TLDR: At The Things Conference 2025, NXP Semiconductors and Edge Impulse (now part of Qualcomm) showcased their differing yet complementary strategies to advance Edge AI. Both companies are tackling the complex challenge of deploying efficient and practical artificial intelligence on low-power, resource-constrained edge devices, addressing issues from model size and accuracy to developer skill gaps and long-term device maintenance.
AMSTERDAM – The Things Conference 2025 in Amsterdam served as a pivotal platform for semiconductor giant NXP and software innovator Edge Impulse, recently acquired by Qualcomm, to highlight their unique contributions to the burgeoning field of Edge AI. Despite their distinct business models, both companies are united in their mission to make artificial intelligence practical and efficient for deployment at the very edge of networks.
Edge AI presents a formidable set of challenges. Models must be sufficiently compact to operate on low-power devices while retaining the accuracy necessary for real-world utility. Development teams face a multidisciplinary hurdle, requiring expertise spanning data collection, digital signal-processing algorithms, machine learning model design, and firmware integration—a comprehensive skill set rarely found in a single engineer. Furthermore, with many edge devices expected to remain in the field for a decade or more, AI models must be robustly updatable and resilient to data drift over time.
Edge Impulse, under the vision of co-founder and former CTO Jan Jongboom (now Senior Director of Engineering at Qualcomm), has been a pioneer in this space since 2017. Jongboom’s early insight that ‘a neural network is just a bunch of matrix multiplications’ laid the groundwork for shrinking models to run effectively on microcontrollers, demonstrating the feasibility of AI on tiny devices.
NXP Semiconductors, a leading chipmaker, is approaching the challenge from a hardware-centric perspective, focusing on enabling advanced AI capabilities, including Large Language Model (LLM) inference, directly on edge devices. Their i.MX-8M+ and i.MX-95 application processors, equipped with on-chip Neural Processing Units (NPUs), are central to this strategy. The i.MX-95, for instance, is designed to handle LLMs with up to approximately 4 billion parameters, catering to automotive, industrial, and smart appliance applications. For more demanding LLMs, NXP has bolstered its capabilities through the acquisition of AI accelerator chip startup Kinara. Complementing its hardware, NXP has also introduced GenAI Flow, a comprehensive toolchain for deploying LLMs and generative AI at the edge. This platform includes a library of functional building blocks such as wake event detectors, speech recognition, and text-to-speech models, crucial for embedded systems that often lack traditional input/output interfaces.
Also Read:
- Generative AI in Automotive Market Set for Rapid Expansion, Reaching $4.58 Billion by 2034
- AI Agents Paving the Way for Smarter Manufacturing: A 2025 Outlook
Both NXP and Edge Impulse exemplify the industry’s concerted effort to overcome the inherent complexities of Edge AI, paving the way for a future where intelligent processing is ubiquitous, decentralized, and highly efficient.
