On-Device Agentic AI: Qualcomm's New Mobile Processors

What happened

Qualcomm has unveiled its latest mobile processors, the Snapdragon 8 Gen 3 and Snapdragon X Elite, which are engineered to power the next generation of smartphones and other devices with advanced on-device artificial intelligence capabilities. These new System-on-Chips (SoCs) are designed to facilitate the rise of "agentic AI" by enabling complex AI models to run directly on devices, moving beyond cloud-dependent solutions. The company highlights these processors as central to a new era where AI agents can perform multi-step reasoning and act autonomously on behalf of users, significantly enhancing personal productivity and device interaction. The Snapdragon 8 Gen 3, intended for flagship smartphones, and the Snapdragon X Elite, aimed at premium computing devices, both feature integrated neural processing units (NPUs) optimised for generative AI workloads. This hardware foundation supports advanced on-device processing for large language models (LLMs) and large multimodal models (LMMs), allowing for more responsive, private, and efficient AI applications without constant reliance on cloud connectivity. Qualcomm's strategy positions these chips as foundational for a shift towards more intelligent and proactive personal digital assistants.

Why it matters

This development signals a critical inflection point for AI governance and business strategy, as the shift from cloud-centric to on-device AI introduces new considerations for data privacy, security, and ethical deployment. By enabling powerful AI models to operate locally, the technology can reduce the transmission of sensitive user data to external servers, potentially enhancing privacy protections. However, it also places a greater onus on device manufacturers and software developers to ensure the ethical behaviour and security of AI agents embedded directly into consumer products. For businesses, the advent of agentic AI on mobile platforms presents opportunities for innovative product development and service delivery, but also demands a robust framework for AI governance. Organisations must consider how these autonomous agents will interact with users, handle personal information, and make decisions, ensuring transparency, accountability, and user control. The ability of these agents to perform multi-step reasoning and act on behalf of users necessitates careful design to prevent unintended consequences and maintain trust, making ethical AI design a paramount concern from the outset.

By the numbers

• 2 new Snapdragon SoCs (System-on-Chips) announced.
• 100% of AI processing for agentic AI is designed to occur on-device.

The Institute take

The push towards on-device agentic AI, as exemplified by Qualcomm's new processors, represents more than just a technical upgrade; it's a fundamental re-architecture of how AI interacts with users and data. While the industry narrative often champions the privacy benefits of local processing, the real challenge for responsible AI leaders lies in understanding the implications of autonomous agents operating directly on personal devices.

The shift to on-device agentic AI decentralises control and decision-making, moving it closer to the user but also potentially obscuring the underlying logic and governance. Responsible AI leaders must not be lulled by the promise of enhanced privacy alone; they must demand transparency from chipmakers and software developers regarding the ethical safeguards embedded within these agents. The focus should be on establishing clear frameworks for accountability, ensuring auditability of agent behaviour, and empowering users with granular control over their AI assistants, rather than simply accepting the default settings of powerful, autonomous systems.