Breakthrough or hype? How WeRide aims to steer past rivals in crowded robotaxi field

WeRide, one of China’s big three robotaxi companies, has cut research and development (R&D) costs by “millions” of US dollars by using artificial intelligence to train its fleet in virtual worlds, its CEO said. While rivals had also developed AI models simulating the physical world, WeRide’s efforts stood out as it was using its world model Genesis to support its global expansion strategy, said Tony Han. “It’s the first real marriage between physical AI and generative AI,” Han told the South China Morning Post. A major challenge for the autonomous vehicle industry has been the high costs associated with developing self-driving algorithms that can generalise across different markets and terrains. In particular, the low tolerance for accidents has meant algorithms must be trained on many low-probability “edge cases” that occur in the real world, as shown by a recent incident in the US where a Waymo robotaxi drove passengers into a police stand-off. This has pushed many of the industry’s leading players, including Waymo and Tesla, to turn to AI world models as they seek to reduce unit costs to drive fleet expansion. In China, Guangzhou-based WeRide and crosstown rival Pony.ai have both emphasised their world model efforts in recent months, as they look to keep up with Baidu’s Apollo Go, the domestic leader by fleet size. Han said Genesis had allowed WeRide to test new self-driving algorithms in a matter of days in a virtual world, where previously the same tests would have required hundreds of cars operating in the real world for months. Crucially, Genesis enables WeRide engineers to simulate rare edge cases that are key to ensuring self-driving safety yet difficult to capture in the real world. This had resulted in a reduction of at least 75 per cent in WeRide’s data labelling and collection costs, Han said, adding that such costs made up a large percentage of the company’s R&D spending – around 316 million yuan (US$46 million) in the third quarter of 2025, or around 73 per cent of total operating expenses. Han said Genesis was almost entirely the result of advancements in generative AI technology in recent years. The company’s early experiments with physical world simulations had been rudimentary, with cars and pedestrians represented by Roblox-like blocks, limiting practical usefulness. This changed as generative AI, which uses machine learning to create models of its underlying datasets, made virtual worlds increasingly accurate depictions of the real world. WeRide trained Genesis using billions of kilometres of real-world driving data collected by the company, including more than eight years of real-world edge cases. Its powerful capabilities mean WeRide could sell the model as a stand-alone product in future – a long-standing “dream” for Han, a former chief scientist of autonomous driving at Baidu. Han said limits on access to computational power, more than rising graphics processing unit (GPU) prices, were now the main obstacle to continued improvement, given the clear upside of investment in this area for WeRide’s bottom line. “Every dollar we spend improving Genesis leads to 10 times more R&D efficiency, because it’s less spent on road testing and recruiting data labourers,” he said, adding they were now “probably” doubling their GPU stock every quarter. “The constraint is the shortage of memory and availability of certain advanced GPUs.” With the help of AI, WeRide aims to ramp up its global fleet from around 1,000 today to up to 3,000 robotaxis by the end of this year. However, investors have shown concern for WeRide’s future in a crowded field. Apollo Go and Pony.ai aside, Chinese start-ups such as AutoX and Momenta have also recently expanded pilot programmes, while Volkswagen announced this week its plans to launch its first robotaxi fleet in the US. Since its November listing, WeRide’s Hong Kong shares have fallen more than 20 per cent, while its Nasdaq shares have slumped almost 70 per cent in the past year. Analysts have highlighted the company’s continued R&D-driven losses as a point of concern, with profitability not expected until 2030. It remains uncertain whether Genesis will give WeRide an edge over rivals, with Pony.ai recently touting that its world model PonyWorld generates 10 billion kilometres for testing each week. Leading US player Waymo also unveiled its world model earlier this month, built on Google DeepMind’s most powerful world model Genie 3. It said the collaboration with its fellow Alphabet subsidiary gave its model an edge over others in the autonomous vehicle industry, as Genie 3 was trained on a huge number of videos and not just real-world driving data. Beyond technical indicators, Han pointed to WeRide’s aggressive global expansion strategy as a differentiator from its competitors, as the strategy could be further enhanced with a powerful world model even if it was not industry-leading. While WeRide still lags far behind Baidu’s Apollo Go in domestic deployment, the company has already expanded to more than 40 cities across 11 countries and regions, one of the highest in the world, compared with eight for Pony.ai and three for Waymo. The company struck a deal last year with Grab, the largest ride-hailing app in Southeast Asia, and announced a major expansion of its partnership with Uber earlier this month to deploy at least 1,200 robotaxis across the Middle East as soon as 2027. Meanwhile, Baidu said on Thursday that Apollo Go had reached 26 cities globally. “Genesis for us makes sense from a business expansion perspective,” said Han. “It allows us to more easily get into a new country or market without having to do all the hard work of collecting new road data.” With earnings coming up on March 23, investors could expect to see a Genesis-powered boost to the company’s financial health, he said. “I think we will have very good numbers.”

Breakthrough or hype? How WeRide aims to steer past rivals in crowded robotaxi field

.css-1c6uqr6{color:inherit;font-weight:inherit;font-size:inherit;font-family:inherit;line-height:inherit;overflow-wrap:break-word;}WeRide says its world model has slashed costs to support global expansion, but investors question if it can deliver a lasting advantage

WeRide says its world model has slashed costs to support global expansion, but investors question if it can deliver a lasting advantage