by Joe Abajian | 05/06/2025

Hardware innovations advance edge AI, but software innovations make it possible. Model optimization technology has brought bulky AI models to low-power devices, expanding the realm of possibilities for AI applications at the edge. Processor-specific optimizations and software toolkits have further enabled edge AI on devices that lack hardware acceleration or high performance CPUs. Thinking back to early February at Embedded World 2025, an annual conference held in Nuremberg showcasing the latest innovations in embedded technology, AI hardware was the star of the show. Several chip and IP providers released new products targeting edge AI.

Regardless of how many TOPS each processor claims to guarantee, however, edge AI software development solutions are necessary to extend the value of embedded AI hardware platforms. Additionally, edge AI software solutions can bring AI functionality to existing systems unable to run full AI models while alleviating the largest challenges facing engineering organizations in developing embedded and edge AI software today.

Hoping to gain an advantage in the AI arms race, several AI software solution providers made waves at Embedded World this year. Qualcomm announced its intention to acquire Edge Impulse, a pioneer of edge AI software development, training, optimization, and deployment for low-power systems. As of now, Edge Impulse will still function independently but announced increased support for Qualcomm Dragonwing processors going forward. The acquisition aligns with efforts from other hardware providers to bolster their AI software offerings. Arm announced that it was making Kleidi, its software initiative designed to enhance AI performance on Arm CPUs, available for AI-based SDV functionality development. Axelera, a Dutch AI processor provider, demonstrated its Voyager SDK’s ability to simplify model optimization and deployment on its new M.2 AI accelerator cards. Embedded World 2025 showed that Edge AI hardware providers have fully embraced AI software tools as a competitive differentiator.

Beyond solutions from hardware providers, several AI software specialists exhibited their ability to bring AI to the edge across hardware form factors. Irida Labs, a Greek end-to-end computer vision development platform provider, showcased its specialized solutions for Industry 4.0 and smart city applications. Enerzai, a Korean edge AI solution provider, announced an expanded partnership with Arm and demonstrated its Optimium AI engine’s ability to make model inference 20x faster without losing accuracy. Edge AI security is also emerging as an important field. Italian startup Accelerat showcased its AI Bunker offering at the event, which is the company’s new AI model storage, isolation, and security solution designed to prevent model IP theft. As the edge AI software landscape has matured, demand for specialized development solutions and services has increased.

Looking ahead, we believe consolidation in the AI software ecosystem will continue over the near term. Growing model complexity will incentivize hardware vendors to abandon their existing AI software development toolkits in favor of acquiring a more functional software suite from an AI development solution provider. Many engineering organizations have struggled to hire and retain AI/ML developers, making specialized or premium AI development tools, expertise, and services more valuable. Qualcomm’s acquisition of Edge Impulse was not the first domino to fall, and it won’t be the last. Recent market innovations  prove that AI model development and optimization tools are an essential piece of the intelligent edge puzzle.

To learn more about VDC’s Embedded AI Development Solutions report, read our Executive Brief here.