Gasgoo Munich- "What is the ultimate form of the automobile?"
In 2015, as Li Xiang embarked on his journey to build cars, he posed this question to himself and his team.
Eleven years later, the new Li Auto L9 launched as a "flagship embodied intelligent SUV." Li Xiang stated on Luo Yonghao's Crossroads that autonomous driving is the first half of embodied intelligence. He added that general humanoid robots are the second half. That answer from a decade ago is moving from vision to reality.
Image source: Screenshot from Luo Yonghao's Crossroads
While most automakers are anxious about Level 3 autonomous driving, Li Auto has mapped an evolutionary path toward general humanoid robots. It uses a comprehensive "two-halves" strategy.
Redefining the Boundaries of the Embodied Intelligence Industry
Public understanding of embodied intelligence is often limited to "humanoid robots." The Equipment Industry Development Center under the Ministry of Industry and Information Technology notes that embodied intelligence refers to intelligent agents. These include robots, drones, and intelligent vehicles. They interact with their environment in real time through physical entities. This achieves integrated perception, cognition, decision-making, and action.
Li Xiang offers a more concrete definition: "Take any machine in the physical world, give it sensors, models, and processors—essentially eyes, a brain, and a heart—and enable it to work autonomously. That is embodied intelligence."
By this definition, a car equipped with advanced autonomous driving capabilities is inherently a typical form of embodied intelligence. It has eyes to perceive the environment, a brain to make decisions, a heart to supply computing power, and limbs to execute operations. Cars and robots are, in essence, different forms of the same species.
Based on this understanding, Li Xiang has proposed a strategic framework of remarkable clarity.
He divides the first half—autonomous driving—into three stages. The first stage (2018–2023) features L2 driver assistance. The second stage (2023–2028) brings L3 conditional autonomy. The third stage (2028–2033) targets L4 high autonomy. China's auto industry is at a critical juncture. It is transitioning into this second stage.
The second half—general humanoid robots—is similarly divided into three phases. The first phase (2030–2035) will possess generalization capabilities equivalent to a 6-year-old child. The second phase (2035–2040) will reach the level of a 12-year-old. The third phase (2040 to around the realization of AGI) will match an 18-year-old adult.
The depth of this framework lies in its revelation that autonomous driving and general humanoid robots are not parallel tracks, but a progression. The core technologies in perception, decision-making, and control accumulated in the first half will serve as the technical foundation for general humanoid robots. In other words, today's investment in autonomous driving is paving the way for the era of general humanoid robots that arrives after 2030.
For the auto industry, this suggests that competition over the next decade will no longer be a simple battle of sales. Instead, it will be about who can successfully migrate their accumulated autonomous driving technology to broader embodied intelligence products. Winning the first half determines the qualification to enter the second.
An Organizational Leap from Building Cars to "Building Humans"
If it were merely proposing concepts, Li Auto would be just another automaker with ideas. What truly rattled the industry was Li Xiang's choice of the hardest path. It involves full-stack in-house development of five core technologies: perception, models, chips, operating systems, and actuators.
Behind this choice lies a judgment about where the technological high ground lies. In 2022, when Li Auto initiated its in-house chip project, the industry consensus was clear. Automakers need not spend heavily to make chips. They could simply buy from NVIDIA or Qualcomm. But Li Xiang's insight ran deeper. "The high ground of future competition lies in the joint design of chips and models."
Image source: Li Auto
Four years later, that judgment has been validated. The new Li Auto L9 Livis is powered by dual in-house Mach M100 chips, delivering 1,280 TOPS individually and 2,560 TOPS combined. Using a data-flow architecture, they are designed specifically for AI-native applications. A paper on this chip was accepted by ISCA 2026, a premier international conference on computer architecture. Li Auto is the first global automaker to have a paper accepted in the conference's industrial track.
This same logic of in-house development extends to perception, chassis, operating systems, and other components. More importantly, Li Auto has established a technological rhythm of "researching one generation while mass-producing another."
Take computing platforms as an example: The original Li Auto L9 from 2022 was equipped with the most powerful computing available globally at the time. Four years later, the infotainment in those vehicles remains smooth. They receive every new feature. Their intelligent driving capabilities remain in the first tier. This is the greatest difference brought by intelligence. Immense computing redundancy gives the vehicle potential for growth and vitality.
In 2025, Li Auto's R&D spending reached 11.3 billion yuan, with AI-related investments accounting for 50%. In 2026, that figure is expected to remain around 12 billion yuan. This spending isn't just about building a car; it is about constructing a technological foundation that can extend to general humanoid robots.
Full-stack in-house development requires organizational capacity to sustain it.
In early 2026, Li Auto overhauled its R&D system. It moved away from dividing departments by software and hardware functions. The company restructured around the concept of "building silicon-based humans." R&D is now split into four systems: the Organ System (chips + datasets + operating systems), the Brain System (perception + models + infra), the Software Body (agents + toolchains), and the Hardware Body (robot hardware). An independent evaluation team was also added to assess the work quality of agents and robots, acting as an "immune system."
The adjustment avoids the imbalance of "a powerful brain paired with weak limbs." It also reduces inefficient internal friction where "every application wants to train its own model."
The results of this transformation are already visible: the training iteration cycle for intelligent driving models has shrunk from two weeks to one day. This is not merely a management optimization, but a substantial boost in organizational capacity for technical iteration efficiency.
This also means that competition in embodied intelligence is not just a battle of technology, but of organizational capability. Whoever can switch their R&D system from "car-building logic" to "human-building logic" first will gain the advantage in bridging the first and second halves.
The Second Half Has Arrived, and Capability Migration is Crucial
The debut of the new-generation Li Auto L9 marks a milestone in the first half. Simultaneously, Li Auto is preparing for the general humanoid robot market of the second half.
Li Auto estimates that general humanoid robots will enter the early market around 2030—a timeline comparable to electric vehicles in 2015, representing a historic window moving from zero to one. Li Auto has initiated two robot products. One is a wheeled robot for factory and commercial scenarios. The other is a humanoid robot focused on hardware control precision and durability.
Li Xiang is equally clear about the evolutionary path of robots: "The first major battlefield for humanoid robots, the move from zero to one, is essentially combining transport with pick-and-place. In the past, factories used AGVs for transport while humans handled picking and placing. In the future, a simple arm will suffice to do both."
The broader picture is that there will be a 90% overlap between future buyers of L4 autonomous vehicles and buyers of home service robots. "Looking ahead ten years, buyers of L4 autonomous cars and housekeeping robots will be the same group. This applies to 90% of them," Li Xiang said.
In other words, the user base cultivated in the first half will naturally extend into the second half. This trend gives the capital market a clear judgment basis. The wait-and-see period where "both halves seem distant" is over. We are at an inflection point. The "first half is accelerating while the second half is brewing."
Image source: Li Auto
Li Auto proposed in 2015 that "the ultimate form of the automobile is a robot." With the 2026 launch of the new Li Auto L9, it has spent eleven years turning this "science-fiction concept" into reality.
This time, Li Xiang has again demonstrated foresight by dividing the two halves of embodied intelligence into three clear stages, each with defined technical goals and timelines. This dual drive of "definition plus execution" allows Li Auto to stay on the correct path for the next decade, even as the rest of the industry chases short-term gains.
For the entire mobility and robotics industry, embodied intelligence is no longer a distant technological vision but a tangible force reshaping industry boundaries. The automobile is the most mature and scaled carrier of embodied intelligence today. It is the first to complete the full-stack closed loop of perception, decision-making, execution, and operating systems. Whoever first masters the full technology stack of embodied intelligence in this "first form"—the car—will hold the power of definition in the era of general humanoid robots after 2030.
In the past, capital markets evaluated automakers based on sales, gross margins, and production capacity. But today, as automakers accelerate their intelligent evolution, the automobile can no longer be defined by traditional industrial manufacturing logic. It must be examined as a frontier technology product.
As the industry transitions toward embodied intelligence, evaluation metrics should shift to "capability base transferability." This means checking if chips, models, and systems from the first half can extend to general humanoid robots. This is not merely technology reuse; it is a fundamental substitution of industrial logic.
This is no longer just a story about cars. It is a long-term narrative about how "silicon-based life forms" move from industrial and mobility scenarios into homes and society. The "two halves" defined by Li Auto are providing a referential blueprint for the entire Chinese intelligent industry. And the first chapter of that blueprint has only just begun.
