Since 2025, Nvidia has made "Physical AI" a recurring theme across keynotes and conferences, positioning it as a foundational pillar for the next phase of artificial intelligence. Among its most promising applications: autonomous vehicles. With the launch of its Cosmos platform, Nvidia is now translating its AI vision into real-world commercial deployment—starting with the automotive industry.
Understanding physical AI
Physical AI refers to systems that mimic human-like reasoning based on logic and an understanding of the physical world. For example, if a self-driving car detects a pedestrian who briefly disappears behind a vehicle, it must infer that the person may still be present and could re-emerge at any moment—or conclude, based on context, that the pedestrian has turned away.
Such inferential thinking, rooted in physics and common sense, is the cognitive framework that Physical AI seeks to emulate. And it's this leap—from perception to prediction—that Nvidia believes will define the next frontier of AI.
Automakers use AI for smarter data and faster development
Nvidia's Cosmos platform is already being used to reshape how automakers manage data and accelerate the development of advanced driver-assistance systems (ADAS). General Motors (GM), for instance, deepened its collaboration with Nvidia in 2025, extending its partnership across AI, simulation, and accelerated computing.
GM is integrating Cosmos into its Super Cruise ADAS system, aiming to advance toward higher levels of autonomy. Yet GM remains one of the few automakers to publicly adopt the Cosmos platform, with many others still in the experimental phase. Industry sources say concerns remain over whether a single platform can effectively serve the wide range of vehicle types and driving environments.
Still, for those using it, Cosmos offers a key advantage: the generation of synthetic data. This capability allows companies to dramatically expand and diversify training datasets using only a small number of real-world samples—enabling faster, more representative model training while helping automakers close the performance gap with more advanced competitors.
Autonomous service providers move faster
While traditional automakers proceed with caution, autonomous vehicle (AV) companies are embracing Cosmos with urgency. Several firms—ranging from AV software startups to autonomous trucking companies and ride-hailing platforms—have publicly committed to the platform.
Nexar, for example, uses Cosmos in tandem with real-world driving data to train its models, while Plus is leveraging the platform to enhance its SuperDrive software for self-driving trucks. Other adopters include OXA, Foretellix, Parallel Domain, Uber, and Waabi—each seeking to bring self-driving technology to market more quickly and reliably.
DIGITIMES analysis suggests that Cosmos is especially well-suited to AV operators due to the centralized nature of their fleets and data systems. Smaller, more controlled operating domains allow these companies to more directly benefit from Cosmos' optimization and reasoning capabilities—cutting development time and speeding up commercialization.
The future of AI moves beyond the cloud
During his keynote at Nvidia's GTC 2025 conference, CEO Jensen Huang called Physical AI the most important evolution in the AI landscape today. He described a future in which AI systems not only process digital information but also interpret and act upon the real world—a shift from passive intelligence to embodied capability.
"Physical AI is about machines that understand the world and take action in it," Huang said. "It's the foundation for autonomous systems in cars, robotics, and beyond."
Ultimately, Physical AI represents more than a new category of artificial intelligence—it signals a new trajectory for the field itself. By connecting reasoning with action, it bridges the gap between digital computation and physical operation.
For automakers, this means stronger data asset management and deeper technological integration. For AV companies, it offers a path to rapid system refinement and market readiness. Together, these applications form a powerful momentum—one that could accelerate the global development of autonomous mobility.
Article edited by Jerry Chen
