Edge AI for IoT systems is a way to cut latency and shape how companies design and run connected systems. Recent signals from chipmakers indicate that this topology is becoming more popular, with more AI workloads handled directly on devices like cameras and embedded systems.
At Embedded World 2026, firms working on edge hardware demonstrated this approach. Among them, Ambarella outlined plans to push more AI processing onto its chips, moving past its roots in camera technology into the broader edge computing market.
In traditional deployments, devices often captured data and sent it to central servers for analysis. That model still works for some cases, but it comes with trade-offs. Sending large amounts of video or sensor data in networks can raise network costs and increase latency metrics. It can also, in some cases, create data privacy issues.
Running AI on-device can change the balance. Devices can process information as it is generated and send, if necessary, any results off-site, reducing bandwidth and improving response times. In industrial environments where machines must react in real time, that distinction should influence system design.
Edge AI shifts cost and system design to IoT devices
Any change is also tied to cost. Cloud processing is not free, and cost tend to be accrued according to the volume of data. As companies deploy cameras and connected equipment, sending data to the cloud becomes harder to justify. Moving AI onto devices can help to reduce OPEX (ongoing compute and storage costs) in large-scale deployments.
Chip design has improved enough to support the model. Processors can handle AI tasks like image recognition and anomaly detection, with some powerful enough to support pattern analysis without the involvement of external systems.
Just such a change is visible in several sectors. Cameras in surveillance systems recognise locally-occurring events and send alerts, rather than constant video for human or off-site processing. In automotive systems, onboard AI helps process sensor data for driver assistance and safety features rather than using unreliable cellular connections. Real-time AI analysis allows machines in robotics and manufacturing to adjust their actions, reducing the need to wait for attenuation instructions from off-site systems.
Industry events like Embedded World suggest that these types of installations are not limited to early technology adopters. Many vendors now offer hardware and software designed for on-device AI, suggesting a mature ecosystem which includes chips and tools to build and manage models at the edge.
The result is a change from hardware components to platforms. Chipmakers are not only selling processors. They are also providing software stacks and development tools, along with support for AI models. The allows companies to build complete systems not piece together separate parts. It also changes how vendors compete, as they move closer to the software layer.
From cloud-first to hybrid AI systems
There are still limits and not all AI workloads run on devices with limited computing power. In many cases, companies will use a mix of edge and cloud systems, choosing to run each task based on cost and necessary speed, as well considerations around the scale of requirements.
Yet Edge AI is starting to become a more common design approach, one not limited to specialised deployments. As devices become more capable, keeping processing close to the source is starting to make more sense. Cloud is going away, but balance is changing. The cloud remains important for initial model training, storing data, and running large-scale analysis. Edge systems handle time-sensitive tasks and reduce the load on central systems.
In practice, this could change how IoT deployments are planned. Instead of designing systems around the data flow to another physical location, companies may start with the assumption that devices will handle many tasks locally. The cloud then becomes a supporting layer.
That change has implications for cost and system design. It also affects how data is managed and governed. It also points to a more distributed model of computing, where intelligence is spread in devices not concentrated in a few locations. Industries that rely on fast decisions and large networks of connected devices may find this model easier to scale over time.
(Photo by Alexandre Debiève)
See also: IoT devices are designed to collect data – edge AI is making them think
Want to learn more about the IoT from industry leaders? Check out IoT Tech Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events including AI & Big Data Expo and the Cyber Security Expo. Click here for more information.
IoT News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.
