On October 21, we learned that during the 2026 Physical AI and World Model Conference, Momenta CEO Cao Xudong pointed out that world models are accelerating the "Intelligent Driving Moore's Law", and predicted that 2028 will be a critical turning point for the large-scale deployment of Level 4 autonomy. At the same time, the growth in data scale provides support for model iteration. Momenta has cumulatively delivered over 1.3 million vehicles and accumulated nearly 2 billion kilometers of driving data. The first 100,000 units took around 24 months to deliver, while the latest 100,000 vehicles took less than a month. The accumulation of driving experience is accelerating. Cao Xudong summarized the trend of every two years doubling the safety and performance of autonomous driving as the "Intelligent Driving Moore's Law," and he believes this rule also applies to broader large models and artificial intelligence fields. He judged that from 2026 to 2028, leading industry companies will achieve a tenfold increase in safety and performance annually. The world model, data scaling, and model scaling are the key factors driving this acceleration. On the world model level, Momenta's R7 world model learns physical laws through predictive learning, combines driving data and post-training, grasps driving knowledge, aligns with excellent driver behavior, and improves defensive driving ability through simulation and reinforcement learning for long-tail scenarios, reducing iteration costs. The generalization capability of the R7 world model also provides support for cross-model and cross-region applications. According to Mr. Cao, after massive data pre-training, models only need to supplement a small amount of new vehicle data or local driving data for post-training to adapt to different vehicles and traffic rules and driving habits in overseas markets. Currently, Momenta has validated the generalization capability of the model in various scenarios such as production passenger vehicles, Robotaxis, Robotrucks, and Robovan. The expansion of model size requires chip support. Mr. Cao introduced that Momenta's "brother company", XHEART (Xinxin Hangtu), defines chips based on the needs of large models, and it is expected to improve performance by about ten times at the same cost and power consumption, with the next X9 chip supporting a 100-fold expansion of edge model sizes. Through the soft and hard collaboration between large models and chips, driven by model needs to design chips, supported by chips to expand model sizes, combined with the continuous growth of production data and evolving world models, the Intelligent Driving Moore's Law is accelerating its realization. At the same time, Mr. Cao judged that 2028 would also be a critical turning point for autonomous driving, from Level 2 to Level 4, with safety continuously improving and opening up a larger market space. (Autohome Zhou Yi) (Autohome Zhou Yi)