物理人工智能，第3部分：移动性的未来

Physical AI, part 3: The future of mobility 26March 2026 Key takeaways •Physical AI is transforming mobility, shifting vehicles from software‑defined machines to AI‑defined platforms with reusableintelligence that spans autos, trucks, drones and industrial equipment. This marks a structural transition where anything thatmoves is becoming autonomous. •Robotaxis, autonomous trucks and drone networks are accelerating toward commercialization, supported by falling hardwarecosts, maturing regulation and scalable software infrastructure. These advancements are reshaping value chains, enabling newbusiness models and expanding multi‑billion‑dollar addressable markets. •In this third and final part of our physical AI series, we spotlight key trends that are gaining rapid commercial momentum andredefining the future of mobility. Physical AI and the future of mobilityToday, physical AI is reshaping mobility, from AI-defined vehicles to highly automated robotaxis and fully autonomous fleets. Those same capabilities are rapidly spreading to trucks, delivery robots and a growing class of machines, signaling a structuralshift: anything that moves is going autonomous. Vehicles represent one of the first scaled deployment environments for physical AI–where perception, world modeling, planningand safety-critical execution converge. While fully autonomous vehicle (AV) technologies are beginning to scale, higher volumedeployments of physical AI in mobility in the near term are driver assistance technologies and reusable software infrastructure.In turn, this enhanced software, when paired with increasingly capable high-powered compute, is the key enabler for higherlevels of autonomy. Future car: From software-defined to AI-definedThe auto industry is moving beyond software-defined vehicles toward AI-defined platforms, whereintelligencerather than code shapes functionality and differentiation. Why? It enables continuously intelligent, data-driven automotive platforms. Thisadaptability is becoming core to the driving experience because it can be changed and upgraded over time–and not limited towhen the car is manufactured. But the trickle-down effect is real: this shift is rapidly impacting vehicle manufacturers, suppliersand technology providers through hardware and software needs required to enable this shift. The implication? Vehicles are becoming part of a broader physical AI ecosystem linking production, energy and operations into acontinuous automated loop. Start with software, end in autonomyAutos offer something no other industry can match at scale: huge production volumes, strict safety requirements and complex operating environments. All of this makes cars the ideal place to mature and industrialize embodied intelligence long before fullautonomy becomes mainstream. Vehicle intelligence is no longer just an “autonomous driving feature” – it’s evolving into a core layer ofphysical AI infrastructurebuilt directly into the vehicle. This intelligence now includes engineering tools, safety-critical operating systems and in-vehicle AIsoftware – not just driving policy. Once developed, these capabilities can be reused across different models, brands and evenentirely different types of machines. This shift mirrors the broader physical AI landscape, where companies are racing to createintelligence that can be certified, scaled and transferred across domains. But physical AI isn’t just software magic–it’s constrained by real-world engineering and hardware realities. Early choices insensors, compute and wiring set the ceiling for future autonomy and define how much intelligence you can run on the hardwarethe vehicle ships with. Most cars can add new features over time, but they can’t be retrofitted for higher autonomy levels if theylack sensors, compute or fail-safes. Unlocking a shared platformAs mentioned inPhysical AI, part 1: The basics, end-to-end AI could enable a shared intelligence platform across both robots and AVs, since the same approach can train systems to manipulate objects or navigate roads (Exhibit 1). This unification promisesfaster rollout, lower costs and more scalable deployment than traditional modular methods, according to BofA Global Research.However, it also introduces trade-offs–requiring far more data and compute and creating tougher safety-validation challengesbecause errors are harder to trace–prompting companies to add extra models or software layers to compensate. Infographic depicting the shift from rules-based to end-to-end AI single-neural network New revenue, new marketsSoftware-defined vehicles unlock new revenue streams: ongoing feature upgrades, services and fleet-level optimizations delivered over the air. A central compute“brain”orchestrates these capabilities. And the benefits don’t stop with passenger cars.According to Applied Intuition, roughly 90% of what’s needed for safety-critical physical AI applies across trucks, industrialmachines, mining equipment and defense sy