In the physical AI era, autonomous driving is evolving from single-vehicle intelligence toward an intelligent flywheel connecting data, models, vehicles, and operations. As the global leader in autonomous driving, WeRide continues to advance physical AI toward its next stage of commercialization.

In China, we continue to strengthen our operational capabilities. With Beijing and Guangzhou as our core operating hubs, our fully driverless robotaxi coverage continues to expand.

Through deployment in complex urban environments and real-world operating scenarios, we continue to enhance system robustness, safety, and operational efficiency. To date, WeRide's autonomous driving footprint has expanded to over 60 cities across 13 countries worldwide.

Internationally, WeRide is accelerating expansion across Europe and the Middle East through our asset-light model. In the Middle East, we have achieved fully driverless robotaxi operations in Abu Dhabi and Dubai, while continuing to expand our operations in Riyadh.

This proven robotaxi commercialization model is now expanding from the Middle East to Europe. Following our rollout in Slovakia, WeRide is rapidly scaling across Europe.

In June, we announced new robotaxi deployments in Madrid and Zurich. In August, we announced plans to deploy Denmark's first autonomous shared mobility services.

Within two months, we advanced public commercial operations across three countries, with Denmark representing our sixth European market. At the same time, we are accelerating expansion into right-hand drive markets.

In April, we partnered with Grab to launch Singapore's first public autonomous driving mobility service. In June, we partnered with Geely, Farizon, and Kwoon Chung Bus to advance right-hand drive robotaxi commercialization, with Hong Kong as the starting point.

This marks a major step toward WeRide's robotaxi mass production and operational capabilities across both left-hand drive and right-hand drive markets. Beyond robotaxi, our Robobus business continues to advance commercialization in Europe, with operational progress across France, Belgium, Spain, Switzerland, and other markets.

Our L2++ and L3 solutions are scaling rapidly. To date, our single-stage end-to-end platform, WRD 3.0, has won mass production commitments across more than 30 vehicle models, with active road testing and market validation underway in key global markets, including Germany, France, and Japan.

Behind these commercial achievements is our continuously evolving physical AI technology foundation. We introduced the WeRide WITT, the physical AI cognitive foundation model, pioneering the concept of the minimum physical fact unit, and building a new AI understanding framework centered on physical facts.

Combined with the Genesis world model, it connects high-quality real-world driving data, model training, and simulation validation to empower L2++ and L4, accelerating large-scale autonomous driving deployment. Operations validate technology.

Technology empowers products. Products drive commercialization.

WeRide continues to advance along this intelligent flywheel path, accelerating the global commercialization of autonomous driving.

Hello, everyone. Thanks for joining us today.

We had a great second quarter this year. To begin, I'd like to highlight three key factors that define our strong performance in the second quarter and emphasize the exciting progress we are making today.

That is overseas acceleration, asset-light scaling, and a clear path to self-sustaining cash generation. We made strong progress across all three areas during the quarter, positioning the company for the next phase of growth.

Turning to our financial performance, total revenue delivered robust growth, nearly doubling year over year and more than doubling quarter over quarter, with growth of 103%. Overseas revenue increased 164% year over year and approximately 170% quarter over quarter.

Meanwhile, gross margin reached a record high of 38%, representing an improvement of approximately 10 percentage points compared with the second quarter of last year. Benefiting from continued improvements in operating efficiency, EBITDA also improved, with the loss narrowing by 8% compared with second quarter of 2025.

On operational metrics, our L4 fleet reached approximately 3,400 units, up 22% since earnings release in April. Within that, our robotaxi fleet grew by 500 units to more than 1 1,800 vehicles, making a 40% growth over the same period.

Furthermore, our one-stage end-to-end L2++/L3 solution has progressed from securing design wins to full-scale mass production. Revenue from this business surged nearly 2,600% year over year and increased 201 219% quarter over quarter in second quarter.

During the quarter, we delivered approximately 30,000 units of our L2++ solutions, making the beginning of a scaled, high-growth commercialization for our L2++/L3 business. Across the industry, we believe we are uniquely positioned with both mature technology stacks and large-scale commercial deployment for both product lines: the multi-sensor fusion L4 autonomous driving solution, and the proprietary one-stage end-to-end L2++/L3 solution for mass production.

Turning to our strategy, this quarter we segmented our business into three areas: L4, L2/L3 ADAS, and AI infrastructure, to provide greater clarity into our long-term strategic planning. In second quarter, we further advanced our strategy as physical AI company, built on proprietary infrastructure and foundation models, with autonomous driving representing the most commercially advanced application of physical AI.

Our physical AI infrastructure enables us to distill the capabilities of large foundation models into efficient, lightweight on-board models. The key lies in our ability to transfer the underlying knowledge and representations learned by large-scale models, not merely their outputs, into smaller models, while preserving the intelligence, generalization, and decision-making capabilities required for real-world autonomous driving.

This enables us to bring increasingly powerful AI capabilities to cost-efficient on-board systems, supporting both the scaling of our L4 business and the continuous evolution of our WRD 3.0, i.e. our L2++/L3 solution.

More importantly, this creates a powerful technology flywheel. Our real-world L4 operations generate high-value data that continuously improves safety, robustness, and generalization, while our growing L2++/L3 fleet generates additional road data, accelerating L4 model development and expanding coverage of long-tail scenarios.

These are not two separate businesses. They are two reinforcing layers of one physical AI platform, where every vehicle deployed and every mile driven makes our technology smarter, safer, and more capable.

Next, I will walk through our specific progress this quarter across technology development and commercial deployment. Starting with our L4 business, specifically robotaxi, our core themes are asset-light overseas acceleration and strengthening regulatory moats.

Our overseas operations are built upon an asset-light model. We do not own operating vehicle assets.

Vehicles serve as a hardware entry point for market expansion, while we work with local partners to deploy and operate the business. This model allows us to scale our footprint with significantly lower capital requirements and greater operational flexibility.

Based on optimal utilization under normalized driverless operations, we estimate the steady-state annualized technology service revenue per robotaxi could exceed $50,000. As our fleet scales, we expect increasing benefits from the data network effects, cross-market learning, and algorithm generalization, which should further improve vehicle-level economics and create meaningful operations leverage.

I'd also like to clarify how we define unit economics. Our overseas UE is not based on a fleet ownership plus ride-hailing platform model, where revenue is derived from fare charging at a gross level.

Instead, our economics are based on a licensed virtual driver technology model. We provide regulated, locally verified, and recognized autonomous driving capability, and charge a recurring fee from autonomous driving technology services and mileage-based fees.

Because our positioning and our business model differ from fundamentally, the same term UE carries very different operational and financial implications from those in traditional ride-hailing platform. Backed by partners like Grab, Uber, GreenMobility, SBB, and etc., we are replicating this proven, asset-light operational model across Mid East, Europe, and Southeast Asia.

This marks that our overseas expansion is now entering a phase of large-scale commercialization. To put into more details, in second quarter, we announced new robotaxi commercial partnerships in Madrid, Spain, and Zurich, Switzerland, and Copenhagen, Denmark.

In the Middle East, we continue to expand full driverless robotaxi operations in Abu Dhabi and Dubai this quarter, now covering over 70% of the core urban areas. In Riyadh, our robotaxi operations zone has expanded to the airport terminals and surrounding central business districts.

By the end of 2026, we will be fully entering a phase of meaningful growth, supported by secure licenses, expanding fleets, recurring orders, solid revenue, and profitability in overseas markets. This will demonstrate the successful execution of our overseas acceleration and asset-light expansion strategies.

For the domestic markets, we continue to strengthen our operational capabilities and extend our driverless operation zones in Guangzhou to key areas of Tianhe District, including Zhujiang New Town, and Haizhu District, including the Canton Fair Complex. This represents a nearly threefold extension in our ODD compared to the end of 2026.

More importantly, our operating efficiency continue to improve meaningfully during the quarter. Average daily rides per vehicle reached 21, up by 24% quarter over quarter, while peak daily rides completed per vehicle climbed to 28.

As a ride-hailing platform with self-operated robotaxi fleets, we recorded nearly 35% quarter-over-quarter growth in registered users in China. This simultaneous expansion in scale, utilization, and user base drove a 140% sequential surge in our ride-hailing revenue, further demonstrating the scalability and strengthening the profitability of our operations.

Looking ahead, as China's regulatory framework for autonomous driving continues to mature, we expect to further expand our robotaxi operations into more cities in near term. At the same time, we will continue to build Guangzhou as our domestic benchmark, with the goal of integrating driverless robotaxi into public transportation system at scale.

Together, these initiatives will provide a strong foundation for the next phase of domestic commercialization. Beyond this, I'd like to take a step back and discuss our regulatory moats.

The operational licenses we have secured overseas are by no means simple administrative approvals. They are combination of years of technology adaptation, ecosystem integration, and rigorous safety validation.

This creates a core competitive moat that puts us two to three years ahead of the market. This moat is built on three core pillars.

First, deep technology localization and regulatory engagement. We adapt our technology to local traffic laws, road conditions, and driving behaviors, while working closely with regulators throughout the process.

Second, ecosystem and infrastructure integration. We establish deep technical integration with local operators and service providers, supported by strong on-the-ground execution.

Third, rigorous field validation and safety performance. Our fleets continuously accumulate localized real-world data, while safety framework and track record provide the foundation for earning regulatory trust and securing commercial licenses.

Importantly, as regulatory frameworks become increasingly aligned across markets, our experience and validation in one jurisdiction can help accelerate approvals in the others. This reduces marginal compliance costs, avoids redundant work, and creates increasing operating leverage, allowing us to scale our asset-light autonomous driving business globally with greater speed and capital efficiency.

As a growing industry consensus on L4 core barriers was also shared by the same mobility platform during its Q2 investor presentation, technical competitiveness depends not on total driving miles, but on the acquisition cost of rare scenarios, scenario diversity, and the practical data quality. Besides, generic large foundation models cannot satisfy autonomous driving needs.

AV systems rely on exclusive vehicle hardware and strict regulatory oversight, which demands customized models matching hardware specifications and compliance requirements. Backed by our robust proprietary end-to-end data toolchain, diversified data set accumulated from long-term cross-border L4 operations, and outstanding self-developed algorithms, we keep upgrading vehicle engineering safety.

Our field-proven safety performance earns sustained trust from global regulators, accelerates regional expansion, and builds an exclusive core advantage for global layout. Then turning to our L2++/L3 business, three themes define our progress: industry-leading technology, rapid commercialization at scale, and a path toward self-sustaining cash generation.

Leveraging on our leading technology advantage in L2++/L3 algorithms, we have secured six consecutive championships at China's intelligent driving competition, reinforcing our leadership position in intelligent driving technology. Commercially, we are entering a clear acceleration phase, with multiple OEM mass production programs advancing steadily.

Recently, we launched an L3 autonomous driving POC program with Mercedes-Benz, making an important validation of our technology by leading global OEM and further demonstrating strong industry recognition of our technological capabilities. As of June 30, cumulative delivery of vehicles equipped with our one-stage end-to-end L2++/L3 solution has exceeded 30,000 units during the reporting period.

Looking ahead, we expect number of vehicles powered by our solution to exceed 100,000 by year end and surpass half million units in cumulative delivers next year, positioning us for significant scale-up in our L2++ ADAS business. Finally, I will provide a brief update on our AI infrastructure.

In this quarter, we officially launched the WITT model, which is a physical fact foundation large model, completing a full capability loop spanning data comprehension, data generation, and end-to-end technical stack. It delivers deep synergy with our Genesis world model.

The WITT model extracts and verifies physical facts from road test data to build a cognitive foundation for machines to perceive and understand the real world. Leveraging validated physical facts, Genesis reconstructs diverse extreme operating conditions and long-tail scenarios via simulation.

The two models work in tandem to perform a closed loop covering data generation, simulation, and algorithm iteration, driving continuous upgrades to autonomous driving technology. Currently, it underpins our two core business lines of L4 autonomous driving and mass-produced L2++/L3.

In closing, in the second quarter, we delivered meaningful milestones and strong results across the business, which lead us to return to three key themes I highlighted at the beginning of today's call: overseas acceleration, asset-light expansion, and self-sustaining cash generation. Looking ahead, our focus remains clear.

On the L4 side, in overseas market, leveraging our regulatory leadership and permit advantages, we will replicate our robotaxi proven model across Europe and other markets, accelerating the transition from successful validation to scaled deployment. In China, we will continue to expand our deep market strategy, building benchmark operations that are fully integrated into urban transportation system, before expanding in a disciplined and repeatable manner across major cities nationwide.

On the L2++/L3 side, our top priority is straightforward: win more OEM partnerships, increase our market share with key OEMs, and accelerate deployments at scale. Growth is the key.

We will invest where we see clear conviction and meaningful scale potential, rather than pursue growth for growth's sake. At the same time, we will remain focus on both long-term cash flow generation and profitability, rather than expand simply for the sake of expansion.

Here beyond the numbers, I would like to thank my team. This is a team that embraces hard problems, confronts reality head-on, and remains relentlessly focused on execution.

That culture of intellectual honesty and operational excellence is what enable us to consistently turn vision into products, products into services, and services into large-scale commercial value, even in uncertain environments. Just as our physical AI foundation model, the WITT, World Intelligence Towards Truth, reminds us, true intelligence must be grounded in observable, verifiable facts about physical world.

The same holds our business. What ultimately matters is not what we claim, but what we continue to deliver in the real world.

Narratives may change, technologies may evolve, and the market cycles may come and go, but facts endure. Real-world performance endures, and a proven track record remains the most powerful language of all.

Next, I will turn the call over to our CFO Jennifer, who will walk you through our second quarter 2026 financial performance in details.

Thank you, Tony. Hello, everyone.

Let me start with a key message from the quarter. Q2 marked an important inflection point for WeRide, with strong revenue growth, continued gross margin expansion, and increasing operating leverage across the businesses.

Importantly, this growth was driven by three area we believe we can support a more sustainable financial model over time: the acceleration of our overseas business, the scaling of our asset-light L4 model, and the rapid commercialization of our L2++/L3 business. We're increasingly seeing the benefit of the strategy in our financial results, particularly revenue growth, gross margin, and capital efficiency.

In Q2, WeRide generated RMB232 million of revenue, up 82% year over year and 103% quarter over quarter. Importantly, growth is broadening across both our businesses and geographies.

Our L4 revenue reached RMB125 million, up 47% year over year, primarily driven by robotaxi. Our focus remain on scaling the core L4 business, particularly robotaxi.

Meanwhile, some of our other L4 businesses have natural seasonal cycle, with commercial negotiation typically complete in the first half and revenue recognized as deployment take place in the second half of the year. As those projects move into deployment, we expect stronger contribution from our broader L4 portfolios in the second half of this year.

Our L2++/L3 business continued to accelerate rapidly, with revenue increasing approximately 26 times year over year and 219 times quarter over quarter. This reflect an important transition from project development into mass production and scaled vehicle deployment.

As more OEM programs move towards mass production, we expect L2++/L3 to become an increasingly important driver of our overall gross revenue growth. Overseas market account for nearly 40% of group revenue and is becoming an increasingly important growth driver as well.

Overseas revenue increased 164% year over year and approximately 170% quarter over quarter. More importantly, the economic of our overseas business are increasingly attractive.

Our asset-light model allows us to scale through local partners and operating ecosystem without requiring a proportional increase in our own balance sheet investment. This gives us three important advantage: faster geographic expansion, improving margin as operation scales, and lower incremental capital requirement.

We believe this model is an important foundation for our long-term path towards self-sustained growth. Turning to profitability at the gross profit level, gross profit increased 143% year over year to RMB87 million, while gross margin expanded 9.4 percentage points to 37.5%, compared to 28.1% in the same quarter last year.

The improvement was driven by continuous scaling of our high-margin, asset-light overseas L4 business, rapid growth in L2++, and an overall shift towards higher-value, AI service-oriented revenue. As this mix continue to improve, we believe gross profit can grow even faster than the revenue over time.

Total operating expense were RMB533 million in Q2, slightly up 9.2% year over year, slower than the revenue growth. This is an early indication that operating leverage is taking hold.

R&D expense increased 36% year over year to RMB434 million, primarily reflecting our additional investments in AI infrastructure and foundation models. At the same time, a much slower growth in the total operating expense demonstrate that we're beginning to achieve greater efficiency as business scales.

G&A declined significantly to RMB69 million, mainly due to lower share-based compensation and professional fee, while selling expense only increased to 29 million as we expand our commercial activities. Overall, we're seeing increasing operating leverage, with operating expense growing well below the revenue.

Putting all this together, our net loss narrowed 1% year over year to RMB401 million in Q2, while EBITDA loss narrowed 8.1% year over year to RMB335 million. These results reflect the early benefit of our asset-light model and increasing operating leverage, particularly as we scale our overseas business, and reinforce our path towards sustained cash generation.

Finally, as of June 30, we have approximately RMB5.4 billion in cash and other liquid financial resource. This provide us a strong financial foundation for continued expansion.

Let me close with four key takeaways. First, revenue growth is accelerating and become increasingly diversified.

Second, our revenue mix is shifting towards higher-margin, physical AI service-oriented business. Third, operating leverage is beginning to emerge, with expense growing significantly slower than revenue.

Fourth, our strong balance sheet and asset-light model provide a capital-efficient foundation for continued expansion. We believe Q2 demonstrate an early financial benefit of the model we have been building: global expansion, asset-light deployment, higher physical AI service revenue, and increasing operating leverage.

Looking ahead, our focus is to convert our commercial pipeline into scale deployment, continue improving margin, and progressively translating revenue growth into stronger cash generation. We believe these are the key building blocks for durable, capital-efficient growth, a leading position in global autonomous driving, and a sustained path to profitability.

With that, operator, we're now ready to take questions. Thank you.

Hi Tony, Jennifer, congratulations with the great results. My first question is that we note as WeRide currently has a diversified business portfolios spanning L4, L2++, L3, as well as AI infrastructure.

The question from me is how does management prioritize resources allocation across different business and regions, and what are your goals for the second half this year? Thank you.

Okay. Thanks for the question.

First, let me start with a little bit explanation about our foundation model. As you have noticed these days, I think almost every industry have their capability boost.

Back to five years ago without large language model and AI progress, and I don't think any company in this world can do L4 and L2++/L3 and AI infrastructure simultaneously. Recently we noticed with our boost in the AI model, especially our Genesis and the WITT, and we found suddenly we can build a lot of technology into data synergies.

That is, by the way, want to explain a little bit about the name of the WITT. This model, WITT, we announced in 2026 AI World Conference.

It is a solution to a famous German philosopher, I think the greatest philosopher, Ludwig Wittgenstein. His famous word like "the limits of my language is the limit of my world" is exactly what we just learned.

We can segment every video into so-called minimal fact units. We basically can dissect every video and analyze every video into a very atomic fact.

Combined with our Genesis model, we finally find out we can generate all kinds of data as we desired, just like in The Matrix, one day Neo just noticed he see the nature of the world. We have this kind of feeling.

Drastically we reduce the resource we need, and we can just with one investment, we can do three things. That is L4, and the L2++/L3 ADAS, and AI infra.

We build up our infra very, very strongly, and generate the data from L4, and also generate data from ADAS, and make this two thing complement each other, build up what we call the double data flywheel. Actually this flywheel is why we are the only company globally that has achieved driverless vehicle operation at large scale.

At the same time, we have many car OEMs adopt our ADAS system. For the goal of our second quarter, we have several aspects.

One is we want toward a accumulated half million installation goal to walk, and we want to really push forward to have a lot of car OEMs adopting our ADAS system, and hopefully in the fourth quarter, Tesla FSD may enter China, and we want to do a directly head-to-head comparison against the FSD, because I'm a heavily FSD user in California, and I drove our own cars based on WeRide ADAS system, WRD 3.0. I think these two things are comparable, I want to see more car OEMs adopt ADAS system, and meanwhile we'll continue expand our robotaxi fleet.

Thirdly, for our AI infrastructure, we want more companies in the humanoid robot industry to adopt our infrastructure. That's the goal.

Here I conclude my answer. Yeah.

Thank you, Tony.

Thank you.

No more questions. Thank you.

Hi. This is Tim from Morgan Stanley.

Thanks for taking my question. Congratulations on the robust top-line growth and the global expansion during the quarter.

Just a quick questions about the overseas business. I think on the call, the management has highlighted accelerating overseas expansion and large-scale commercialization.

Just wondering could you share more detail, how should we view your overseas strategy, and in the meantime, can it be replicated across different regions? That's my question.

Thank you.

Thank you, Tim. Like we just mentioned, overseas market is a is important growth driver engine for us, and it's already contributed approximately for almost 40% of the group revenue.

We do believe there are significant room to run, and the structural case is very compelling. Those markets we're entering into, they're all facing acute labor shortage and rising labor costs, which make which creates a strong natural demand for autonomous mobility.

If we can let me give you a sense of the pace. For the Q2, in the in the past quarter, we launched three new robotaxi deployments, in Spain and in Switzerland and in Denmark.

In the Middle East, our robotaxi fleet has roughly doubled to around 400 vehicles since last quarter. If you recall from the Q1 like earnings, at that time we have like around 200, and even like like even given the considering the recent geopolitical tension in the region, and we have doubled the fleet in the quarter.

The momentum is real and it's accelerating. Also on the replicability, this is where we think our model genuinely differentiates.

We we operate overseas, we don't own any of the vehicles in operation. What we do is we handle the localized tech adaptation and the regulatory compliance, then we license our physical AI driver to local partners.

We capture the recurring technological service revenue as especially as our operation scales. This can lead us to let us expand very quickly without having heavy capital.

Also we can maintain a good margin and good cash flow. What's more important if we're looking at on the unit economics, the annualized recurring revenue per vehicle, it exceed $40,000 to $50,000.

It's also has meaningful upside as density builds up in each city. For us, our primary goal is just to increase the fleet size in cities we we're already operate.

We're very disciplined about where we go next. Usually we pick the market with proven like commercial potential and reasonable regulatory barrier to entry.

The playbook is well established. Everywhere we go, we secure the permit, we adapt the technology locally, we we set up like a benchmark project in the region and flagship project in the region, and then we replicate at a lower marginal cost.

We can we genuinely believe this is going to be repeatable at scale. Today we have active L4 operation across 12 countries, all well on track.

We're confident in the model and in the in our pipeline, and we look forward in serving passengers in more markets soon. Thank you.

Great. Thank you very much for sharing all the details.

It's really helpful. That's all my question.

Thank you.

Thank you, Tim.

Thanks, Tony, Jennifer. I have two questions.

The first question is regarding the ADAS. Can you please update the latest progress and the future plan for your Level 2++ and Level 3 businesses?

Given the relatively small size of WeRide's team on this front, what is your competitive advantage in this highly intensive competition? My second question, if I may, is how do you see the OEMs and also the ADAS companies claim that they can build a robotaxi by leveraging the data accumulated through the ADAS?

How do you see these trends? Does this pose a threat to your robotaxi business?

Thank you.

Okay. Thank you so much for this very two important questions.

Let me answer them one by one, okay. The first question is about our latest progress and plan for our L2++/L3 ADAS system.

To make it short, I will just say ADAS to mean our business, including both L2++ and L3 systems. Also the question is about our currently competitive edge.

I think I have to say the competition in ADAS system in China and also globally is quite intensive. Now everybody is trying to compare their ADAS system to the FSD version 14.

According to our internal evaluation, I think we are currently definitely in the urban scenario in China, we are on par with FSD. I'm not sure whether you have noticed or not, just last week, I did a live stream with our ADAS experience in Zhujiang Xincheng, okay, I.E.

the Pearl Harbor New Town, which is the central CBD area in Guangzhou. I drove the M60 based on our WRD 3.0 ADAS system.

At the same time, I do the live stream answering questions from all audience from my live stream room. I do this live stream for two hours.

Just driving in rush hour in Guangzhou without any intervention. Okay.

The video is online, you can check it. I think according to my best knowledge, this is the first time a founder of a ADAS company or autonomous driving company do a live stream in the rush hour in first-tier city without any intervention and answer the questions.

That shows the breakthrough of our technology. In terms of the installation units, right, our revenue grew by 2,600%, that is 26 times year over year, and the 20 and 220% quarter over quarter.

In a very short of time, we have already have 30,000 units installed in a quarter. Our goal this year is to get 100,000 units installed, and next year cumulatively we want to get to half million.

It is by far the fastest speed I have ever noticed in the ADAS growth in terms of the installation unit. This is actually why can we do that?

Thank you for your high praise we are doing with a relatively small team. The key, the secret sauce is our two important large language model: Genesis model and the WITT model.

I do believe, as you guys already noticed, with AI, people can form a one-person company. With our very good AI infrastructure and the foundation model, we can actually build the best ADAS solution in China and in the world with a team of 200 or 300 people.

That's the power of AI, okay? That is what I want to mention about our current progress.

The second question, thank you, it's outstanding question, that is we have seen many OEMs, which are car OEMs or some company with ADAS solutions, claim in a very high tone that we are entering robotaxi sector and we will do great things, a lot of PR. My answer to this kind of claim or this risk goal is, actually first of all, we welcome this kind of competition, and because this competition will make us stronger, WeRide as the leader and the first mover in robotaxi industry.

We expand globally, and we know that we also have a very strong background in ADAS. We know the big difference.

We know the requirements for redundancy, for the reliability, there's a 1,000 times difference. As I mentioned at the beginning of this year, I proposed a qualification threshold for robotaxi, that is, a company need to operate at least for half year with a fleet of 100 driverless robotaxi without severe accidents, they can claim themselves a robotaxi company.

Otherwise, it's just a claim. People may claim every day, and I don't need to respond to this kind of claim.

I just want to mention there's a big difference between ADAS system and L4 robotaxi. People will say, okay, with this progress in large language model, we can solve the problems takes 10 years of your efforts, we can solve probably in half year.

Now I make another claim that is: hallucination in digital AI may be entertaining, but it is fatal in physical AI. In physical AI, our society, our public roads cannot stand a fatal accidents, if we can avoid it.

In digital world, you just restart computer or just do it again. In the real world, a life lost is lost, you can never get it back.

With this kind of philosophy, with this kind of priority on safety, we know that we have very, very strong competitive edge. We have very deep moat for our L4 business.

Here are my answers to your two great questions, thank you.

Thank you very much, that's very clear. Thank you.

Hi, Tony and Jennifer, this is Tianyu from CITIC Securities. I have only one question.

How does management view the overall trajectory of future operating expenses, particularly R&D expense? Thank you.

Hi, Tianyu, I'll take your question. Like we just explained our financial results, you can see the operating leverage is already emerging.

Revenue grew 87% this quarter and R&D only increased 36%. Our R&D is largely shared like a technology platform supporting L4, L2++, L3, and all the other product category.

It does not really scale linearly with revenue or the vehicle deployment. As we can see as we scale across different business, the R&D cost per vehicle will continue to decline.

We have also moved past the peak investment phase for the AI infrastructure build-up, and where the steady deployment costs taper significantly once operations are established. Looking ahead, we think we will remain very disciplined in R&D spending and continue to invest in the technology mode, like say for the foundation model, like on the Genesis and WITT what Tony just described.

As our asset-light L4 business scales and L2++, L3 enter into mass commercialization, we expect the operating leverage to accelerate. We remain very well on track to achieve a positive cash flow in a single quarter by 2028, and we aim to have a break-even in the full year 2029.

Thank you.

Thank you very much.

Thanks for taking my questions. The company just launched the WITT model, so, Tony, could you please share more about on this?

Thanks.

I would like to share more about the WITT model. You can see from the WITT model, basically I will say the key part of the WITT model provide WeRide a unique analysis tool for video, and combined with Genesis model in tandem, first of all, give you a video, it has many facts, and you cannot simply do causal inference across different videos.

It's just like I give you a machine, and WITT model can help you to dissect the machine and find out the relationship between different components. Genesis give you another capability of learning the functionality of components and put them together to build another machine.

Basically just like if I give you, maybe say, for example, a Lego toy of a jeep, and then I can dissect it and make it into a cargo ship. Okay.

By putting these two together, we believe we find the very unique path to analyze hundreds of videos, thousands, millions of videos, and find out the causal relationship between all the facts, and then use them to generate the relevant data of very long-tail data for our autonomous driving. By this way, we actually boost our training capability and lower our training cost, and build up the desired distilled on-board model.

In a urban challenge, we won six consecutive championships. Our on-domain controller is relatively slower, and the TOPS is much lower than our competitors.

They may have 2,000 TOPS domain controller, we only have a 200 TOPS controller. We are winning.

Why? Because we can distill model according to our special needs, and we can get most long tail.

That's the power of our two foundation model. I will stop here.

If you have further questions, you can also ask.

Thanks a lot, Tony. That's super clear.

Thanks.

Thank you. Thank you.

we're

Thanks for taking the question. On the asset-light model and the platform relationships, under this structure your partners own the vehicles and the platforms basically own the riders.

WeRide basically sitting in the middle licensing the driver, that makes the durability of those relationships kind of the whole thing. Then on Uber, one of your partners, obviously in some markets, their track record here is obviously you had them rotating Waymo out in Phoenix, they just wound down their Serve Robotics relationship, I guess if you can just review what exclusivity or minimum volume commitments that you have in the Uber markets and in the asset-light model where you don't own the cars or the customers, what ends up being your leverage if the platform, Uber or whoever, decides to route demand to other partners?

Jennifer, you want to take this question or you want me to take this question?

Sure, I can take this question. Walter, we have this asset-light business model so that we can expand relatively faster in the global market, and especially wherever we go, we try to build a ecosystem with the local partners and of course also with the platform partners.

We work with multiple platform partners and we work with multiple and local partners. For us and this is really a huge market out there, this is really at the beginning, and we think setting up a healthy ecosystem, it's going to set a good foundation for everybody.

If you are looking at the real deployment figures globally, we believe we have the largest robotaxi fleet outside China and US among all of the AV competitors by a lot. We think our strategic moat is not only just deploying the asset there and that's it.

Basically if you're looking at what we do everywhere, we sell the asset to the local partners or the fleet operators, and because we do have a bomb advantage from China. We're going to keep have that advantage.

Then what's more importantly, because WeRide has a superior safety records, that's why we can get autonomous driving license in so many countries. Right now we have autonomous driving license, official license from eight countries, we operate in 12 countries.

Nobody else based on my knowledge has such an operational scale. For us, a key responsibility is to get the regulatory permit.

That responsibility is with WeRide. That's going to be the key for the longer term.

Everyone can be on the platform, but not everyone can get the regulatory permit. For regulator, everything is a merit basis.

For the AV player who keeps having the good safety record locally, globally, they definitely can get more permits as business scales. For some of them who doesn't have a such a good record, in fact some of the competitors are getting they will get questions, suspension from the regulator when their safety accidents continues to happen.

We have a strong confidence in our tech capability, in our local deployment capability, our regulator communication, and our state capability, and we have strong confidence in WeRide's safety record. With all of that, that's how we build the strategic moat globally, yeah.

I hope that answers your question.

Thanks, that's actually very helpful, especially on the regulator side in terms of who has the power in the market. I just have one follow-up on the OEM side, and specifically with Mercedes.

On the call, you just framed the L3 proof of concept as kind of a validation of the leading OEMs, but then Mercedes is also an Nvidia flagship partner doing this autonomous program on the full Nvidia stack, they're not just using Nvidia's open model, they're obviously doing the full stack. Can you just help us understand how this OEM as an example is evaluating WeRide which Nvidia wouldn't already cover?

Is this just a Chinese market thing difference where Nvidia stack wouldn't necessarily be accepted? Is it about cost?

Is it again on the regulatory? Is it that the OEMs will have multiple sources for L3 or L4 or L2++, whatever it is?

Then I guess the bigger question is over time, if Nvidia has a lot of money and they mature, what's ultimately going to have your software win at a company like Mercedes, obviously Nvidia with its open model and doing everything else, whether that's China regulatory or just cost?

Let me answer this question. I think, first of all, under the contract, we cannot reveal too many detailed things, but I really appreciate this question.

I always make a joke internally, I said, "Look, if you look at the nature of the company, WeRide doing so a lot of things is just like DeepSeek." I think previously I always heard Sam Altman answer the questions from an Indian engineer saying, okay, if you don't have such amount of money, I don't remember that exact figure is, maybe if you don't have $20 billion, never think about try large language model.

Look what DeepSeek has done, okay? I always believe the genius, the great engineers, the great minds can beat funds, okay?

That's why I like Silicon Valley, I believe the creativity of human nature. Therefore, I think WeRide is such a company with a great creativity, with a lot of genius, and we will build up the best ADAS system, the best robotaxi, which have already proved by the market.

Look at the resource we put in in our company, and look at the results we have got so far. I think the answer to this question is very straightforward, and I really appreciate this great question.

Hello? Can you hear me?

Thanks

Thank you very much.

Thank you very much. Bye.

Thank you.