Physical AI Strains Computing Infrastructure to Its Limits

Physical AI is forcing a fundamental rethink of computing infrastructure as robots, autonomous systems and intelligent devices demand economical inference, enhanced security and capabilities beyond conventional cloud systems. Companies like Rafay Systems are addressing costs through shared GPU orchestration, while Positron AI focuses on tokens per dollar and watt as key efficiency metrics. As agentic workloads consume exponentially more tokens than traditional chat applications, infrastructure demands are intensifying across factories, vehicles and distributed environments. Firms including Axiado, Liquid AI and Protopia AI are developing silicon-level management, compact edge models and data protection tools to support AI moving deeper into the physical world.