News Reporter |
News Editor | Wen Shuqi
Self-driving company WeRide just dropped a quarterly report that shows it’s “running faster than ever” — but Wall Street sent its stock sprinting in the opposite direction.
On August 12, WeRide posted its Q2 2026 earnings. Quarterly revenue hit ¥232 million, up 82.2% year-over-year and 103.1% quarter-over-quarter. Gross margin also improved nicely, climbing from 28.1% a year ago to 37.5%. Overseas business was the real standout, jumping 164% year-over-year and now accounting for nearly 40% of total revenue.
But here’s the kicker: right after the earnings release, WeRide’s shares plunged more than 10%. By the close of trading that day, the stock was down 9.7%, settling at $5.72.
That said, if you look purely at the operational metrics of its core business, this was hardly a quarter where the Robotaxi story hit a speed bump.
As of the end of July, WeRide’s L4 autonomous fleet stood at roughly 3,400 vehicles, with over 1,800 of those being Robotaxis. Back in April, when the company reported Q1 results, its Robotaxi fleet was around 1,300 — so that’s an addition of about 500 vehicles in just three months. When the annual report came out in March, the number was a mere 1,125. Clearly, the deployment pace has shifted into high gear recently.
The cars are also working harder than before. In Q2, WeRide’s Robotaxi average daily orders per vehicle climbed to over 21, up 24% quarter-over-quarter, with the single-day peak hitting 28 orders per vehicle. Meanwhile, registered users for its domestic Robotaxi service grew nearly 35% quarter-over-quarter, and domestic ride-hailing revenue rose about 140% sequentially. The L4 segment brought in ¥125 million in Q2 revenue, up 47% year-over-year, which the company attributes mainly to the Robotaxi business.
The overseas footprint is getting bigger too. WeRide’s Robotaxi fleet in the Middle East doubled from about 200 vehicles to 400 in a single quarter. Over the past few months, it has also announced entries into European cities like Madrid, Zurich, and Copenhagen. Q2 overseas revenue grew 164% year-over-year and about 170% quarter-over-quarter.
The catch? The company’s ability to make money hasn’t quite kept up. WeRide’s net loss for Q2 came in at ¥401 million, nearly flat compared to ¥406 million in the same period last year. Stripping out some non-cash items, the Non-IFRS adjusted loss actually widened from ¥301 million to ¥339 million. R&D expenses hit ¥434 million, up 36% year-over-year — that’s almost double the quarter’s revenue.
According to the company’s disclosures, WeRide held approximately ¥6.225 billion in cash, time deposits, wealth management products, and restricted cash at the end of Q1. By the end of June, that figure — cash plus other liquid financial resources — stood at around ¥5.4 billion.
Commenting on the stock’s sharp decline, Xie Siyuan, Managing Director at Yijing Capital, told NUPIAO that the core issue is WeRide’s position outside the top tier of autonomous driving. “The first tier consists of Huawei, Horizon Robotics, and Momenta,” he said. “Second-tier players like WeRide and Pony.ai have to carve out differentiated paths. WeRide hasn’t established a monopoly position in either its robobus or sanitation vehicle businesses either — that’s the crux of the matter.”
“The earnings report is just a catalyst; what it reflects underneath is the market re-evaluating the company’s competitive standing,” he added. “Right now, capital isn’t just asking how good the technology is — it’s asking whether a company can establish clear dominance in a specific niche and build durable, sustainable competitive moats.”
WeRide is also trying to make the Robotaxi business model less asset-heavy. Overseas, the company is increasingly expanding through a model where local partners own the vehicles, platforms like Uber and Grab supply the passengers, and WeRide provides the autonomous driving system. Founder and CEO Tony Han described this on the earnings call as selling a “virtual driver” that comes with regulatory approval. The company estimates that under stable fully driverless operations and high utilization, a single overseas Robotaxi can generate $40,000 to $50,000 or more in recurring technical service revenue each year.
This approach isn’t unique to WeRide. As NUPIAO previously reported , Pony.ai — which also operates Robotaxis — has similarly begun leaning more on partner-deployed fleets after initially validating its per-vehicle economics. For autonomous driving companies, the most obvious benefit is cutting the capital expenditure tied to vehicle ownership, while making fleet expansion feel a lot lighter in an increasingly crowded Robotaxi market.
But saving on vehicle costs brings another question into focus: if the cars belong to operating partners and the passengers are held by ride-hailing platforms, how much bargaining power does the technology provider actually have? Consider this: Uber, one of WeRide’s partners, once swapped Waymo out of its partnership system in Phoenix. And just recently, it ended its collaboration with Serve Robotics.
WeRide’s answer is what it calls its “regulatory moat.” In an industry heavily dependent on policy, CFO Li Xuan noted that WeRide has already obtained autonomous driving permits in eight countries. “Anyone can enter the platform, but not everyone can get regulatory approval,” she said.
That said, WeRide isn’t planning to rely solely on this one kind of “driver” to make money.
On this earnings call, Tony Han for the first time clearly divided the company’s business into three pillars: L4, L2/L3 ADAS, and AI infrastructure. Compared to the past, when WeRide was mostly seen as a Robotaxi company, its self-definition is clearly broadening.
Among the new businesses, L2 is the first to gain real traction. In Q2, WeRide delivered approximately 30,000 units of its WRD 3.0 L2++/L3 solution, with related revenue surging 2,593.8% year-over-year and 219.3% quarter-over-quarter. The solution has secured mass-production design wins for over 30 vehicle models, and the company plans to scale installations to 100,000 vehicles this year, with a cumulative target of over 500,000 by 2027.
Han’s expectations for this business are nothing short of ambitious. In a recent interview, he claimed that if Tesla’s FSD scores 95 points in intelligent driving capability, WeRide can hit 80 — while other domestic players’ products “might only get 30 or 40 points.” On this earnings call, he doubled down, saying that based on internal testing, WRD 3.0 can already go head-to-head with FSD in Chinese urban environments.
But L2 is a market where frontrunners have already emerged. In July this year, Momenta announced that its mass-produced installations had surpassed 1 million vehicles. Qingzhou Zhihang also crossed that threshold earlier this year. As leading autonomous driving companies make their way to the capital markets one after another, the L2 intelligent driving market is starting to look like “the game is already decided.” And as the industry shifts from pure technology competition to mass production and commercialization battles, concentration will only intensify, further amplifying the scale advantages of the top players.
WeRide’s counter-argument? It says it’s not building a separate L2 technology stack from scratch.
In Han’s vision, L4 and L2 share the same underlying R&D architecture. The 1,800-plus Robotaxis bring back rarer and more complex corner cases, while the much larger volume of L2 production vehicles down the road contributes more everyday road data. Both streams feed into the same model training system — something Han calls a “dual data flywheel.”
Just a month ago at WAIC, WeRide unveiled its physical AI cognitive foundation model, WITT . According to the company, WITT is responsible for extracting, identifying, and validating “physical facts” of learning value from continuous real-world road videos. The previously launched GENESIS world model then uses that data to generate simulation scenarios and long-tail situations that are hard to capture repeatedly in reality, which are then fed into vehicle-side model training.
What WeRide wants to do now is take the data, simulation, and training tools originally built for internal autonomous driving R&D and make them serve both L4 and L2 — and potentially sell them to others as well.
As NUPIAO previously reported, WeRide is pushing to spin off its data business into an independently operated subsidiary called Jingshuo, established in 2024. The business would expand from internal autonomous driving data loops into data generation, collection, simulation, and model training, and is already reaching out to embodied intelligence companies.
This earnings call didn’t confirm Jingshuo’s independent operational arrangement, but Han did formally include AI infrastructure as one of WeRide’s three core businesses. He also expressed hope that more robotics companies — especially humanoid robot makers — will adopt WeRide’s infrastructure in the future.
Xie Siyuan told NUPIAO that these pipeline businesses can indeed be spun off as standalone products to empower automakers. As intelligent driving becomes a decisive battleground and a must-have in the future automotive industry, many OEMs want to develop independent capabilities to gain strategic leverage in negotiations with first-tier suppliers like Huawei and Horizon Robotics. Second-tier autonomous driving companies see this opportunity, and spinning off products that don’t involve core algorithms to empower automakers is a sound business logic.
Li Xuan also emphasized on the earnings call that WeRide’s R&D spending is largely shareable across L4 and L2/L3 operations, and won’t scale proportionally with fleet size or revenue. She added that AI infrastructure has already passed its peak investment phase.
As for the losses that keep refusing to narrow and the sustained R&D spending that comes with them, Li Xuan laid out a clear timeline: WeRide plans to achieve at least one quarter of positive cash flow by 2028, and full-year breakeven by 2029.
WeRide has clearly mapped out its growth for the coming years. It’s just that from now until 2028, investors will have to keep waiting.
