<p>This article examines key challenges in computing systems research under the emerging paradigm of Physical Intelligence on the Edge (PIE), in which raw sensor streams are transformed into real-time, safety-critical intelligence that can act in the physical world. It traces the evolution of computing architectures from centralized systems to distributed systems and edge computing, and argues that PIE constitutes a qualitative shift: the edge becomes the primary platform for tightly integrating sensing, reasoning, and actuation under stringent real-time constraints. The article identifies five emerging research thrusts—embodied spatial reasoning, embodied temporal reasoning, edge-native customization, symbiosis, and sustainability. Using a hypothetical PIE scenario, it exposes a fundamental gap between the capabilities of current systems and the requirements of future PIE-enabled autonomy: while today’s edge platforms can execute individual components of perception and inference, they remain unable to autonomously close the sense-think-act loop with certifiable guarantees on timing and safety. This vision is further substantiated by recent industrial progress, including several compelling demonstrations showcased at CES 2026 by leading companies such as NVIDIA and AMD. The article concludes by calling for a paradigm shift in systems thinking—from efficiently transporting and processing data (bits) to predictably and safely influencing the physical world (atoms)—thereby positioning edge-native system design as a foundational enabler of next-generation autonomous and robotic systems.</p>