Wu Xinzhou Responds to NVIDIA's L4 and Physical AI Strategy

Wu Xinzhou's First Comprehensive Response: L4 Is Indeed Getting Closer

Wu Xinzhou, NVIDIA's Vice President of Autonomous Driving and Robotics, recently publicly addressed the industry's many concerns about L4 autonomous driving and physical AI. He stated clearly: "L4 is indeed getting closer." This assessment stems not only from the objective trajectory of technological evolution but also from NVIDIA's comprehensive readiness in computing platforms, software ecosystems, and data closed-loop systems.

Since joining NVIDIA in 2023, Wu Xinzhou has been regarded as a key figure bridging the autonomous driving industry and the AI chip giant. His remarks have sparked a new round of industry discussion about the L4 deployment timeline.

Physical AI: NVIDIA's Next Trillion-Dollar Narrative

Physical AI is a core concept that NVIDIA CEO Jensen Huang has repeatedly emphasized in recent years. Unlike large language models that focus on understanding and generating text, physical AI aims to enable AI systems to truly understand the laws of the physical world and make decisions and take actions in real environments. Autonomous driving and embodied intelligent robots are the two most critical application scenarios for physical AI.

In his response, Wu Xinzhou further elaborated on NVIDIA's understanding of physical AI. He believes the key challenges of physical AI lie in three dimensions:

• Perception: AI needs to understand three-dimensional space, dynamic objects, and complex scenes just as humans do
• World Models: AI needs predictive capabilities for the physical world, understanding causal relationships rather than merely performing pattern matching
• Decision-Making and Control: AI needs to make safe and reliable real-time decisions in highly uncertain real-world environments

NVIDIA is building a complete physical AI technology stack through the Omniverse simulation platform, the DRIVE Thor in-vehicle computing platform, and the Isaac robotics platform. Wu Xinzhou emphasized, "These three platforms do not exist in isolation but form a complete closed loop from simulation training to real-world deployment."

Why Is L4 "Getting Closer"?

Regarding his optimistic assessment of L4, Wu Xinzhou offered several supporting arguments:

First, the maturation of end-to-end models. Over the past year, end-to-end autonomous driving solutions have achieved significant breakthroughs. Integrated neural networks spanning from perception to planning have dramatically reduced information loss and engineering complexity inherent in traditional modular approaches. Practices from Tesla FSD, Waymo, and multiple Chinese automakers are all validating the viability of this technical path.

Second, exponential growth in computing power. NVIDIA's next-generation in-vehicle chip, DRIVE Thor, will deliver over 2,000 TOPS of computing power, sufficient to support multiple large models running simultaneously on the vehicle. Wu Xinzhou noted that computing power is no longer the bottleneck — the real challenge lies in how to efficiently utilize that power.

Third, the acceleration of the data flywheel. As vehicles equipped with advanced intelligent driving systems hit the road at scale, the industry's accumulation of high-quality driving data is growing exponentially. This data feeds back into model iteration through cloud-based training, creating an ever-faster evolutionary cycle.

Fourth, breakthroughs in simulation technology. The combination of Omniverse and the Cosmos world foundation model enables autonomous driving systems to undergo large-scale, high-fidelity testing and validation in virtual environments, significantly reducing the cost and timeline for L4 verification.

China Market: A Critical Battleground in the L4 Race

Notably, the Chinese market plays a pivotal role in the L4 race. On one hand, China has the world's most complex traffic scenarios, making it the ideal proving ground for testing L4 technology maturity. On the other hand, Chinese automakers' investment intensity and iteration speed in end-to-end intelligent driving are among the world's leaders.

Wu Xinzhou previously served as Vice President of Autonomous Driving at XPeng Motors and has a deep understanding of China's intelligent driving market. In his response, he also acknowledged that the fierce competition in the Chinese market is pushing the entire industry to accelerate toward L4. NVIDIA hopes to play the role of "infrastructure provider," empowering global partners through open platforms.

Challenges Remain

Despite the optimistic outlook, Wu Xinzhou also candidly pointed out the key challenges that L4 deployment still faces:

1. Long-tail scenario coverage: The last 1% problem in autonomous driving remains thorny — extreme weather, rare traffic events, and other long-tail scenarios require stronger generalization capabilities
2. Safety validation frameworks: How to prove with industry-accepted methodologies that L4 systems are safer than human drivers still lacks unified standards
3. Regulations and liability definitions: L4 means vehicles drive fully autonomously under specific conditions, and the relevant legal frameworks and liability allocation are still being refined
4. Business model sustainability: Whether the high costs of sensors and computing can be reduced to consumer-acceptable levels after scaling up

Outlook: The Prologue to the Physical AI Era

From a broader perspective, L4 autonomous driving is merely the tip of the iceberg in the grand narrative of physical AI. NVIDIA's ambition is to build a universal physical AI platform that enables intelligence to exist not only in the cloud and on screens but also to permeate every corner of factories, warehouses, homes, and even cities.

Wu Xinzhou's response sends a clear signal: NVIDIA is advancing autonomous driving, robotics, and industrial simulation as a unified physical AI strategy. When computing power, algorithms, data, and simulation all reach their critical points simultaneously, L4 may truly no longer be far away.

For the entire intelligent driving industry, 2025 could be the pivotal year for the leap from L2+ to L4. And NVIDIA is positioning itself to be the "foundational enabler" behind that leap.