Overview
In today’s industrial sector, from employee cafeterias to boardroom meeting rooms, new technologies such as generative artificial intelligence, large language models (LLM), agents, agent AI, and edge AI are widely discussed, heavily promoted, and even raise concerns.Embedded AI, as another form of industrial AI, is gradually gaining attention in the field of industrial automation, albeit in a more low-key and unobtrusive manner.
Embedded AI has entered the practical application stage in the field of process industrial sensors, supporting smarter and more autonomous operations. Vendors claim that smart sensors can achieve more complex real-time analysis than currently possible and can perform self-calibration tasks — whether connected or not.
By conducting local analysis on the device side, embedded AI in sensors will help the process industry achieve smarter, faster, and more autonomous operations, and its importance will increase accordingly.
Current “Smart” Sensors
Although sensors equipped with embedded AI are still in the development stage, smart sensors have already captured a significant market share, and the market size continues to grow steadily.According to the ARC “Smart Sensor Research Report,” the annual growth rate of the smart sensor market is 6%, far exceeding the overall sensor market growth rate.
But what exactly is a smart sensor? Does every smart sensor have AI capabilities and can perform advanced data analysis? The answer is not absolute. In industrial practice, the definition of a smart sensor depends on two key capabilities:
Self-diagnosis and status reporting: For example, if the lens of a photoelectric sensor is contaminated, it can directly signal the operator.
Remote configuration via electronic interfaces: For instance, when installing a new sensor, the operator does not need to configure it from scratch; they can simply migrate settings from the replaced sensor.
So why do most smart sensors not have advanced diagnostic and analytical capabilities built-in? The reason is often related to the application scenario.For most “traditional” application scenarios, uploading sensor data to gateways, industrial computers, monitoring and data acquisition systems (SCADA), or dedicated data platforms, or even directly to the cloud for analysis or other tasks, remains a simpler, faster, and more cost-effective approach. This method not only meets most industrial needs but also simplifies device cluster management.
Embedded AI in Sensors
Embedded AI refers to data processing occurring locally on the device’s own chip — regardless of whether the device is connected to external cloud or edge infrastructure.
Embedded AI allows “smart” decisions to be made directly at the data source (i.e., the sensor) without the need to transmit data to external computing units or the cloud. Its goal is to make pre-trained machine learning (ML) models sufficiently compact, fully reliable, and resource-efficient so that they can run directly on the microcontroller unit (MCU) inside the sensor. In this context, the term “smart” is somewhat misleading, as it does not refer to adaptive systems but rather to fixed reasoning logic. Nevertheless, embedded AI remains a crucial part of driving process monitoring towards decentralization and autonomy.
In scenarios where hard real-time decision-making, limited connectivity, or energy efficiency is critical, the value of embedded AI in process sensors is most pronounced — it enables immediate local inference without relying on edge or cloud infrastructure.
The development of embedded AI primarily focuses on inference rather than training. This means that while sensors can utilize large datasets and advanced algorithms to make real-time decisions based on their embedded models, the actual training of the models occurs elsewhere. After training, the models are compressed and then deployed to the sensors, a process that may complicate subsequent development and maintenance.
The separation of training and inference presents unique challenges. Before deploying the compressed model to the sensor, data collection, processing, and modeling must be completed externally, requiring a well-defined workflow and careful integration (i.e., matching the model with the hardware, primarily with application-specific integrated circuits (ASIC) or microcontroller units (MCU)). This approach enables rapid, autonomous decision-making on-site — a feature that is increasingly valuable in industrial environments where low latency and high reliability are critical.
Implementation and Impact
Despite the inherent limitations of embedded AI systems (especially in terms of computational power and updatability), recent technological advancements are opening up new application possibilities. Innovations in chip design have led to more powerful and energy-efficient MCUs and ASICs that can be seamlessly integrated into compact sensor packages. For the latest developments in embedded systems, it is recommended to follow events such as the 2026 Embedded World exhibition in Germany.
Regulatory requirements and operational expectations are driving the market demand for smarter, more autonomous systems. Smart sensors with local inference capabilities can not only help the process industry meet these demands but also enhance overall resilience and efficiency.
Currently, embedded AI in sensors is mainly limited to research laboratories or pilot projects, but the first sensors equipped with this technology have already been launched in the market. In industries such as oil and gas, chemicals, water treatment, pharmaceuticals, and power generation, AI-equipped sensors can directly detect anomalies and perform calibrations, thereby optimizing processes while enabling predictive maintenance. Most importantly, many of these applications can operate independently without relying on external infrastructure — highlighting the value of local intelligence in environments where connectivity may be limited or unstable.
Despite various limitations, its advantages are still significant: lower latency, higher system stability, and greater operational autonomy.
Recommendations
Although technological limitations still exist — particularly in computational power and ease of model updates, ongoing advancements in hardware and model compression technologies are rapidly expanding the capabilities of embedded AI. As the industrial sector’s demand for smarter, more efficient systems continues to grow, embedded AI is expected to become an essential component of modern process automation.
For industry stakeholders, the core message is clear:Embedded AI is a new form of AI technology that deserves attention, and it has already demonstrated value in applications beyond the process industry. As part of a comprehensive, multi-layered AI strategy, embedded AI solutions will play a key role in shaping the next generation of process automation. ARC anticipates that more sensors equipped with embedded AI will be launched in the coming years.
For more information or inquiries regarding ARC Advisory Group please contact:
Rita Liu
General Manager
ARC Advisory Group China
For inquiries please contact:
Rita Liu
General Manager
ARC Advisory Group Chin
For inquiries please contact:
Rita Liu
General Manager
ARC Advisory Group Chin
For inquiries please contact:
Rita Liu
General Manager
ARC Advisory Group Chin