revealed that Moxa will present five core directions: end-to-end unified edge access and data uplink solutions, a multi-redundancy network resilience system, security foundation capabilities based on the IEC 62443 standard, a visualized panoramic O&M system, and compliance design and global service capabilities for enterprises' global expansion.
These five directions cover the full scenarios of edge access, resilient operation, security, O&M, and global expansion compliance, precisely addressing the core needs of enterprises' intelligent upgrades.
Currently, the intelligent transformation of the manufacturing industry has entered a deep-water zone, and the value of industrial networks is being re-recognized across the industry. From connectivity pipelines to AI foundations, every upgrade of industrial networks is paving the way for the deepened implementation of AI applications. Moxa's judgment of "network first" provides a clear underlying logic for manufacturing enterprises exploring their intelligent transformation. The development of industrial intelligence cannot ignore the construction of underlying infrastructure. Only by building a solid industrial communication foundation can AI technology release lasting value on the production floor.