Gasgoo Munich-Li Auto has always wanted to "get on board."
The AI bandwagon.
Yet on the race to catch that ride, Li Auto seems far from composed, always leaving the impression that it’s just falling short.
That pressure lends a strange air of persuasion and self-justification to the way Li Auto and its founder, Li Xiang, discuss AI.
Li Xiang’s "Career Pivot" From Super Product Manager
This push to prove and persuade began in late 2024.
As the year drew to a close, Li Xiang sat down for a marathon interview, repeatedly insisting that "Li Auto is not a car company, but an artificial intelligence enterprise."
To deliver on that promise quickly, Li Auto spent all of 2025 in overdrive: stripping "Huawei-style" elements from its organizational structure, open-sourcing Xinghuan OS, advancing in-house chip development, and launching its first AI glasses.
But that didn't seem to be enough.
Early this year, Li Xiang called an impromptu all-hands meeting without notice. It lasted nearly two hours. For most of that time, he shared his vision for AI trends and emphasized several key milestones.
Yet when it was over, employees were left with one confused refrain: "I don't understand what the boss is talking about."
That sentiment wasn't isolated. At the time, numerous Li Auto employees took to internal social media platforms to express their confusion or question the meeting's purpose.
Image Source: Li Auto
In other words, at least internally, Li Xiang had yet to fully convince his own staff that "Li Auto is an AI company."
Why?
"Li Auto’s team has always been focused on short-term goals," notes one auto industry analyst. So when Li Xiang suddenly started talking long-term, employees’ first reaction was that it didn't fit his usual "pragmatic" persona.
Li Xiang is indeed pragmatic.
Before 2024, as a product manager known as a "hit-making machine," his decisions consistently hit the market's sweet spot, delivering returns in record time.
That is the pragmatism of a product manager.
But in an era where everyone talks AI, with trends and concepts laid bare, that product manager identity needs a shift. He must become a tech-savvy CTO whom the market trusts.
Li Xiang’s CTO-like transformation was on full display at the Li Auto Software and Embodied AI Launch event, Livis Day.
Image Source: Li Auto
June 15, Beijing.
Standing on stage, Li Xiang set the tone for Li Auto’s next decade. He said that over the past ten years, the company built a "mobile living room"; for the next ten, the goal is to "breathe life" into the car and the home.
Moving from "building the house" to "fine interior design," Li Xiang needs not only recognition for his own shift in identity but also validation of the tool animating Li Auto: its AI capabilities.
The event was dense with information: the in-house Mach M100 chip, the Mach Mind large model, the Mach VLA autonomous driving system, a year-long OTA roadmap, and the promise that safety and efficiency would "surpass humans." There was also an appealing upgrade schedule: a new version in Q3, and capabilities rivaling Tesla’s FSD V14 by Q4.
At a glance, these capabilities don't seem all that remarkable when compared with other automakers offering intelligent driving.
So, what was the point of the event?
Li Xiang’s criteria for holding a launch are clear: "If users would miss out on one important thing by not holding it, then it should happen. If it doesn't provide valuable information, then it shouldn't exist."
Image Source: Li Xiang's Weibo
Measuring the June 16 Livis Day against this standard, simply unpacking technical specs wasn't enough. Li Auto desperately needed a "hardcore tech showcase" to counter the stereotype that it only knows how to make "fridges, color TVs, and sofas."
So, who was this embodied AI launch really for? Did it achieve what Li Xiang wanted?
To some extent, yes — but only just.
The primary audience was Li Auto owners; the secondary, institutional investors and the core algorithm team. It was also a roadshow aimed at talent and capital — a defense against poaching and short-selling.
One early adopter of both the Li Auto L8 and i6 told Gasgoo after the event: "In smart driving, Li Auto is leagues ahead." He added that his third car would undoubtedly be another Li Auto.
Clearly, the event convinced at least one owner. The "not enough" part showed up in the stock price: Li Auto shares fell 1.5% at the close on June 16, and dropped another 3.75% on June 17.
The implication is clear.
Embodied AI That Isn't "Sexy": Old Wine in New Bottles?
But that is understandable. Unlike churning out hit products, tech commercialization takes time. A company used to quick returns must inevitably endure a quiet period.
The question is: In today's market, is Li Auto's concept of "embodied AI" really all that sexy?
To answer that, we first need to define what Li Auto’s embodied AI actually is.
On June 15, the day before the event, Li Xiang posted on Weibo: "Many people came to know Li Auto through 'fridges, color TVs, and big sofas.' Today, please remember this image:"
Image Source: Li Xiang's Weibo
Li Xiang said: "The core difference between an embodied AI vehicle and a smart car is this: protecting human safety, completing tasks independently, and being more efficient than humans. That is our definition of an embodied AI vehicle, and it is what Li Auto must achieve over the next decade."
In other words, unlike XPENG’s humanoid robot walking the catwalk at its previous event, Li Auto’s concept of embodied AI seems somewhat "abstract." Onlookers will need time to accept this less tangible philosophy.
In fact, while ostensibly a tech launch, this was more about rewriting the rules of the game. Li Auto is trying to use the term "embodied AI" to elevate the competition to a new dimension: "whose car has more life."
Li Xiang defines the embodied AI vehicle as a "four-in-one" entity: an EV, a professional chauffeur, an AI computer, and a life assistant. The EV and AI computer are the "embodiment"; the chauffeur and assistant are the "intelligence."
The brilliance of this definition lies in its repackaging: it takes a pile of industry buzzwords and bundles them into a proprietary concept. "Professional chauffeur" is autonomous driving; "AI computer" is the computing base; "life assistant" is cabin AI.
Tesla talks about this; XPENG talks about it; NIO talks about it. But Li Auto has given it a collective name — "embodied AI vehicle" — and declared: this is true intelligence; the rest is mostly "function-driven" intelligence.
It’s like a restaurant boasting "three Michelin-star chefs + organic farm direct supply + temperature-controlled wine cellar + butler service," then slapping a label on itself calling it a "new dining species."
Everything is done well, but together they don't create a new species.
1,280 TOPS of Computing Power: How to Revitalize an Expensive "Brick"?
If the concept of embodied AI fell flat simply because Li Auto missed the beat on communication, what about its heavy-hitting tech? Is that impressive enough?
Take the Mach M100 chip. The specs are undeniably solid.
In May, the Mach M100 entered mass production, becoming the world's first mass-produced dynamic data-flow AI chip. Built on a 5nm automotive-grade process, it delivers 1,280 TOPS of single-chip computing power with an actual operating efficiency exceeding 82%. Compared to the industry-standard Nvidia Orin-X at 254 TOPS, the Mach M100 boasts significantly higher compute density. The new Li Auto L9 Livis, equipped with dual Mach M100 chips, achieves a total computing power of 2,560 TOPS.
Li Auto announced it has achieved full-stack in-house development spanning chips, compilers, operating systems, AI algorithms, and domain controllers.
But behind those specs lies a brutal truth: the core of AI is data, and Li Auto’s data scale is falling behind the market leaders.
Tesla’s FSD V14 is backed by millions of vehicles on the road, uploading real-world driving data daily. Huawei’s ADS relies on a vast matrix of partner automakers.
By comparison, Li Auto’s assisted driving mileage data still lags by orders of magnitude. Catching up in six months is a massive time crunch.
But Li Auto has its own solution: using "computing power" to compensate for the lack of "data."
Image Source: Li Auto
The true value of the Mach M100 isn't the number 1,280 TOPS; it’s the ability to define its own data pipeline. No longer constrained by third-party chip iteration cycles, Li Auto can customize its computing architecture to fit its data needs.
Training world models through synthetic simulation data is the Mach M100’s raison d'être. It isn't built to handle today's L2+ driving, but to power tomorrow's high-density simulations.
This is an expensive strategy of "overtaking on a bend." In essence, it admits this: the speed of physical sales expansion cannot keep pace with the demand for data.
But the risks on this path are extreme.
Over at XPENG, the Turing chip entered mass production in Q3 2025, with cumulative shipments exceeding 200,000 units. The target for 2026 is nearly 1 million units, and the entire lineup is switching to in-house chips in Q2. Li Auto’s chip has just "launched," with large-scale verification yet to begin and the data loop not yet closed.
If the simulation data route fails, the Mach M100 will be nothing more than Li Auto’s most expensive "brick."
Conclusion:
As mentioned earlier, Li Auto’s stock fell for two consecutive days following Livis Day.
It’s not that the market doesn't understand technology; it’s that the market understands it all too well. It knows the gap between a 5nm chip launch and mass verification involves many yield climbs. It knows the distance between defining a "four-in-one" embodied AI entity and delivering the mass-market experience involves many gaps between OTA promises and actual delivery.
As Li Xiang completes his identity switch from "product manager" to "tech evangelist," have the company behind him and the users before him kept pace? This cognitive disconnect is becoming the hardest chasm for Li Auto to bridge in the AI era.
