The physical AI wave is finding a more concrete application in autonomous driving technology: embedding models, sensors, and operating systems into real vehicles to continuously execute transportation tasks.
This trend is also fueling industry financing. According to incomplete statistics, by August 7, 2026, six domestic companies involved in unmanned delivery vehicles and heavy-duty trucks had completed distinct funding rounds totaling approximately 78 billion yuan based on disclosed minimums. Several other deals are still in progress. In comparison, the entire year of 2025 saw the top players in the unmanned delivery sector raise nearly 10 billion yuan in total.
Additionally, WeRide plans to secure independent funding for its Robovan business, while Pony.ai recently launched an unmanned light-duty truck and is pushing forward with the mass production of its next-generation autonomous heavy-duty trucks.
Participants have expanded from autonomous driving companies to include OEMs, logistics platforms, and industrial capital. The focus of investment has shifted from last-mile unmanned delivery last year to heavy-duty trucks that can achieve mass production and operational loops, as well as passenger car companies entering urban delivery projects.
Capital attention has also moved beyond autonomous driving software to include whether vehicles can be delivered in batches, whether cargo sources can be consistently secured, and whether transportation services can be billed based on results.
The industry must answer whether each vehicle can consistently generate transportation revenue, thereby enabling the next phase of expansion.
The Surge of Interest
Since the start of the year, the latest moves in the autonomous driving industry have increasingly focused on vehicles capable of carrying cargo.
In April, White Rhino and Xin Yuan established a joint venture combining autonomous driving technology, vehicle manufacturing, supply chain, and distribution channels. Changan Kaicheng began batch deliveries of its Robovan and partnered with JD Logistics. In June, WeRide pushed forward independent financing for its Robovan business, with sources indicating investors valued the business at around $400 million. By the end of July, Momenta's Robovan began operations in Suzhou, using the R7 world model and a mapless solution validated in its passenger car production business.
The fleet size of low-speed unmanned delivery vehicles has also changed. Following the integration of Nine Intelligence and Cainiao's unmanned vehicle business, the RoboVan fleet exceeded 20,000 units, covering hundreds of cities. The market now includes fleets available for operation and procurement, rather than just single-point demonstration vehicles.
Financing is the most apparent signal. In February, Nine Intelligence completed over $300 million in Series D funding. In March, Karl Dynamic secured over $100 million in Series B financing, and Zero One Vehicle raised 1.2 billion yuan. In April, DeepWay disclosed its Pre-IPO round had accumulated over $310 million. In May, Zero One Vehicle completed another $200 million Series B2 round. In July, SiNian Intelligent Driving raised 300 million yuan in Series C financing. Based on disclosed minimums, these six rounds total approximately 7.8 billion yuan.
This is not an exhaustive list of capital movements. The amount for White Rhino's new round was not disclosed, and WeRide's Robovan independent financing is not included. There are also several undisclosed rounds, potential independent financings, or strategic integrations in the market.
The sources of funding are also changing. Zero One Vehicle's new round included industrial capital such as CATL, Zijin Mining, Yankuang Capital, and Sanhua Holdings. DeepWay's Pre-IPO round featured Middle Eastern capital, Australian pension funds, a Singaporean impact fund, and local state-owned capital. Investors are no longer just looking at algorithms, but also at electric chassis, vehicle manufacturing, logistics scenarios, and customer orders.
However, an insider at an autonomous driving company noted that investors are optimistic about the long-term development of unmanned freight, so there is no rush to push for business spin-offs and IPOs.
In the past, financing for autonomous driving companies was primarily used for training algorithms, accumulating test mileage, and purchasing sensors. Now, funds must also cover the costs of vehicle mass production, remote operations, depot construction, and cargo sourcing, making investors more concerned about whether capital can flow back via vehicles, orders, and cash flow.
Beyond financing, the industry is organizing unmanned freight through strategic integrations, business splits, and new vehicle models. White Rhino formed a joint venture with an OEM, Changan Kaicheng partnered with a logistics platform, and WeRide is separating its Robovan business for independent funding. Pony.ai launched its first unmanned light-duty truck in August and is advancing the mass production of its fourth-generation Robotruck. Technology companies, OEMs, logistics platforms, and capital are now dividing labor around the same fleet of vehicles.
This wave of enthusiasm is not just about valuing software. In scenarios like ports and mines, where boundaries are relatively clear, heavy-duty trucks can first establish stable operations. Trunk line transport revolves around road rights, vehicles, and cargo sources. At the same time, an aging driver population and high-intensity work are pushing logistics companies to seek alternative transport capacity. When autonomous driving companies enter new vehicle categories, they can reuse existing systems, operational platforms, and supply chains. These conditions are enabling capital to fund vehicle production, remote operations, and scenario construction.
Vehicle Models Are Just the First Step
Technology reuse is a prerequisite for the rapid launch of new vehicle models.
On August 5, He Xing, Vice President of Pony.ai and Head of its Truck Business Unit, stated that the technology sharing rate between light-duty trucks and Robotaxi exceeds 95%, with an overall reuse rate, including operations, supply chain, and other aspects, exceeding 90%. Code-level reuse for the Robotruck exceeds 80%, and hardware sharing exceeds 90%.
The boundaries of reuse differ between the two product types. For urban road products, reuse is primarily in perception, decision-making, operations, and policy pathways. For Robotrucks, which face long-haul, port, and bulk transport, the vehicle structure, charging infrastructure, and road conditions vary more significantly, making reuse more concentrated in code, sensors, and supply chains.
To turn this capability into revenue, it must be installed in compliant vehicles, placed into real cargo networks, and finally, the transport results must be measurable for pricing.
The unmanned light-duty truck recently launched by Pony.ai is an urban delivery product, distinct from the Robovan class. The vehicle, co-developed with CATL, has a cargo volume of approximately 18 cubic meters, a top speed of 70 km/h on city roads, and targets B2B urban distribution scenarios like express sorting, supermarket replenishment, and cold chain catering.
From a vehicle classification perspective, N1 light-duty trucks and low-speed unmanned delivery vehicles belong to different product systems. The N1 is a motorized cargo vehicle, while low-speed unmanned delivery vehicles primarily serve parks, neighborhoods, and last-mile delivery.
This product difference directly impacts the business scenario. Low-speed vehicles have lower costs and speeds, making them suitable for last-mile delivery. Unmanned light-duty trucks can handle larger cargo volumes and longer urban delivery routes, but their automated operation on public roads requires compliance with product approvals, defined-area operation permits, and transport business licenses. The pilot document from the Ministry of Industry and Information Technology and three other departments states that certified intelligent connected vehicles can only conduct road trials in designated areas, and those involving transport operations must also meet the operational qualifications and management requirements of the transportation authorities.
Robotrucks follow a different product path. Their goal is not urban last-mile delivery, but rather placing autonomous heavy-duty trucks into trunk lines, dedicated routes, and port transport. In April, Pony.ai received approval to conduct a "1+N" platooning demonstration with unmanned following vehicles on the Beijing-Tianjin and Beijing-Tianjin-Tangshan expressways. The lead vehicle still has a safety driver for dispatch and anomaly handling, while the following vehicles operate without a driver.
After entering the unmanned freight market, the operational capabilities of autonomous driving companies must also improve. Pony.ai's cooperation with Sinotrans has extended from heavy-duty truck trunk line transport to urban delivery. The logistics partner provides cargo sources and operational scenarios, while the autonomous driving company handles the system, fleet management, and remote assistance, putting the vehicles on real routes. He Xing mentioned that the routes for the light-duty truck overlap with already covered Robotaxi areas, and the truly new operational aspects are the loading and unloading processes.
In a stage where customers are reluctant to own vehicle assets, technology companies must first build out this operational capability, using real orders, uptime rates, and maintenance data to price their technology.
Pricing models also evolve with product maturity. According to a research report from Huatai Securities, Robotrucks primarily have two pricing methods: charging a transportation service fee per kilometer or per ton-kilometer, or, after selling the vehicle, charging an annual or per-mile subscription fee for autonomous driving capabilities.
In the Robovan direction, Cao Cao Mobility has introduced vehicle sales, leasing, and a RaaS (Robot-as-a-Service) model. Customers can buy vehicles, lease them, or directly purchase intelligent transport capacity. Vehicle assets, autonomous driving systems, and transport operations are beginning to be priced separately, but its first Robovan only started operations in Changsha in July of this year, meaning the model is still being validated.
Vehicle launch is just the transformation of a system into supply. It is real cargo sources and the resulting transport economics that will determine if it is a viable business.
What the Market is Calculating
Once vehicles are in operation, the autonomous driving system becomes just one variable in the per-vehicle financial equation.
Cargo density determines whether a vehicle has work. Loading rate and empty mileage determine how many kilometers are billable. Waiting and loading/unloading times determine how many trips a vehicle can complete per day. Transport revenue must then be reduced by depreciation, energy consumption, maintenance, remote assistance, insurance, and downtime losses, leaving the per-vehicle cash flow contribution.
Xiao Ping, Product Manager at Pony.ai's Truck Business Unit, illustrated this with an example. Express and logistics companies have core operating hours from 5 AM to 11 PM, typically requiring two drivers per vehicle to ensure availability, but there are often 10 hours of daytime idle time waiting for orders. An unmanned light-duty truck, after completing its primary transport tasks, could use this 4-5 hour idle period to accept orders from freight platforms.
In the Robovan model, like low-speed unmanned delivery vehicles, the remote monitoring ratio and vehicle operating hours are another set of key parameters. According to Huatai Securities' estimates, in scenarios like direct transshipment, the per-unit transport cost for an unmanned delivery vehicle could drop from 0.47 yuan under a traditional manned model to 0.22 yuan. However, this model assumes a remote monitoring ratio of 1:100 and that the vehicle's effective daily operating time increases from 8 hours to over 10 hours.
This calculation shows that cost advantages rely on a single remote operator managing a sufficient number of vehicles and on vehicles achieving longer effective operating hours. Simply moving the driver from the cab to a control room does not automatically reduce per-vehicle costs.
Robotrucks primarily face challenges related to route and operational efficiency, meaning the "N" in a "1+N" platoon cannot be reduced to a simple numbers game. Xiao Ping noted that while technically possible to have "1+4" or "1+5" platoons, the platoon size must be based on specific operational requirements. If loading/unloading speeds are insufficient, "1+2" or "1+3" platoons should be used; "a larger N is not always better."
For trunk line transport, road rights themselves are a factor in expansion speed. New routes require permits along the entire corridor. The Beijing-Tianjin-Tangshan route only obtained permission for unmanned following vehicles after accumulating sufficient test mileage. Autonomous light-duty trucks entering urban roads similarly require product approvals and local operating permits.
Therefore, the real market competition is not about which company can get a vehicle to drive, but which company can place its vehicles into a higher-density cargo network.
Technology companies ultimately aim to return to the position of an autonomous driving solution provider, handing over vehicle assets and transport operations to partners. However, whether customers are willing to purchase vehicles first depends on the vehicles' uptime rate, accident rate, and payback period. Capital can prepay for R&D and fleet construction, but it cannot replace customer orders. Whether a vehicle can cover its depreciation, insurance, maintenance, and remote labor costs through transport revenue determines whether subsequent financing from leasing, insurance, and financial institutions will follow.
Insurance companies are also waiting for the same set of operational data. He Xing from Pony.ai explained that early insurance premiums for its intelligent connected vehicles, for a coverage amount of 5 million yuan, were once tens of thousands of yuan. Now, the per-vehicle premium is lower than for manned fleets. This change in premiums is still driven by accident rates, repair costs, and actual uptime records.
The dividing line for unmanned freight will ultimately hinge on three numbers: daily effective paid mileage, empty driving and waiting time, and the per-vehicle investment payback period.
Comments