‘AI at Edge’ was the key theme for the Embedded World 2025 show, held in Nuremberg, Germany, from March 11 to 13. All the key MCU/MPU players showcased solutions that had on-device AI for low-power computing. We saw interesting use cases for object detection and sensor fusion being run on these MCUs/MPUs without the need for cloud processing, enabling low latency and reduced bandwidth along with enhanced privacy and security.

Parv Sharma, Senior Analyst at Counterpoint Research, and Taimur Zafar, Research Analyst at Counterpoint Research, had a quick chat with Vitali Louiti, Senior Director of Segment Strategy, Product Management, at Imagination Technologies, on the sidelines of Embedded World 2025. The company focuses on power-efficient solutions for GPUs, particularly for RISC-V designs and edge use cases. We discussed new GPUs, Edge AI and other key themes from Embedded World 2025.

The Interview

Key takeaways from discussion

Introduction to Imagination Technologies

• Imagination Technologies is a GPU IP supplier with billions of units sold. Its partners include SoC makers in segments like mobile, automotive, PC and consumer IoT.

• The company is positioned as the go-to GPU provider for RISC-V designs.

• Imagination focuses on edge use cases requiring power efficiency and combines this with the trend of artificial intelligence.

• The company’s DXTP GPUs are highlighted as being 20% more power-efficient than previous designs. Future designs will continue this focus.

Heterogeneous compute for Edge AI

• The integration of AI workloads with the GPU is being considered an integral block for running future ML tasks due to its parallel-computing capabilities, unlike CPUs that are not deemed capable of taking up advanced workloads.

• Running AI on GPUs offers advantages in performance and flexibility compared to CPUs, and in some cases, it can negate the need for a dedicated NPU, especially in entry-level segments where cost is a concern.

• GPUs also offer flexibility in adapting to new AI requirements, unlike NPUs which may need to be redone.

Key takeaways from Embedded World 2025

• Loads of innovations were seen across various areas. AI was a core theme, with a focus on running AI on end devices to avoid the cloud.

• There was a notable presence of European companies with interesting use cases and advancements in niche areas like neuromorphic computing and robotics.

• A significant trend observed was localization, with companies emphasizing design and manufacturing within Europe due to geopolitical factors.

• Another interesting development was the combination of UWB and Wi-Fi 7 in enterprise solutions, enhancing connectivity, location tracking and security.

Analyst take

• ‘AI at Edge’ is being integrated without affecting existing experiences.

• There is a hybrid approach for deploying AI based on heterogeneous compute.

• Power efficiency versus performance is a key metric for the edge. Multiple players showcased innovative solutions for running these AI and ML tasks at the edge using either CPU, GPU or NPU.

• The level of complexity dictates the need for dedicated AI accelerators. Heterogenous computing is important for low-power edge AI use cases and there are multiple players that are already showing AI inferencing being run on CPUs, like ARM Kleidi, Imagination Technologies on its GPUs, and NXP on its NPUs.