Jiushi Expands Physical AI Push as Fleet Passes 30,000 Vehicles Across 300 Cities

The company presented the move as an expansion rather than a sudden turn. Jiushi CEO Kong Qi has said complex open urban roads generate high-value data, and a business model must make money to keep collecting data and advancing technology. The company says its fleet has entered a cycle in which data drives iteration and commercial revenue feeds research and development. The physical AI it describes is an intelligent system that already senses, decides and executes in the real world and has generated commercial revenue, according to the report.

Jiushi was founded in Suzhou in August 2021, during a downturn for autonomous driving. The unmanned delivery sector had long lacked a mature commercial scenario. In 2021-2023, the company focused on technology and on persuading customers to pay for L4 technology. It compared traditional vans with unmanned vehicles and defined a RoboVan without a driver's cab.

In May 2023, Jiushi launched the Z5, described as the first mass-produced L4 urban delivery vehicle. Selling it was difficult. Co-founder and CCO Zhou Qing said customers worried about whether the technology was mature and whether it could operate stably. Courier customers have low margins and value every bit of cash flow. Jiushi lowered the entry barrier by introducing interest-free installments from the commercial vehicle industry into unmanned delivery and by offering autonomous driving subscription services that shifted customers from one-time purchases to monthly payments. Zhou said the goal was to let customers use the actual value of unmanned vehicles at a lower initial cost. The Z5 also had all-scenario adaptability and was the industry's first unmanned cargo vehicle to run in urban motor vehicle lanes, according to the report.

Jiushi then iterated more Z-series models and added product lines with higher payload and larger cargo boxes, as well as lower-priced, lighter models. Sales grew from the first vehicle in 2023 to hundreds that year, to the thousands in 2024, and to more than 15,000 by the end of 2025. The global fleet now exceeds 30,000 vehicles in more than 20 countries and 300 cities.

After the fleet passed 10,000 units, Jiushi extended its capabilities to the scale of intelligent driving systems, allowing its autonomous driving capability to leave its own vehicles and combine with chassis and vehicle manufacturing from OEMs. In June this year, Dongfeng launched four L4 unmanned delivery vehicles using Jiushi's Zelos Inside solution. The solution provides general L4 autonomous driving technology to OEMs, special vehicle makers and special equipment manufacturers. The two sides jointly define products, develop and test, deliver mass production and build a national service system. Jiushi recently signed a Zelos Inside cooperation project with GAC. Zhou summarized the model as letting those who understand cars build cars and those who understand L4 provide support.

Jiushi also launched a Jiushi Car Rental mini program at the conference, offering customers stable transport capacity. When the unit of purchase changes from a vehicle to transport capacity, unmanned driving no longer needs to borrow the commercial logic of automobile products, the company said. It argues that the core of physical AI commercialization is efficiency improvement. Customers will keep buying only if the technology creates real labor value and reduces human working time. Jiushi says it has achieved physical AI value first in logistics through its unmanned vehicle products.

Zhou said judging whether a physical AI company can land requires three things: whether the technology works in the physical world, whether it can scale into real scenarios, and whether it creates real commercial value. He said the core is not the quality of a single product but whether AI can enter the physical world and form a complete loop from perception and understanding to decision-making and execution. Physical AI cannot stop at understanding the physical world; it must make correct decisions based on that understanding and ensure safe and stable machine operation. It must also scale to factories, airports and parks, cover enough scenarios, serve enough customers and run for long periods. Most importantly, it must create real commercial value. A company whose technology can only be demonstrated, cannot run long-term, or runs but customers will not keep paying is not yet a mature physical AI company, Zhou said.

Zhou told Leiphone that Jiushi's experience is not a transformation from delivery to physical AI. More accurately, after building commercial capability in unmanned delivery, Jiushi is extending validated technology and commercial capability to many industries. It relies not on a single advantage but on what Zhou compares to a barrel: technology scaling, product functionalization, vehicle delivery and manufacturing, operations, safety systems and customer service. Each plank must be strong enough. These planks grow from the same base, which starts with real data.

Jiushi's unmanned vehicles have driven 250 million kilometers. Each completed task and interaction on real city roads iterates its autonomous driving technology. The company set up a safety committee this year to coordinate functional safety and cybersecurity. Technically, vehicles are given considerable authority to handle complex situations: a vehicle automatically pulls over and alerts the cloud; the cloud makes an initial judgment and pushes a handling suggestion to a human operator; after one-click confirmation, the vehicle completes the safety response quickly.

Safety and technological maturity have turned the commercial flywheel, according to the report. Demand for transport capacity is emerging, and governments are drafting policies and standards to promote scaled use of autonomous driving. Rising demand and Jiushi's autonomous driving technology drive fleet expansion, model enhancement and wider operating range. These three factors reduce unit operating costs and further stimulate demand for transport capacity. Public information shows Jiushi achieved positive business cash flow and gross margin at the end of 2025. By January 2026, it had helped customers reduce operating costs by an average of 66%.

For the flywheel to copy faster to more places, technology must be standardized. Many L4 companies struggle to standardize technology and products even after commercialization, the report said. Jiushi has broken L4 autonomous driving capability into a general solution used in Dongfeng and GAC models. On hardware, different models share a multi-sensor fusion pre-control unit and gateway unit, with a complete calibration SOP to ensure stability. On software, the solution includes full-stack self-developed autonomous driving technology, a vehicle management platform, an operation system, a mobile app and an open autonomous driving interface for rapid vehicle management and scheduling. On services, Jiushi provides operations, customer service and after-sales personnel and a complete operating system, including operation guides, personnel training and on-site practical upgrade guidance. Different models differ in hardware layout and body structure, but the underlying driving capability, operation system and standardized development process can be reused, the report said.

With standardized capability spread across 30,000 machines, the test becomes large-scale device management. Tens of thousands of machines running every day require not only models but vehicle scheduling, fault handling, customer service, remote operations, charging, task allocation, OTA and data return. Around Jiushi Car Rental, the company built three platforms for operations, dispatch and supervision. The operations platform handles daily alarms and supports one-click vehicle movement and return. The dispatch platform automatically finds idle vehicles near customer demand points and dispatches them. The supervision platform shows the location and status of each vehicle. As scale expanded, the share of remote labor cost fell. In the second half of 2025, it dropped below 3% of Jiushi's average monthly total cost. Zhou said the decline did not come from cutting staff but from standardizing, platformizing and automating operations. Vehicle management, task scheduling, status monitoring, anomaly detection, data analysis and some operational decisions can now be handled by platforms and AI, with humans handling more complex problems.

Jiushi's expansion beyond unmanned delivery broadens its market. RoboVan addresses the unmanned urban delivery market, while general L4 technology and transport capacity services without vehicle purchases can extend to commercial vehicles, special vehicles, sanitation, airports, parks and environments that need mobility. The market has already expanded to light trucks. Jiushi's new Z20 light truck cannot yet enter trunk line transport because of policy, but parks and factories have generated significant transport demand. At exhibitions, people ask whether the vehicle can carry more than four tons or have a larger cargo box. Previously, customers might not have thought of using unmanned vehicles for such transport needs; as unmanned vehicles scale on roads, potential customers begin to imagine whether unmanned vehicles can solve problems in their own scenarios.

Revenue structure is also being reshaped. In the past, Jiushi mainly earned from selling vehicles. As more OEMs produce unmanned delivery vehicles, hardware and software costs decline and hardware profits are diluted. Zhou said selling vehicles is the basis for scale, Zelos Inside is capability output, and the transport capacity platform and Jiushi Car Rental turn capability into recurring service revenue. Jiushi hopes to move gradually from earning once from selling a vehicle to outputting a set of capabilities and continuously earning from them.

The scope for reusing R&D investment is also expanding. In the past, trained algorithms could only improve the competitiveness of Jiushi's RoboVan. Once the Zelos Inside model scales, the same R&D investment can serve multiple OEMs, vehicle models and scenarios. More served models can in turn feed data accumulation, and data is the foundation of technological progress. Zhou stressed that data authorization is a precondition for cooperation with automakers. Jiushi wants feedback from more real operating scenarios to enter its technology iteration system under compliance, security and authorization. The larger the vehicle fleet and the richer the scenarios, the more chances Jiushi has to capture long-tail problems in the real world and, through unified algorithms, simulation validation and version iteration, turn a problem encountered by a single vehicle into system-wide capability. Zhou said that if more vehicles carry Zelos Inside and provide more data, the growth efficiency of the Jiushi brain model will be higher. This positive cycle between data and models determines whether Jiushi can grow from a large unmanned vehicle company into a platform-like physical AI company, according to the report.