At the 2026 edition of the embedded world Exhibition & Conference in Nuremberg, edge AI specialist Ceva, Inc. walked away with one of the show’s top honors. The company’s Ceva‑NeuPro‑Nano neural processing unit (NPU) won the Artificial Intelligence category of the embedded award, highlighting the growing importance of ultra-efficient AI processing at the device edge.
The annual award recognizes standout innovations in the embedded systems ecosystem and is judged by an independent panel of industry and academic experts. In this year’s AI category, the judges spotlighted NeuPro-Nano for packing meaningful AI inference performance into an unusually small and energy-efficient silicon footprint—a combination increasingly critical as AI workloads move from the cloud to low-power edge hardware.
For Ceva, the recognition underscores a broader industry shift: AI is no longer confined to data centers or high-performance processors. It’s rapidly becoming a core capability inside everyday devices—from smart sensors to wearables and industrial equipment.
Edge AI, Without the Power Budget Headache
The Ceva-NeuPro-Nano NPU is built specifically for edge devices where power consumption, silicon area, and cost are tightly constrained. Instead of relying on external compute resources, the processor enables devices to run AI inference locally—reducing latency, improving privacy, and minimizing cloud dependency.
That capability is increasingly important as edge devices are expected to do more with less. Smart cameras must recognize objects instantly. Wearables analyze biometric signals continuously. Industrial sensors interpret environmental data in real time. Each of those tasks requires AI acceleration—but often on chips that must run for months or years on limited power.
According to Yaron Galitzky, executive vice president of Ceva’s AI division, the goal is to make practical AI possible even in the smallest embedded systems.
“NeuPro-Nano enables a new generation of devices that can perceive, understand, and act on real-world data locally,” Galitzky said in the company’s announcement, describing the architecture as a foundation for emerging “physical AI” applications in power- and cost-sensitive products.
In practical terms, that means edge devices can interpret sensor inputs—vision, audio, motion, or environmental signals—without sending raw data back to the cloud.
What Makes NeuPro-Nano Different
While AI accelerators have become common in smartphones and PCs, designing one for tiny edge chips is a different challenge entirely. The Ceva-NeuPro-Nano architecture aims to strike a careful balance between compute performance and silicon efficiency.
Key capabilities include support for modern neural networks used in:
- Computer vision workloads
- Speech and audio processing
- Sensor fusion
- Contextual awareness applications
These capabilities allow devices to interpret multiple data streams simultaneously—for example combining camera input, microphones, and motion sensors to understand user behavior or environmental conditions.
The NPU’s small silicon footprint also makes it easier for chipmakers to integrate AI acceleration into cost-sensitive designs. That’s a critical requirement for large-volume markets like consumer IoT, smart home products, and industrial monitoring systems.
Part of a Broader Edge AI Platform
NeuPro-Nano sits at the smallest end of Ceva’s NeuPro lineup, a scalable portfolio of neural processing units aimed at different tiers of edge computing performance.
Across the NeuPro family, AI acceleration ranges from tens of GOPS (giga operations per second) up to hundreds of TOPS (tera operations per second). That spectrum allows semiconductor companies to choose architectures suited to everything from ultra-low-power sensors to high-performance AI systems.
The strategy mirrors a broader industry trend: scalable AI architectures that span multiple device categories. Rather than developing separate solutions for each market segment, chip designers increasingly rely on flexible IP blocks that can be scaled up or down depending on power and performance requirements.
For companies licensing AI processor IP, that flexibility can significantly reduce development time while ensuring compatibility with evolving machine learning models.
Rising Demand for On-Device AI
Ceva says adoption of its NeuPro technology accelerated in 2025, with 10 customers licensing the architecture across markets including:
- Consumer IoT devices
- Industrial automation systems
- Automotive platforms
- Infrastructure equipment
- PC applications
This surge reflects growing demand for on-device AI inference—a computing approach where machine learning models run locally rather than relying on centralized cloud processing.
The shift is driven by several factors.
Latency is one. Applications like real-time vision recognition or predictive maintenance require immediate responses that cloud processing can’t always deliver.
Privacy is another. Keeping data processing on the device reduces the need to transmit sensitive information over networks.
Power efficiency also plays a role. Constantly transmitting data to the cloud can drain battery life in edge devices, making local AI inference far more practical.
As a result, edge AI silicon has become one of the most competitive segments in semiconductor design, with companies racing to deliver specialized processors optimized for machine learning workloads.
The Embedded Systems Industry Is Betting on AI
The recognition at embedded world also highlights how deeply AI has become integrated into the embedded systems industry.
Historically, embedded processors focused on deterministic workloads—control loops, signal processing, or device management. But modern embedded platforms increasingly incorporate machine learning capabilities, enabling devices to interpret data rather than simply process it.
That shift is reshaping chip architectures, development tools, and software ecosystems across the sector.
Industry analysts expect edge AI deployments to accelerate sharply over the next several years as semiconductor vendors release more specialized NPUs, and as machine learning frameworks become easier to deploy on small devices.
For companies like Ceva that specialize in licensable silicon IP, the opportunity is substantial. Instead of manufacturing chips directly, Ceva provides processor architectures that other semiconductor companies integrate into their own designs—a model similar to the way ARM popularized CPU licensing.
Winning a high-profile industry award doesn’t automatically translate into market dominance, but it does signal that the company’s technology is resonating with designers working at the front lines of embedded innovation.
Why This Matters
Edge AI is entering a phase where efficiency matters as much as raw performance. While data-center accelerators dominate headlines with ever-larger AI models, the next wave of intelligent computing may actually occur in tiny devices scattered across homes, factories, and cities.
If processors like NeuPro-Nano can deliver meaningful AI capabilities within tight power and cost limits, the result could be a massive expansion of intelligent edge hardware—from autonomous sensors to context-aware consumer devices.
That vision—millions of devices interpreting the world around them in real time—is exactly the future Ceva hopes to enable.
And if the recognition at embedded world is any indication, the industry is paying close attention.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
