The Future of AI: Cars as the New Data Goldmine
As the digital landscape evolves, a new frontier in artificial intelligence looms on the horizon—automobiles. Wang Xiaogang, co-founder and chief scientist of the Chinese AI startup SenseTime, recently made headlines by asserting that when the internet’s data pool runs dry, automobiles will take center stage as continuous sources of real-world data for AI models.
From Internet Dependency to Physical Data Generation
Wang explained how, in this mobile internet era dominated by smartphones, large AI models have heavily relied on internet data. However, he insists that the value of this data has diminished significantly, reaching a critical bottleneck. “The next stage requires generating new data through interaction with the physical world,” he stated, highlighting the transformative potential of cars. Automobiles serve as ideal platforms for generating new forms of intelligence, given their capacity for real-world interactions.
SenseTime’s Vision for Automotive AI
At SenseTime’s Shanghai headquarters, Wang articulated a bold vision, claiming the company is the most automotive-focused AI firm. Dressed in a shirt emblazoned with ‘SenseAuto,’ he metaphorically described their partnership with automakers as a thoroughbred horse galloping alongside them—a synergy aimed at accelerating progress in the sector.
Since partnering with Honda Motor in 2017, SenseTime has extended its collaboration to over 30 car manufacturers, producing more than 3.6 million vehicles equipped with smart technology. This journey underscores the rapid evolution of Chinese AI firms in the smart vehicle arena.
Navigating a Competitive Landscape
The smart driving industry is ferociously competitive. Major players, including Horizon Robotics, DJI, Huawei, and Momenta Technology, are all striving for market dominance alongside automakers like BYD and Xpeng Motors. “Maintaining innovation and technical leadership is paramount,” Wang noted, emphasizing the challenges ahead.
For instance, SenseTime introduced the concept of end-to-end autonomous driving back in 2016, a time when lidar technology was prohibitively expensive. Focusing on a vision-based solution rather than conventional rule-based systems, Wang’s team worked with Honda to create cars that could operate without lane markings or high-precision maps.
The Evolution of Autonomous Driving
Wang believes that cars should emulate human behaviors in driving. By leveraging visual inputs, today’s end-to-end autonomous solutions can accurately predict a vehicle’s trajectory. Despite the obstacles posed by outdated networks and insufficient computing power, SenseTime has emerged as a leader in this realm.
Five years later, the firm introduced UniAD, the first integrated perception and decision-making model in autonomous driving. This innovation won the Best Paper Award at the International Conference on Computer Vision and Pattern Recognition 2023, an industry milestone.
Data Scarcity: A Significant Challenge
A glaring issue with end-to-end systems is their heavy reliance on high-quality data for training. However, Wang pointed out that a mere 1% of human driving data is effective for model training, making data filtering a highly subjective process. To combat this, SenseAuto developed a world model—an extensive mechanism to generate invaluable driving data.
Wang explained, “The key lies in precision and control,” highlighting how inputting a vehicle’s 3D trajectory can produce corresponding video content. This significant advancement overcomes traditional data collection limitations, allowing for the generation of training scenarios for hazardous driving conditions.
Launching Innovative Technology
In November, SenseAuto unveiled the Kaiwu world model, upgraded to a near real-time interactive version by April. Yet, the industry remains divided on the commercialization of advanced technologies like world models, with some experts advocating for a more cautious approach.
Shifting Paradigms: From Passive to Proactive AI
Wang anticipates that intelligence will define the second half of the electric vehicle race. He envisions a shift from passive AI functionalities to ambient computing, where AI develops a human-centric approach that proactively delivers services.
This vision blossomed within SenseTime’s concept of ‘A New Member for U,’ which quickly spurred similar initiatives in the industry. Noteworthy competitors have already started following suit, showing the rapid pace of technological evolution.
The Unique ‘1+X’ Model
Central to SenseTime’s success is its peculiar ‘1+X’ model. The ‘1’ symbolizes robust computing infrastructure and large-scale model capability, while ‘X’ denotes deep applications across various sectors. Wang believes that as cloud infrastructure needs swell, the emphasis will be on simplifying in-vehicle systems, ultimately highlighting SenseTime’s strengths in data generation and analysis.
Investing for the Future
SenseTime’s early investments laid the groundwork for its current success. In 2018, long before homegrown cloud services could support large graphics processing unit clusters, the company established an AI supercomputing facility that now ranks among China’s elite.
When ChatGPT triggered the large language model boom in 2022, SenseTime’s prior investments paid off, granting them a competitive edge in generative AI and intelligent driving applications.
Recent Collaborations: A Step Forward
In March 2025, SenseAuto partnered with Guangzhou Automobile Group to unveil the first Horizon Robotics Journey 6M assisted driving solution, marking a pivotal moment in the AI automotive sector. More solutions based on this innovative technology are set to be released with other manufacturers, showcasing the rapid advancements in autonomous systems.
Cars: The Stepping Stone to Embodied Intelligence
Wang is clear that while cars are not the end goal, they play a crucial role in the journey toward broader ambitions such as robotics. The transition from a mobile internet environment to vehicles and, eventually, to robots, presents exciting opportunities for data-driven AI.
He explained, “The mobile internet’s wisdom has peaked, yet the potential for interaction-driven data remains immense.” Automobiles, as physical entities, are uniquely positioned to gather interaction data that simply cannot be obtained through the internet alone.
Challenges Ahead: From Vehicles to Robotics
While the transition from vehicles to robots offers incredible promise, it doesn’t come without challenges. Wang pointed out that robots come in various shapes and sizes, lacking uniformity. This variability necessitates heightened versatility and robust simulation capabilities.
Laying the Groundwork for Tomorrow
Despite the hurdles, SenseTime is preemptively investing in both upstream and downstream enterprises. Current work in automotive technology also sets the stage for future advancements in robotics, showing how interconnected these industries really are.
Wang’s perspectives reflect SenseTime’s strategic orientation as an AI platform company, adept at navigating the waves of transformation in the AI landscape. As AI evolves across different sectors, SenseTime has strategically positioned itself to leverage this growth.
The Journey of AI: Lessons Learned
Wang’s journey from image intelligence to robotics illustrates the diverse trajectories AI can take. He astutely recognized that AI’s evolution is contingent on the industries driving it forward at any given moment.
Conclusion: Embracing the Future of AI
As SenseTime carves out its niche in the interplay between artificial intelligence and automotive technology, it remains poised for the future. Cars are becoming a linchpin in the transition to embodied intelligence, leading us all towards groundbreaking innovations in robotics. By merging real-world data with cutting-edge AI technology, SenseTime exemplifies how the automotive sector is not just adapting but leading the charge into the next era of intelligent living.
