Industrial Edge AI is moving into a new phase. According to the International Federation of Robotics, the number of industrial robots operating worldwide reached a record five million in 2025, while annual installations exceeded 600,000. The organization also reported that shipments of professional service robots grew by 24% over the same year. These figures do not mean that every AI application has moved beyond the pilot stage. They do show that Physical AI—AI embedded in machines that perceive and act in the physical world—is being deployed in operating environments.
For OEMs, the question is therefore changing. It is no longer only whether an AI model can run on an edge device. It is whether the complete system can coordinate several workloads, connect with industrial infrastructure, support applicable security requirements and evolve across multiple product generations. Four shifts may make those questions increasingly important in 2027.
AI workloads are becoming coordinated systems
Early Edge AI projects may focus on one visible capability: an object-detection model, a navigation algorithm or an inspection task. A commercial system may need to support additional functions. An autonomous mobile robot may combine perception, sensor fusion, route planning, application logic and time-sensitive control. An inspection platform may process several image streams while exchanging data with production equipment and a higher-level system.
These workloads place different demands on compute, memory and bandwidth. They may also have different latency and power requirements. Selecting a platform around the peak performance of one model may therefore create integration challenges when the complete application comes together.
OEMs should consider workload partitioning early in the design process. CPUs, GPUs, NPUs and dedicated control resources can serve different roles. Treating them as parts of one architecture can help engineering teams allocate workloads and limit the need to create separate processing islands for new functions. This approach may make it easier to add capabilities without requiring an early system redesign.
IT/OT convergence is redrawing the industrial edge
The International Federation of Robotics identifies the convergence of information technology and operational technology as a trend shaping robotics. The same shift is reshaping industrial automation, as intelligent machines become connected nodes within the factory's data and control architecture.
That does not mean moving every workload to the cloud. Real-time inference, decision-making and control may need to remain local, while selected data and services connect to wider systems.
Teams must establish those boundaries deliberately: which functions need local autonomy, which data should leave the machine and how operation continues when connectivity changes. AI processing must also connect to cameras, sensors, industrial networks and control equipment through the interfaces required by the application.
The objective is a resilient topology that protects local responsiveness while allowing data and services to move where they create the most value.
Lifecycle security is becoming a design requirement
Security can no longer be treated as a feature added shortly before release. A connected industrial product may remain in operation for years, during which vulnerabilities, software dependencies and regulatory expectations will continue to change.
The EU Cyber Resilience Act makes this lifecycle responsibility more explicit. It introduces cybersecurity requirements covering the planning, design, development and maintenance of products with digital elements. Its reporting obligations have applied since 11 September 2026, while the main obligations will apply beginning 11 December 2027.
For OEMs, the implication reaches beyond compliance documentation. The underlying platform must support a software foundation that can be maintained throughout the expected product life. Security processes, vulnerability handling and updateability all become architecture considerations.
Addressing them early may help reduce the risk of discovering, late in development, that a prototype software stack cannot support the commercial product's security and maintenance requirements. Lifecycle readiness is therefore becoming part of platform selection and procurement, not only an operational concern after deployment.
OEMs are designing product platforms, not one-off devices
Many industrial products develop into families. A robotics manufacturer may need several payload or sensor configurations. A machine-vision supplier may serve different camera counts, performance levels or inspection tasks. An HMI platform may be adapted across multiple machines.
Rebuilding the compute and software architecture for every variant increases engineering effort, validation work and lifecycle cost. It also makes future product generations harder to plan.
A modular approach can change the economics. A reusable carrier concept, a consistent software foundation and scalable compute options can help teams reuse portions of their work across related products. Not every component will transfer unchanged, and every final design still requires application-specific validation. The potential value lies in reducing unnecessary reinvention.
This makes modularity more than a form-factor decision. It can provide a way to help preserve engineering investment, shorten the development of product variants and create a more manageable path for future upgrades.
From trends to platform decisions
These four shifts can give engineering and product teams a practical way to assess an Industrial Edge AI platform.
Can it coordinate the mix of AI, application and time-sensitive workloads required by the complete system? Can it connect to the relevant sensors, networks and control infrastructure without weakening local operation? Can its software foundation be maintained and secured throughout the product lifecycle? Can the architecture be reused or adapted across future variants?
Peak AI performance remains important, but it is only one input. A production platform must bring compute, I/O, software maintainability and modularity together. Evaluating those elements as one system helps expose integration risks before they become expensive redesigns.
A production-oriented foundation for Industrial Edge AI
These design criteria are not only future requirements. Following initial engineering evaluations, SECO is making a limited first batch of SOM-SMARC-Dragonwing-IQ8 engineering samples available to selected OEMs and development teams. SECO is the first company to make an industrial SMARC system-on-module based on Qualcomm Dragonwing™ IQ8 Series available for OEM evaluation.
Powered by the Qualcomm Dragonwing™ IQ-8275 processor, the module is built to deliver up to 40 TOPS of AI acceleration through the Qualcomm® Hexagon™ NPU. Gigabit and 2.5 Gigabit Ethernet, PCIe Gen4, camera interfaces and CAN support are designed to provide the connectivity needed to evaluate vision, networking and application-specific industrial architectures without treating AI compute as an isolated component.
As part of the Works with Arduino™ ecosystem, SOM-SMARC-Dragonwing-IQ8 is structured to support a more continuous path from application development on Arduino® VENTUNO™ Q board to industrial product design. Rather than swapping one board for another, each plays to its strengths: Arduino and Arduino® App Lab aims to accelerate prototyping and validation, while SECO offers carrier-board design, software integration and the long-term requirements of the final product. Clea OS is designed to add a foundation for security-focused device management and long-term software maintenance.
Design for the product lifecycle, not the first demonstration
The 2027 opportunity is not simply to place more AI inside a machine. It is to build intelligence designed to remain supportable as the product evolves. Evaluating the full architecture early may help OEMs reduce redesign risk and reuse more of their engineering work in future systems.
With a limited first batch of SOM-SMARC-Dragonwing-IQ8 engineering samples available, selected OEMs and development teams can assess whether Dragonwing IQ8 technology, industrial connectivity and a lifecycle-oriented software foundation are appropriate for their next Industrial Edge AI or Physical AI product.
Request an engineering sample