Re-raising $100 million! Huawei's strongest competitor in the intelligent driving field: Constantly promising customers will damage the industry's reputation

Ask AI · Why does Yu Qian firmly believe that autonomous driving needs 20 years of deep cultivation?

Autonomous driving is moving toward physical-world AI. Lightboat Zhihang wants to “stick it out for 20 years” at the table.

Byline｜Aoran

Edited by｜Yang Buding

After the intelligent driving industry went on a tear, it has entered a brutal reshuffling phase. Most companies compete for buzz and projects, treat order volume as the only answer to survival, and what survives is usually the player that dares to wage a price war and knows how to do marketing.

Lightboat Zhihang seems to have taken a different path.

This company was founded in 2019. Its core founding team came from Waymo, a global benchmark in autonomous driving. In its early days, it focused on L4 autonomous driving. With the rapid boom of the intelligent assisted driving market, in 2022 Lightboat began mass-production business for L2+ assisted driving, starting to “walk on two legs.”

Over the past stretch of time, Lightboat has seemed like an “invisible” company hiding behind automakers. Since its founding, it has not followed the crowd to pile on people or parameters, and it has had no brand fanfare. All of its limited resources were poured into the only mass-production customer at the time—Li Auto.

In early 2023, at the recommendation of Yu Kai, founder of Horizon, Yu Qian met Li Xiang, founder of Li Auto, at Lakefront University. After just a chance meeting, the product teams of the two companies began a months-long conversation—more accurately, Li Auto investigated Lightboat Zhihang, which at the time only had a Demo, for half a year. In October 2023, Li Auto decided to hand the AD Pro project to Lightboat Zhihang, while AD Max would be developed in-house.

For quite a long time, Lightboat had only one customer—Li Auto. A business model that is overly dependent on a single core customer carries potential pressure. Especially then, the division between automaker in-house development and supplier roles was not yet clear. Many OEMs were increasing their in-house development investment, and the survival space for third-party suppliers was being continuously squeezed.

On the first day of the 2025 New Year, we had already held a talk with Lightboat Zhihang’s co-founder, chairman, and CEO Yu Qian. At that time, everyone in the intelligent driving industry was on edge. But as Li Auto sold strongly, Lightboat became the first intelligent driving solution provider to deliver at the 500,000-unit-per-year level for mass production. However, due to the combined impact of fluctuations in the financing pace and the pressure of mass production delivery, the company had to move forward at times through difficulty and adjustment.

Yu Qian said that at the time, the team size was around 400 people. To control business risks, the company made adjustments to some personnel, while making every effort to keep the core R&D team intact.

So when the industry was still competing by stacking compute and hyping concepts, Lightboat took a differentiated route. In April 2025, they officially released the world’s first end-to-end city NOA solution based on the single J6M platform. On January 21, 2026, it achieved mass production and went on vehicles. The first batch was equipped on Li Auto’s L-series refreshed AD Pro models. On a 128TOPS compute platform, this solution delivers an intelligent driving experience comparable to that of a 256TOPS platform.

As of now, Lightboat Zhihang has reached cooperation with 10 major automakers, including Li Auto, Chery, SAIC, Geely, GAC, and others, with a cumulative total of nearly 30 mass-production models deployed.

Exactly one year later, when we met Yu Qian face to face again, Lightboat’s deployed volume of assisted driving had already surpassed 1 million units, making it—after Huawei—the second autonomous driving company in the industry to break through the million-unit mass-production threshold. Just a few days ago, Lightboat officially announced the completion of $100 million in Series D financing.

Unlike a year ago, today’s intelligent driving industry—after quickly saying goodbye to hype—has entered a pragmatic cycle of competing for safety, delivery, and large-scale deployment. In this late-February discussion, we talked about how Lightboat views industry ecology and competition, how it plans L4, and whether this company has truly secured a seat at the table.

It is reported that in 2026, Lightboat’s city NOA is expected to achieve mass production and go onto about 50 models. In 2027, mass production and rollout of NOA are expected to exceed 3 million units. At the same time, Lightboat plans to pilot Robotaxi on a small scale in 2026 and conduct scaled deployment in 2027.

Yu Qian said that the road to autonomous driving is still long. If you keep “painting a pie” for customers and hoodwinking them, it will ruin the industry’s reputation. He predicts that autonomous driving is the best gateway to physical-world AI, and Lightboat must “stick it out” for at least 20 years at this table.

“Living longer may be more important than living better. Continuous, stable long-term operation is the key step that leads us to the era of fully driverless driving.” he said.

The following is the full transcript of 《Titan Talk》 and Yu Qian’s conversation. After Lightboat Zhihang’s financing news was officially announced on March 23, we added several additional questions. (For easier reading, the content was compiled and edited):

“Eating from a hundred households,” at least “sticking it out” for 20 years at the table

Titan Talk: The last time we talked was on January 1, 2025. When I met you this time, your state felt completely different. Back in January last year, you said Lightboat Zhihang would become the first assisted-driving supplier to ship one million vehicles. Now you’ve achieved it—what’s the most direct change in your mindset?

**Yu Qian: **I can feel that the confidence across the entire company—from top to bottom, whether employees or executives—has become stronger and stronger. On the other hand, the company’s requirements for products have been raised to higher standards, and our sense of responsibility has become even stronger.

I think one million units is just the beginning. With the big trend of electrification and intelligentization of automobiles, there will be even more room for growth in our installed base.

Titan Talk: In 2026 and 2027, what installation volume do you expect?

**Yu Qian: **We expect to add at least 50 new cooperative vehicle models this year. By 2027, the company’s total installed base will cross the 3 million threshold.

Titan Talk: Last year you were still very concerned about the intelligent driving industry entering the stage of getting “too much air,” and you even thought Lightboat Zhihang might be the part squeezed out. Now can you confidently say that the company has already secured a seat at the industry table?

**Yu Qian: **When it comes to competition, we don’t look at whether some short-term point in time is “good” or “bad.” We remain firmly committed to long-termism and long-term value, so we don’t overly focus on “the table talk.” Many people are currently wondering how many companies will remain in the end. We care more about how to create more value than about competition itself. As long as we stick to this, we can always remain at the industry table.

Of course, in the course of development, we will definitely build our own differentiated advantages. This is important. For example, we were the first to identify that the single J6M chip could enable city NOA functions. We were also the first to mass-produce it and deliver at scale to users. That is our differentiation.

Cage (Yu Kai, founder of Horizon) also said something that day—“Keep sticking it out at the table.” I strongly agree. We do autonomous driving with a long-term perspective—at least 20 years. Lightboat has already stuck it out at the table for 7 years. The next step is to keep it going for another 13 years.

Titan Talk: When you gave your speech, you said that if an intelligent driving company can last for 20 years, it will definitely become an amazing company. Why 20 years? In your mind, what should an amazing company be like?

**Yu Qian: **We say this because doing autonomous driving must be viewed with a long-term lens. You can’t let short-term changes in products or technology shake your long-held belief.

Looking back to the point in time when the company was founded in 2019, we believed that a 20-year period is enough to cover the stage of widespread adoption of driverless driving, so we set that timeline. If you look at autonomous driving on a 20-year scale, then fluctuations in the capital market and ups and downs in technological development are actually not that important. So adhering to long-termism and continuously creating value for customers and consumers is the fundamental principle by which we counter industry uncertainty and respond to industry cycles.

Titan Talk: In 2024, when the company was at its worst, how many employees did you have?

**Yu Qian: **At that time, it was roughly 300 to 400 people—about 400. That period was indeed pretty difficult.

To control business risks, we mainly made adjustments to people outside the R&D team. The strength of our core R&D team was always well protected. At the time, we distributed year-end bonuses via stock options—many people already know this. Looking back now, that arrangement led many team members’ mindsets to shift in a positive direction, which turned out to be a good decision.

Titan Talk: I talked with your cooperation partners and employees a couple of days ago, and they all felt that those tough experiences are an extremely valuable asset for Lightboat Zhihang, making the company more mature like a person. Do you have that feeling?

**Yu Qian: **I think that’s true. Many excellent companies have experienced challenges like this—it’s actually completely normal; there is nothing new under the sun. As business operators, what we need to do is avoid risks, never dance on the edge of a cliff, and ensure stable and sustainable operations.

I often say we don’t want to pursue how cool or high-profile the company can be. Instead, we hope the company can live in peace and develop in a stable and sustainable way. That’s what we pursue. So living longer may be more important than living better. Continuous stable long-term operations is a key step on the road to the era of fully driverless driving.

Titan Talk: During our conversation last year, Chery was not yet one of your investors. Back then, you were the only intelligent driving company in the industry without an automaker behind you. Now Chery has become your investor. From your personal mindset and company operations, what has changed?

**Yu Qian: **Chery is both our shareholder and our customer. We will do everything we can to provide support. But at the same time, General Manager Yin of Chery often says that we are “eating from a hundred households,” so we can’t just serve only Chery. We also need to serve a broader range of customers so more consumers can enjoy safe, reliable, stable, easy-to-use intelligent driving products that are not overpriced. On this point, our consensus with Chery is very strong.

Titan Talk: How long did it take from the first time both sides got in touch to finally reaching cooperation and investment?

**Yu Qian: **The process was actually very fast. I remember the media meeting in 2025 when I couldn’t attend because I had an important matter. But in fact, at that time, I was already talking with Chery.

Chery is a well-known “science-and-engineering men” company. Our cooperation, in essence, is Chery recognizing our product technologies, and also recognizing the company’s underlying values.

Actually, before that we didn’t have many intersections or any deep relationship foundation with Chery. What we relied on more was recognition of our technology, product strength, and values, which won Chery’s acknowledgment.

We also have cooperation plans on areas including unmanned logistics and future Robotaxi businesses. Both sides have many plans in cooperation.

Titan Talk: Which of Chery’s vehicle models are you currently providing intelligent driving solutions for?

**Yu Qian: **There are many partnered models. Some have already started delivery.

Titan Talk: You also mentioned GAC and SAIC. When did those cooperation projects start being approached and secured, roughly?

**Yu Qian: **GAC has been our relatively early customer—earlier even than Li Auto. In a sense, GAC was the first customer that brought us into the mass-production domain. They gave us the first mass-production model opportunity, which allowed us to get the product properly deployed.

Titan Talk: Looking at the several intelligent driving suppliers currently at the table in the industry, are the gaps between them still as large as before? Or are the gaps gradually narrowing?

**Yu Qian: **I don’t think you can view this statically, and you also can’t just look at a single point. The industry develops with ups and downs, and everyone’s overall direction is moving upward.

Around this time last year, the industry’s core was still delivering high-speed NOA products. This year, you can already see that even 100,000-yuan models will be equipped with city NOA solutions close to “standard.” That’s industry progress. I believe more new product functions will appear in the future as well.

The ultimate form of intelligent driving is the L4 level, but during this process, every company will bring out its best tricks, realizing the same goal from different angles—so there will be all kinds of solutions. Each company will have its own differentiated advantages. Reaching the L4 stage will still take some time.

Titan Talk: From the beginning you mentioned that industry competition is extremely intense. Your peers often say the same thing. I’m especially curious where exactly this intensity shows up—when抢单（winning orders） is fierce, or when customers put forward stringent requirements, or in other ways?

**Yu Qian: **I think competition exists at every level. Nowadays everyone talks about industry “involution.” The core reason is that the market cake is not big enough. If you can make the overall cake larger, then everyone won’t have to fight over a limited cake.

How do you make the cake bigger? It’s by improving the experience of intelligent driving products, and also making the product’s safety value measurable, statistically verifiable, and precisely calculable—then even combining it with the insurance industry.

From the perspective of society as a whole, when there are tens of millions—even up to hundreds of millions—of vehicles equipped with intelligent driving actually driving on the road, then you can statistically determine that vehicles equipped with intelligent driving solutions have higher safety than ordinary vehicles. Once the safety value of intelligent driving gets verified, it means fewer scrapes, fewer collisions, and fewer casualties, which in turn leads to lower insurance costs for society. This is the bigger value created by intelligent driving, and the share that everyone can enjoy becomes bigger too, and the industry will move toward a better future.

Titan Talk: In this new round of financing (March 23), top automakers plus global automotive electronics giants all entered at the same time. How does Lightboat ensure equal treatment for all customers and prevent resource skew?

**Yu Qian: **Like the joking analogy I made earlier, we insist on “eating from a hundred households,” and we don’t rely on a single customer. For resource allocation, we use platform-based products to meet differentiated service needs. For example, the technical base for single J6M city NOA is unified; when adapting to different customers, we only do customized adjustments.

At the same time, we open up the three major platforms from NVIDIA, Qualcomm, and Horizon, giving customers the choice. From start to finish, we speak with products and delivery.

Titan Talk: The intelligent driving industry is still in a reshuffling phase. Is this billion-dollar-plus financing for Lightboat “ammunition” for expansion, or a safety cushion to stay alive?

**Yu Qian: **Ammunition and a safety cushion are fundamentally the same thing—it allows us to be more at ease in adhering to long-termism. We will use this financing to deepen the “world model + reinforcement learning” technology roadmap, accelerate the commercialization of L4 unmanned logistics and Robotaxi, strengthen the company’s organization and talent building, and accelerate our global expansion.

“The road to autonomous driving is long. If you keep painting a pie for customers, you’ll ruin the industry’s reputation”

Titan Talk: I have an observation. Starting around the end of 2025, many intelligent driving companies have become more high-profile, including Lightboat Zhihang. You used to be a very pragmatic company, and personally you’re also quite tight-fisted, yet this year you held the first brand event. What’s behind this shift?

**Yu Qian: **Lightboat Zhihang has always been a pragmatic company, and it still is. It’s just that this year the company’s investment in the market has been a bit higher.

We believe that “good wine doesn’t fear the deep alley.” But sometimes good wine also fears the deep alley. Especially for C-end consumers—many people don’t really understand intelligent driving. We need to do education through publicity and jointly with OEMs, so that more consumers can understand intelligent driving and be willing to use it. In essence, we hope that through market promotion events, we can help OEMs sell cars well. That’s the logic behind why we do brand events.

Titan Talk: Previously, OEMs were not willing to co-promote with you. When you officially announced the J6M this time, for the first time you added Li Auto to the title of the official press release. In the past, it seemed like this had never happened. Why is that?

**Yu Qian: **Yes. In the intelligent driving field, there are two models: suppliers developing in-house and OEMs developing in-house. Now the industry is getting more and more pragmatic. Consumers don’t really care about which model provides the product. They care most whether the product is good to use, whether the experience is good, whether it’s safe, and whether the cost is high or not.

Now everyone has returned more to commercial rationality, and many OEMs have become more pragmatic and open. For example, our client Li Auto has very strong in-house development capabilities, and we respect them a lot. But even so, they still need suppliers’ support. These two models don’t conflict, and they shouldn’t be mutually exclusive. Everyone starts from real needs to solve specific problems.

Titan Talk: You just mentioned that you will connect with OEM 4S store sales teams. Specifically, what work will you do?

**Yu Qian: **Consumers’ final car-buying behavior mostly happens at 4S stores. But I’ve found that many 4S store salespeople from OEMs don’t explain intelligent driving properly. That’s because intelligent driving penetration is not that high yet, and people’s understanding of the product is still lacking. So we will work with the OEM’s sales team to conduct training, and discuss how to better showcase the capabilities of intelligent driving products. Intelligent driving products themselves have a certain barrier to entry.

Titan Talk: Will OEMs pay separately for these training sessions?

**Yu Qian: **This isn’t our paid offering.

Titan Talk: Based on the cooperation model currently, can you achieve profitability? Many are free, and automotive companies’ own profit margins aren’t large either.

**Yu Qian: **If you look only at a specific link, our costs are indeed not low. So the essence of the auto industry still has to win through scale and volume. We’re very pragmatic about that.

We tie our destiny to that of OEMs. Everything we do is to help OEMs sell cars well. We want to maximize overall benefits.

Titan Talk: At the roundtable forum, they also discussed win-win topics—like how both automakers and intelligent driving suppliers can make money. Chen Yudong (former CEO of Bosch China) also said that when he sold a solution to Chery back then, the price was indeed a bit high. So in terms of cost reduction for intelligent driving products, what new thinking do you have?

**Yu Qian: **I’ve always believed that cost reduction isn’t something one department can handle—it’s a systematic engineering effort. It involves product definition, the procurement supply chain, and supplier cooperation, and more across all these links. This requires automakers to do platform-based design well. Especially during the product design phase, you need deep integration with suppliers, and avoid fragmented product design so that the product becomes a standardized platform. Only then can everyone bring costs down together and achieve win-win results.

For example, Tesla sold millions of vehicles, but it has very few model variants. Its platform-based approach is truly excellent. In this respect, Tesla really is ahead of the industry.

I’ve always believed that product cost isn’t “saved” or “hacked down” by squeezing; it’s more often “designed out.” Technology cost reduction is the more fundamental and important approach. For example, if we can achieve city NOA functionality on a single J6M chip, that’s a reflection of technology-driven cost reduction.

Titan Talk: When did you start to anticipate that J6M could enable city NOA in urban areas?

**Yu Qian: **We made this judgment around the end of 2024. At the 2025 Shanghai Auto Show, we publicly announced it to the industry. After the announcement, there wasn’t much reaction in the industry. People thought this was just another company doing PR and painting a pie. We didn’t explain too much. Instead, we focused the bulk of our effort on product refinement—making the product truly deliverable and stable and reliable.

There is a fundamental difference between a Demo version and a mass-production product. Our products are for the broad user base. They must be as stable and reliable as a steel bucket. That requires spending a lot of time and effort. So from the Shanghai Auto Show release to final mass production, we spent more than half a year and put in a great deal of effort.

Titan Talk: In 2024, when you proposed that you could achieve single J6M city NOA, how did Li Auto respond? After all, the final rollout landed on Li Auto’s models.

**Yu Qian: **At the time, we discussed it together with Li Auto’s team. At the beginning, it wasn’t only Li Auto. Basically all OEMs are the same: they don’t want to see “futures,” and it’s not a joke. If you say you can do it, they still want to see spot deals.

So we did a lot of preparation work in advance so they could see the real possibility that this product would be mass-produced. And we maintained very close coordination with Li Auto’s entire team.

Li Auto publicly announced this to users in a release event around last August or September. The biggest value is improving customer satisfaction. Because many customers worry that their models will be forgotten. This kind of upgrade helps customers feel the brand’s care.

Titan Talk: Some people have done it; some haven’t. If this is the “air” in the industry in 2025, do you think there will still be such air in 2026?

**Yu Qian: **There are always differences within the industry in how people understand different chip platforms and different functions. So the industry is still far from the stage where solutions converge. There will be many changes ahead.

What we can do is do our own things well and deliver what we promise. This is also why many OEM customers recognize us. We never make commitments we can’t deliver just to win projects. We avoid overpromising. Even sometimes we are more conservative in our commitments. But we hope that at delivery time, we will exceed customers’ expectations.

We often say “under-promise and over-deliver.” Because getting a project in one or two instances can’t undermine our long-term trust with customers. The road to autonomous driving is still long. If you keep painting a pie and hoodwinking customers, it will damage the industry’s reputation. So we are very cautious when it comes to commitments.

Titan Talk: You’ve kept emphasizing that with limited resources you achieve stronger functionality. So what will your differentiated competitive strength be when you make products with 500TOPS or 1000TOPS compute?

**Yu Qian: **Actually, this year we achieved breakthroughs in products with one to a few hundred TOPS compute. By the end of this year, we will mass-produce products with 500TOPS compute.

And at that point in time, 500TOPS compute may become an industry standard. The industry’s compute level keeps moving upward—we won’t stop. In the future, we may even launch products with thousands of TOPS, even 2000TOPS compute. Now the industry already has related planning.

Titan Talk: You’re that confident?

Yu Qian: That’s because of what I just mentioned—achieving monthly takeover, meaning a level where a human takes over once per month.

Titan Talk: Is it using Horizon’s J6P chip?

**Yu Qian: **It’s not convenient to disclose this right now. Our principle is: once it’s done, we will announce it externally. At this point, we don’t want to talk about it yet.

“Lightboat doesn’t make chips; we give OEMs the choice”

Titan Talk: You said Lightboat is the first to step up in supporting domestic chips, and that you’ve done it the best. Why are you now strengthening this point so much?

**Yu Qian: **Supporting domestic chips isn’t just about pushing out lots of homogeneous domestic chips and letting enterprises engage in vicious competition. It’s about building a good domestic chip ecosystem—using domestic chips to deliver excellent product experiences and user value, so that people feel domestic solutions are also great. That’s the real support.

For example, in the past the country strongly supported the development of domestic operating systems, and in the PC era it invested a lot of effort. But those operating systems didn’t form a good application ecosystem. In the end, consumers still chose the operating system with a better experience and richer applications. So we believe that supporting localization means, like Lightboat Zhihang, using domestic chips to deliver better experiences—possibly even better than overseas comparable chips.

Titan Talk: Now some intelligent driving suppliers are starting to make their own chips, and some OEMs are also developing chips in-house. What do you think of this trend? Will Lightboat Zhihang make chips?

Yu Qian: We don’t make chips. Our understanding is that we should give the choice to the customer. We won’t require that if customers use our software, they must also use our chips.

In the early days, some industry players may have taken a bundling strategy like that. But now with AI technology progressing and the industry evolving quickly, giving OEMs the choice so they can select based on product requirements and market needs is more appropriate. We will focus on the areas we are good at.

Titan Talk: Between Horizon and you, is there some competitive relationship? Are there any red lines neither side can cross?

**Yu Qian: **We are absolutely more in cooperation than in competition with Horizon, and there is deep cooperation across many levels. Horizon’s chips aren’t only used by themselves. Many companies in the industry use them, so our cooperation approach is very open-minded.

In the end, it’s still the same sentence: we don’t overly focus on competition.

Titan Talk: The day when Dr. Yu Kai had that big screen, you were “Chicken Chop Brother’s” big brother. Before he put that content up, did he talk to you about it?

**Yu Qian: **No. He didn’t tell me about it.

Titan Talk: When the media sent you this content, what did you feel?

**Yu Qian: **I think it’s fine. Ke Ge himself is quite humorous. From another perspective, this also shows that the autonomous driving industry needs more everyday “real-life sparks”—it doesn’t have to be so high and mighty.

For technology companies, that’s what we say when we shut the door on ourselves and talk about “being a tech company.” Fundamentally, it means providing people with better products so that everyone can drive more comfortably and more safely. So having a bit of real-life “smoke and fire” is also a good thing.

Titan Talk: Dr. Yu Kai also mentioned that Horizon has the most ecosystem partners among current intelligent driving suppliers. How do you understand the term “intelligent driving ecosystem”?

**Yu Qian: **The core of an intelligent driving ecosystem is forming a diversified and thriving partnership network. Horizon’s product line is very rich—from J2 and J3 in the early years, to J5 and J6. The product layout is very complete, so it’s natural that many partners exist. If a chip is used by only one company, that would actually be abnormal.

Titan Talk: When you take on a new cooperation project now, how long does it generally take?

**Yu Qian: **Building a connection with a customer is a long process. But now the time it takes to land a customer is actually gradually getting shorter. This is also positive market feedback on us.

Titan Talk: Shortened to one year or to half a year?

**Yu Qian: **Much shorter than that. Chinese OEMs develop at a very fast pace. Basically within a few months—sometimes even measured in months rather than years—they can reach cooperation with customers.

Titan Talk: When you negotiate and communicate with high-level executives of automakers, what do they care most about? Is price still the main factor in intelligent driving cooperation?

**Yu Qian: **Price is obviously important too, because the scale and profit of the entire auto industry are a focus for the whole industry. But compared with before, everyone is no longer only price-driven. Now people increasingly focus on the core value of the product, including safety, stability, and reliability. Not just looking at PPTs or one or two demo cars.

In addition, the country has issued mandatory standards that have raised the safety requirements for intelligent driving products to a very high level. Whether it’s OEMs or intelligent driving suppliers, everyone must put safety first.

Titan Talk: I previously chatted with Wu Yongqiao from Bosch, and he said that once during an order scramble, a startup suddenly quoted a price 20% lower than theirs. Have you encountered something similar? If the other side cuts prices aggressively, would you follow?

**Yu Qian: **We never engage in price wars, and we never initiate price wars. Price wars harm the industry. Intelligent driving R&D requires huge investment. If everyone competes by cutting prices, in the end inferior money drives out good. So we firmly won’t wage a price war.

Titan Talk: In your impressions, is there any customer you had to win over that was especially tough? For example, needing multiple rounds of having executives test drive, and repeated communication.

**Yu Qian: **Actually, the process of winning over every customer has challenges. **Now OEM requirements are extremely high, and it’s different from before. Previously, you might just show PPTs. Now, many OEM executives pay special attention to intelligent driving products—**very high requirements for product stability, reliability, and user experience. So there is no customer that was won easily; we have to invest a lot of time, effort, and hard work.

Titan Talk: You’ve said that when you communicate with customers of fuel vehicles, they become more and more willing to listen to your opinions. When did that change start?

**Yu Qian: **Actually, from the start of doing intelligent driving, we have been communicating with many fuel-vehicle OEMs and participating already in the product design stage. In the proposal confirmation stage, we will share our understanding of the product with OEMs, and offer suggestions based on the vehicle’s positioning and target customers, so that the product aligns better with market needs and is easier to sell. In essence, it’s like that.

Also, when we suggest improvements, we never do it from our own perspective—whether it’s easy for us to deliver or whether it saves us trouble. We do it from the OEM’s perspective. For example, if you want to make a car a blockbuster, what functions should it have, what product form should it take, and how to fit the brand tone, price range, and consumer needs—these are the focus of our attention.

Titan Talk: In your view, what should a healthy, sustainable cooperation model and business model look like? The premise is that you need to achieve profitability.

**Yu Qian: **The essence of the auto industry is that you win through scale. So we will achieve profitability through scaled development. At the same time, I believe that new business models will emerge in the future. For example, Tesla no longer uses a one-time buyout model; it uses a subscription model. In the China market, subscription models may also emerge.

In addition, I mentioned earlier the combination of intelligent driving and insurance. For vehicles equipped with our intelligent driving solutions, the annual premiums can be reduced by a lot, which is very attractive to consumers.

The core reason premiums decrease is safety. Because intelligent driving can reduce the occurrence of accidents. Our product safety indicators are already close to that target. It’s just that the current penetration rate isn’t high enough yet to reach the level of large-scale statistics and actuarial precision. That’s a problem the industry currently faces. If we can achieve information sharing across society—do data statistics at the national level—then the safety value of intelligent driving can truly be revealed.

Titan Talk: What penetration rate of intelligent driving do you think is needed to realize the kind of scenario you described, where intelligent driving and insurance are combined?

**Yu Qian: **I think when the intelligent driving penetration rate of new cars’ annual sales reaches 30% to 50%, it’s pretty much there. If it’s for the total vehicle stock across society, it likely needs an even higher penetration rate.

Titan Talk: Then does it need to be at the L3 or L4 level to achieve that kind of combination?

**Yu Qian: **My understanding is that it doesn’t necessarily have to be L3 or L4. L3 and L4 are more industry-standard definitions. Even if it’s just L2++ or L2 assisted driving today, as long as the vehicle is equipped with intelligent driving functions and has higher safety than ordinary vehicles, beyond improving the driving experience, there will be real and tangible help on the safety side—then it can be combined with insurance.

For example, parking functions. For vehicles equipped with intelligent driving parking, the scrape rate is only one-third of that of human parking. And many insurance claims come from parking-lot scrapes. If intelligent parking can significantly reduce scrape rates, premiums will eventually be adjusted accordingly.

Titan Talk: Have you internally done calculations on what intelligent driving penetration rate the industry could reach this year?

**Yu Qian: **Actually it’s hard to predict penetration rate because many results have to be “done” rather than predicted. If we didn’t take the city NOA functions and bring them down to 100,000-yuan vehicle models like we did, the penetration improvement would be much slower. But I believe this year the industry penetration rate will see a big jump—an increase of 30%, even 50% or more is possible.

Titan Talk: Earlier, some industry people said that low-compute solutions will be eliminated sooner or later. I remember you also said something similar before. Is that the case?

**Yu Qian: **Yes. Our “low-compute” is a relative concept. 128TOPS is certainly lower than 500TOPS. But compared with 30TOPS, 128TOPS is actually high. When making intelligent driving products, first you need to ensure the safety red line and the safety baseline. On that basis, you can optimize. I talked about this view last year: safety comes first. All optimization cannot go below the safety baseline, otherwise it has no meaning.

Different companies have different safety baseline standards. We draw this line very high. Based on that high safety standard, we optimize. For example, compute of 30TOPS or 40TOPS may also achieve certain intelligent driving functions, but it is likely optimization that sits below the safety baseline—this is something we firmly won’t do. The core of intelligent driving products is safety, reliability, and stability. These standards not only must be met, but also must be greatly exceeded.

“We came out of Waymo, but chose Tesla’s route”

Titan Talk: Recently, everyone has been paying close attention to Waymo’s development and also its competition with Tesla. From the perspective of today, how do you see this?

**Yu Qian: **I’ve always believed that Waymo has made many breakthrough progress in the field of autonomous driving. Waymo and Tesla are both industry leaders, and their promotion of the entire autonomous driving industry has been very significant. That must be acknowledged.

Of course, Waymo and Tesla have taken different development paths. Their ultimate goal is the same: achieving fully driverless driving. But Waymo’s path is more challenging and more difficult, while Tesla achieves a commercial closed loop by combining products and technology, step by step pushing forward the development of technology and the market.

These two routes have the same ultimate goal, but different paths. Both companies deserve deep respect—they are both amazing enterprises.

Titan Talk: It seems that the industry is not too optimistic about Waymo’s route now, because Tesla has already achieved commercial success.

**Yu Qian: You could say that I came out of Waymo, but Lightboat Zhihang has taken the Tesla route. Because for startups, Waymo’s path is too hard to learn. **Waymo has strong funding, a strong team, and the founder’s ambition, so it can take this high-difficulty route. And only if you reach the final stage can this route create value; it’s very hard to realize commercial monetization in the middle process.

Tesla’s route is different. Its assisted driving products have already achieved large-scale sales, generating significant commercial and user value. It’s a step-by-step process. But I believe that whether it’s L2 or L4, the underlying technical capabilities need to keep improving. What we insist on is building a unified technical foundation, so technology can move forward toward “higher” levels. At the right time, we launch products that meet customer and consumer needs, achieve a commercial closed loop, and ultimately achieve fully driverless driving—this is our understanding.

Waymo has also made many long strides in autonomous driving, but they rarely disclose it externally. I have great respect for both companies.

Titan Talk: Looking at Waymo’s products today, there seems to be lots of redundancy design, and it looks rather bulky. Doesn’t everyone think this form factor is itself outdated?

**Yu Qian: **Actually, this is how I see it. Tesla’s approach is indeed smarter and more economical. For startups, you definitely need to learn Tesla’s path. Why? Because once your driving technology capability reaches a very high level, only then can you assign the product logic and achieve efficient deployment.

The core difference between L2 and L4 is the difference in product logic, not the difference in underlying technology. The underlying technologies all aim to make the vehicle’s driving capability stronger and stronger. But different product logic leads to huge cost differences.

As early as 2009, Google began exploring Waymo-related work. Back then, the driving technology level was still low. If you had to achieve L4 in that stage, you would have to do a large amount of redundant design. Even then, it would still be difficult to reach the goal.

As technological capability improves and driving capability reaches a higher level, then if you match the appropriate product logic, you can build products that meet market needs more quickly. That’s a smoother startup path, and it’s actually a very simple logic.

Titan Talk: Do you think Waymo will give up on the L4 route?

**Yu Qian: **I don’t think so. Even though its investment volume is clearly much larger than Tesla’s, its driverless effect has already been very good. It’s said that Waymo’s monthly order volume has already surpassed 2 million orders, and it already covers multiple cities. In the area of driverless operations, they have indeed done a lot of solid work.

I often compare Waymo and Tesla’s routes to the two paths up Mount Everest. Waymo chooses the north slope: extremely difficult, almost like climbing a rock face with bare hands, where every step has to be very solid. Tesla chooses the south slope: relatively smoother. The two routes are completely different, but the final goal is to reach the summit.

Titan Talk: Tesla is speeding up its catch-up in the Robotaxi field. Have you recently test-driven any of Tesla’s related products?

**Yu Qian: **Of course—we’ve been continuously tracking Tesla’s progress, including the latest v14 version. But objectively speaking, I think it will be very difficult for Tesla in 2026 to turn v14 into a fully driverless complete Robotaxi form. This year, it probably can’t be done, no matter how they describe it officially. Because driverless isn’t just a question of driving capability. It’s a complete product system that involves many dimensions.

For v14 versions that are currently available to buy, let alone being able to take over once per month, you might need to take over multiple times in a week. This takeover rate far fails to meet the requirements for large-scale driverless operations. And China’s road conditions are even more complex, making this takeover rate even harder to satisfy. Imagine it: if hundreds of thousands—or millions—of driverless vehicles are on the road, needing to take over several times per week. At a societal level, that is simply not acceptable.

Titan Talk: Specifically, what is it still missing in terms of driverless capabilities?

**Yu Qian: **The core is the takeover rate. Its takeover includes safety-related takeovers as well as efficiency-related takeovers, and efficiency takeovers are also a major problem. Also, L4 is an operational state, not a single standalone product—it requires providing a full service system, and there’s still a lot of work to do in that regard. Of course, we also acknowledge that Tesla has made tremendous progress.

Titan Talk: You just said you’ll frequently test drive Tesla products. What points do you mainly focus on?

**Yu Qian: **We focus on comprehensive product experience, which is part of our product benchmarking. Not only takeover rate, but we’ll study all its various product characteristics. This is a comprehensive benchmarking process.

Titan Talk: How often do you experience it?

**Yu Qian: **For everyday experiences, we also have colleagues in Silicon Valley, and it’s basically on a weekly cadence.

“World models are the more core essence, but this thing itself is hard”

Titan Talk: At the end of 2025, many intelligent driving companies and OEMs mentioned technical routes like VLA and world models. What changes do you think will happen in 2026 in terms of technical routes?

**Yu Qian: **If I’m making a judgment, I think the world model is the more essential core. Once the world model makes a major breakthrough, the entire autonomous driving industry will enter a phase of rapid development. Let me give a simple example. AI can beat humans’ strongest brain in many digital world scenarios like Go—such as AlphaGo and AlphaZero. But why can ordinary people easily drive cars, while AI finds it hard to do so?

The underlying reason is that physical-world AI needs complete closed-loop validation. In the digital world, AlphaZero can find optimal solutions quickly by playing virtual hands against itself—reinforcement learning—in a virtual environment. But in the physical world, we must deploy algorithms and models onto real vehicles for validation. There is no virtual environment that can fully replace the physical world, which makes it hard for R&D efficiency to improve quickly.

Once world models truly get realized, we can use powerful compute to run large numbers of algorithms simultaneously in a pure virtual environment and rapidly improve driving capability. So world models are extremely critical. But it’s also hard, because once you achieve a true physical-world model, in a sense you’ve already achieved autonomous driving.

I believe the development of vehicle-carried models and world models will advance alternately, ultimately reaching the ideal state. In the models we publish now, we’ve already integrated the world model and the vehicle-carried model in a very good way. Through mutual iteration between the two, we drive overall technical progress.

Titan Talk: You said the vehicle-carried model is VLA?

**Yu Qian: **Yes. For the vehicle-carried model, we use a VLA model. But we also have a world model that runs on the vehicle side. Its core role is to understand and predict the physical world.

Titan Talk: Then when you integrate VLA and the world model into one architecture, what’s the main purpose? This seems different from other companies that only emphasize a single technical route.

**Yu Qian: **I’m a technical person. I don’t want to make things overly mystical, and I want to explain it in plain language. The core of the vehicle-carried world model is to better understand the physical world and predict how the physical world will change over the next period of time.

For example, when driving, you see a soccer ball by the roadside. By understanding it through the world model, you can predict that in a few seconds, a little kid might run out to pick up the ball. That enables defensive driving in advance. That’s the value of the vehicle-carried world model: it can comprehensively understand the current state of the physical world and future trends.

Titan Talk: Starting in 2025, many intelligent driving companies have started bringing up L4 frequently again. Before that, for a while everyone collectively talked about L2 and rarely mentioned L4. Why do you think that is? In your recent release conference, you also mentioned Robotaxi-related goals.

**Yu Qian: **Lightboat Zhihang started with L4. Our mission and vision are to bring fully driverless driving into reality. We never gave up on this goal. It’s the core starting point for why the company exists.

But during the process of achieving this goal, you can’t just focus on L4. You need to create user value at different stages. Through matching technology, products, and the market to achieve a commercial closed loop, and then through one commercial closed loop after another, gradually push the realization of fully driverless driving. This is the logic we have adhered to from the beginning.

The development of driverless driving must follow industry rules. Humans’ dream of driverless driving has lasted for decades, but along the way there have been twists and turns. Only when sensors, compute, algorithms, data, and all other links have developed to a certain stage can technology, products, and the market form a good match, and finally achieve successful commercialization. If you force L4 in a stage where the technology is not mature, you’ll end up with half the effort for double the result, or worse.

Titan Talk: In your products, you placed emphasis on unmanned delivery vehicles. From a business-model perspective, is it already established now?

**Yu Qian: **I think it already has a trend of becoming established, but it hasn’t reached the stage where you can calculate the books with 100% clarity. The entire industry is still developing rapidly. Right now, the market size for unmanned delivery vehicles is only around several tens of thousands of units. In the future, it will gradually grow to 100,000 units and even a million-unit scale. The market potential is very large.

Also, from a practical logic standpoint, commercializing unmanned logistics will achieve faster than Robotaxi. Because if you can’t complete item delivery safely, reliably, and at low cost even within a city, how can you achieve safe transportation of people? It’s a very simple logic.

Titan Talk: Is your Robotaxi also built on mass-production vehicles?

**Yu Qian: **Yes. In the early days, we had some Robotaxi prototype vehicles mainly to refine the technology foundation. The Robotaxi we are launching now looks almost identical to mass-production vehicles from the outside. That’s because Robotaxi ultimately also needs large-scale mass production to enable broad market applications. Those older approaches that carry a huge number of LiDARs might be only a transitional form. Eventually, it will certainly move toward mass production.

In the future, autonomous driving will become a universal physical AI core

Titan Talk: The other day I saw data released by the CAAM. In the third-party market, Momenta’s market share already exceeds 60%, while Huawei is close to 20%. How do you view this data?

**Yu Qian: **Different statistical definitions can yield different results. As long as the descriptors are specific enough, you can find a statistical angle that matches expectations. What we care more about is the product market with NOA (navigation-assisted driving) functions. In that area, our market share is very high.

Also, this year our city NOA functions have already achieved large-scale mass production. The growth rate is very fast and shows a strong growth trajectory. I generally don’t pay much attention to all kinds of ranking lists, because there’s always a set of descriptors you can use to reach the desired outcome. What we value more is value creation itself, not these superficial rankings.

Titan Talk: You previously said that the future of the intelligent driving industry will be a pattern of “one superpower with many strong players.” In this, who is the “one superpower”?

**Yu Qian: **Actually, everyone in the industry can feel it. There are some very top-tier companies, and in a sense you could call them national brands. The existence of such companies is a good thing. It helps more consumers recognize intelligent driving products and makes the whole market cake bigger. Of course, the competitive landscape changes continuously. Today’s “one superpower with many strong players” structure doesn’t necessarily represent the future. In the end, it won’t be only “one superpower.”

From Porter’s five forces model, the upstream and downstream of the supply chain are inherently also competitors. For example, between OEMs and suppliers. Chinese OEMs can’t end up with only one. Because the auto market is highly fragmented: price ranges from tens of thousands to hundreds of thousands, and different vehicle types like sedans and SUVs, covering consumers across different age groups and with different needs—so multiple OEMs can coexist.

As long as there are multiple OEMs, they won’t want only one supplier on the supply chain. They will actively choose multiple partners and strengthen their supply chain control capability. So in the intelligent driving supplier space, there will also necessarily be a pattern of multiple players coexisting.

Titan Talk: You previously also predicted that in the future the industry might end up with 4 to 5 mainstream intelligent driving suppliers. Has this judgment changed now?

**Yu Qian: **The endgame structure of the industry mainly depends on the size of the market space and the capability to create value. If the market space is small, it might not be able to hold too many players. But if value can be created enough, it can accommodate more participants.

In the future, with the popularization of L4-level driverless driving, the market size will become bigger and the covered regions will become broader—not just China’s market, but also global markets—so the number of players that can be accommodated may also change. We’ve already entered a competition stage at the “World Cup” level, not regional competition. In the future, we’ll face a broader and higher-level competition.

Especially in the future general-purpose physical world AI field, it will attract excellent players across the globe. That will further expand the market space.

Titan Talk: If Lightboat Zhihang were to fall in the future, what do you think the reasons would be?

**Yu Qian: **We always insist on putting customer value first. Quality, stability, reliability, and safety are our bottom lines, and safety is even the top principle. If one day we abandon these core principles, we’re not far from failure.

Titan Talk: Recently, your competitor also introduced some talent from large model companies. How do you view the trend that talent overlap between intelligent driving, robotics, and large model companies is increasing?

**Yu Qian: **I think this talent overlap is inevitable. Because what we ultimately want to move toward is complete general intelligence in the physical world. Look at DeepMind. It started by playing games. Its core advantage isn’t being able to beat one particular game, but to beat all games—and even better than the strongest brain in humans. OpenAI started with chat. Chat is one of the most basic needs of humans, but it ultimately makes it possible to achieve general AI in the digital world.

I believe that general intelligence in the physical world will very likely start from the autonomous driving domain. In the future, what we do in autonomous driving won’t be just a machine that can drive. It will become the core of general physical intelligence. It can not only drive, but also help you do household chores and handle various complex situations in the physical world. That’s our final pursuit.

Under this grand objective, it naturally attracts many excellent people to join and work together to realize this dream. I think this dream, to some extent, is a key milestone for humanity on the path to higher-level AI.

Titan Talk: From an investor’s perspective, will they still ask you like before—when exactly can you achieve fully driverless driving?

**Yu Qian: **This is how I see it. Many times people are overly optimistic about short-term progress. Once they can’t meet expectations, they become disappointed. But then they tend to ignore the big long-to-medium-term trend. So we must look at industry development over longer time horizons. We can’t forget the long-term direction just because there are short-term ups and downs.

This big direction is very clear and heading in a good direction. After we move toward general physical intelligence, society will undergo tremendous changes. As long as we keep making achievements in this area, we will definitely become an amazing company.