The Transformation of Four Emerging EV Companies into AI Enterprises

On June 5, 2026, during the Qualcomm Automotive Technology and Cooperation Summit held in Shenzhen, NIO founder William Li made a striking statement: “Today’s car companies must transform into AI enterprises. Today’s smart cockpit must evolve into an AI cockpit.” According to TMTpost, such remarks might have been dismissed as “marketing fluff” three years ago, yet today, no one laughs. This is because, in the past six months, nearly every major Chinese EV startup has been shifting from a traditional carmaker to an AI-focused enterprise.

This transformation goes beyond marketing rhetoric. It encompasses sweeping organizational restructuring, the development of proprietary chips, foundational AI models, and even the mass production of humanoid robots. This article, based on insights from TMTpost, examines how Li Auto, XPeng, NIO, and BYD are steering toward becoming AI enterprises. It also explores the underlying logic of this shift and its implications for the industry.

Li Auto:

Organizational Overhaul and the Logic of “Species”

On January 26, 2026, Li Auto CEO Xiang Li convened an online all-hands meeting and outlined three critical decisions. First, 2026 would be the “last boarding opportunity” to become a leading AI company. Second, Level 4 autonomous driving must be achieved by 2028 at the latest. Third, fewer than three companies worldwide will be capable of simultaneously deploying foundational models, chips, operating systems, and embodied intelligence. Based on these judgments, Li Auto undertook a “bone-deep” organizational restructuring.

Specifically, the traditional autonomous driving division was dissolved, and a foundational model team was reorganized under the leadership of Zhan Kun, a senior algorithm expert in autonomous driving. This team oversees the development of VLA (Vision-Language-Action) models. The software core team integrated autonomous driving and smart cockpit R&D, while humanoid robotics was elevated into an independent hardware line. In essence, Li Auto stopped treating “autonomous driving” as a standalone department, integrating everything under the umbrella of “AI.” The car and robot are seen as carriers of AI, sharing the same foundational recognition model.

On the hardware front, Li Auto’s proprietary cockpit chip, Mach 100, began mass production in May 2026, boasting a computational power of 2,560 TOPS. Xiang Li explained, “Apple excels in user experience not because individual components are the strongest but because all parts work cohesively without weaknesses. Li Auto aims to replicate this paradigm in the physical AI world.” As Xiang Li puts it, Li Auto is moving from creating “mobile homes” to the domain of embodied intelligence, driving a comprehensive overhaul of its R&D systems and organizational structures. The term “silicon-based lifeform” now reflects a mindset of creating “species” rather than merely vehicles.

XPeng: Mass-Producing Humanoid Robots by Year-End

XPeng’s transformation differs from Li Auto’s approach. According to TMTpost, at the company’s global product launch on January 8, CEO He Xiaopeng declared, “2026 will be a pivotal year for XPeng in realizing the commercialization of physical AI. AI will become the core growth engine for our automotive and robotics businesses.”

Three key points stand out. First, the introduction of the second-generation VLA model. While previous smart driving systems relied on a Vision-Language-Action architecture, the language conversion phase caused information loss and delays. This year, XPeng’s second-generation VLA eliminates the intermediate phase, generating motion commands directly from visual signals in an end-to-end process. He Xiaopeng revealed that more than 2 billion yuan was invested in training the model with over 100 million data clips, culminating in the “moment of emergent intelligence” in the second quarter of this year.

Second, the mass production of the IRON humanoid robot. The new IRON robot, standing 178cm tall, features a spine-like structure, pseudo-muscles, flexible skin, and three proprietary Turing AI chips with a total computational power of 2,250 TOPS. According to the latest update in June, the mass-production version has entered the ET2 integration phase, with plans to launch commercially in XPeng’s stores by year-end and export globally starting next year. He Xiaopeng anticipates that “robot hardware revenue and AI model revenue” will become key income sources for XPeng beginning next year.

Third, the mass production of flying cars. The Huitian land-air carrier is on the verge of mass production, with worldwide pre-orders exceeding 7,000 units. In a company-wide meeting, He Xiaopeng stated, “The automobile industry has officially entered the era of cross-domain integration with AI. Smart cockpits and smart driving are now technologically unified, forming super-intelligent entities.” XPeng’s confidence in robotics stems from its expertise in autonomous driving and the advanced capabilities of its “brain.” Its VLA large-scale models, proprietary AI chips, and end-to-end perception, decision-making, and control loops are all integral components of a unified technological stack.

NIO’s Cognitive Cockpit: From Passive to Active

NIO’s William Li captured attention at the Qualcomm Summit not only by discussing the transition to AI enterprises but also by clearly articulating the trajectory for smart cockpit evolution. He stated that AI is reconstructing the next generation of in-car experiences, ushering smart cockpits into the era of cognitive cockpits. The core experience of future smart cockpits will be fully agent-driven.

What is the difference between a “cognitive cockpit” and a “smart cockpit”? Simply put, a “smart cockpit” executes commands given by the driver, whereas a “cognitive cockpit” assesses the driver’s needs and acts autonomously. The former responds passively; the latter provides proactive service. The AI cockpit is the central carrier of this evolution, elevating the experience from “command execution” to “needs understanding.”

Specifically, NIO released version 3.3.0 of its Banyan system this January, upgrading over 60 features. Its foundation lies in the NWM world model, China’s first generative embodied driving model based on a multi-modal autoregressive architecture. This model can simultaneously simulate 216 potential driving scenarios in under 100 milliseconds. Far from being a passive tool that merely executes commands, these cars can now “think.”

Data Reveals the Penetration of Smart Cockpits

According to data from Gasgoo Automotive Research Institute, smart cockpits were installed in 83% of passenger cars in China during the first quarter of 2026 (January–March), reaching 94.5% in new energy vehicles. Today, purchasing a new energy vehicle almost guarantees a smart cockpit—it is no longer optional.

What will the next stage of competition focus on? Gasgoo Automotive Research Institute identifies a shift from competing over “component lists” to “experience reconstruction.” In multi-modal interactions, voice commands have an 87.3% penetration rate, facial recognition around 18%, and gestures approximately 5%. Human-machine interaction is extending from inside the car to outside it, with external voice controls and light projection interactions emerging as new competitive domains.

The most fundamental change lies in the comprehensive integration of large-scale AI models into cockpits. At the 2026 Beijing Motor Show, Alibaba’s Tongyi Qianwen announced that over ten major automakers had simultaneously integrated the “Qianwen” AI assistant into their vehicles. ByteDance’s Doubao revealed that over seven million smart cars equipped with Doubao are now in use across 50 brands and 145 models, facilitating over 30 million daily cockpit interactions. These figures illustrate that smart cockpits have already formed a massive data feedback loop.

The Changing Logic of Competition

Roland Berger’s automotive industry outlook report released in January highlighted the evolving nature of competition in 2026. It identified six key axes, with the final two—“technology wars determine outcomes, AI wars determine superiority”—as the most critical. The report emphasized that automotive product definitions are evolving from traditional transportation modes to “AI-driven intelligent entities.” Cars are no longer merely autonomous vehicles but intelligent entities capable of perception, decision-making, and execution.

Gasgoo Vice President Wang Xianbin put it more bluntly: “The industry is transitioning from ‘software defines cars’ to ‘AI defines cars.’” He summarized the essence of this shift in one word: “proactivity,” where cars sense, judge, and execute tasks autonomously. Through voice, vision, touch, and environmental changes, they instantly understand scenarios and deliver services as if the vehicle itself possesses a “soul.” In other words, competition is shifting from “what this car can do” to “how well this car understands me.”

Editorial Perspective

Short-term impact: From late 2026 to 2027, the mass deployment of vehicles equipped with proprietary AI chips and foundational models by China’s emerging EV makers will dramatically accelerate competition in smart cockpits and autonomous driving. In a market where smart cockpit penetration already exceeds 83%, the next differentiating factors will be the “proactivity” and “understanding” of AI models. Companies such as Li Auto, XPeng, and NIO are driving this differentiation through organizational restructuring and chip development. XPeng’s mass production of humanoid robots could directly influence its automotive AI functionalities.

Long-term outlook: Over the next 1–3 years, this transformation could redefine the boundaries of the automotive industry itself. As Li Auto’s Xiang Li posits, the concept of “silicon-based lifeforms” signifies viewing cars not merely as transportation tools but as physical manifestations of AI. The convergence of autonomous driving and robotics shares a common technological stack that magnifies returns on innovation. Future automakers are poised to strengthen their roles as “AI platform companies,” with revenue streams increasingly derived from AI model licensing, robotic hardware, and data services alongside vehicle sales.

Editorial Question: The AI pivot by emerging EV makers mirrors Tesla’s ambitions for “robotaxis” and “Optimus.” However, such transformations require substantial R&D investments and robust data infrastructure, leaving no guarantee that all players will survive. Readers are encouraged to discern whether these moves by Chinese manufacturers represent “flashy marketing” or genuine technological shifts. Additionally, how should Japanese automakers respond to this trend?

References

• TMTpost - New Energy Automakers Quietly Transforming into AI Companies — Published June 6, 2026