It’s not about maximizing compute. It’s about optimizing it.
At Computex 2026 in Taipei, NXP Semiconductors, along with Ajith Mekkoth, executive VP of AI and chip engineering, advocated for “right-sized AI,” arguing that AI adoption should be driven by application requirements rather than the pursuit of ever-bigger models and greater computational power.
“Right-sizing AI means that different types of models have different roles,” Mekos said, emphasizing that many applications can be effectively served by smaller AI models rather than generative AI models. This is a fit-for-purpose approach that focuses on selecting the appropriate model applicable to the current application.
Mekos noted that while cloud-based AI remains relevant for some workloads, “the token cost is too high and the latency is too high” for many use cases. Advances in silicon technology allow large amounts of computing to be performed locally, reducing dependence on cloud infrastructure.
NXP believes the transition to edge AI will accelerate, especially in consumer applications. But Mekos said implementation challenges often have more to do with the overall system design than the AI itself. “We believe this is not just an AI problem, but a system design problem.”
To explain NXP’s strategy, Mekkoth highlighted industrial motor control, voice-enabled human-machine interaction, and robot control as examples where AI can operate efficiently within tight power and latency budgets. He also said that the Kinara acquisition strengthens NXP’s portfolio by adding generative AI capabilities and discrete NPUs that can be combined with the company’s processors to scale edge AI according to customer requirements.
Going forward, Mekos predicts that concerns around data ownership, privacy, and security will further strengthen the case for edge AI as organizations seek greater control over their data and real-time decision-making capabilities.
