On September 22, in Qipanjing, Ordos, a line of heavy trucks moved at a steady pace along a mining road: a single pilot sat in the lead vehicle's cab, while an unmanned heavy truck followed closely behind, keeping its distance without any tension behind the wheel—only real-time data flickering on the dashboard. On this day, Kalpower launched its "Scaled AI Freight Network" here—from the first platooning route five years ago to the current routine operation of nearly 100 L4 autonomous trucks, this company, deeply rooted in mining zones, is putting to the test a phrase the industry has repeated for years: Can driverless heavy trucks make money on their own?

A firsthand experience of an L4 truck platoon run at Kalpower's unmanned freight operations site in Qipanjing, Ordos, revealed that the most striking impression was the "lack of drama." On real roads, the following vehicles completed tasks like tailing, turning, and climbing slopes, making the whole process resemble a normally functioning truck fleet. This perhaps offers a lens into the commercialization of autonomous heavy trucks. As the industry has evolved, proving that "the truck can drive itself" is no longer enough. The real question is whether it can be cheaper and safer than human-driven trucks, and whether it can scale from one route to many more. This is also the direction the entire embodied intelligence field is exploring.

How the Unit Economics Turn Profitable Without Subsidies

During the on-site discussions, Kalpower CEO Wei Junqing repeatedly emphasized "economic efficiency." According to the company's calculations, the cost structure of traditional long-haul freight roughly follows a "33211" pattern: tolls and energy each account for about 30%, labor about 20%, and maintenance and depreciation about 10% each. With relatively rigid costs like tolls and bridges, the most immediate change autonomous driving can bring is reducing labor costs. This is a key reason Kalpower chose the platooning model.

Kalpower COO Li Xiaoxiao stated frankly that for commercial fleets, even adding driver-assistance features like NOA does not truly eliminate labor costs, as long as a driver must still remain in the cab. Therefore, the company aims to combine a "piloted lead vehicle with unmanned following vehicles," allowing a single pilot to manage multiple trucks. The on-site test featured a "1+1" configuration, but with current technology, the platoon product can adopt a "1+5" setup, meaning one pilot leads five unmanned trucks. The company estimates that with these configurations, labor costs could drop by up to 67%, and current operational projects show a 10%-18% improvement in economic efficiency compared to traditional human-driven fleets.

Notably, Wei Junqing said the current projects do not rely on special operating subsidies and use the same freight pricing standards as traditional manned fleets, with their unit economics turning positive last year. This is far more significant than discussing technical parameters. Passenger vehicle intelligent driving can command a premium through better user experience, but commercial vehicles have little room for such "storytelling." A heavy truck is essentially a production tool, and customers ultimately care about how much revenue a truck can generate annually, how much cost it saves, and how quickly the investment in autonomous driving hardware and software can be recovered. A transport company executive attending the event also offered a straightforward benchmark: choosing an autonomous driving service provider depends on production efficiency, cost, and safety, with particular attention to real route data and the ability to quickly handle rare events. In other words, the real commercialization barrier for autonomous trucks is not simply "activating autonomous driving features," but rather transforming technology into a profit statement for freight companies.

After 100 Trucks, Replication Becomes the Harder Task

Running one route successfully is just the first step. According to Kalpower's disclosed data, it currently operates around 100 autonomous trucks, with cumulative L4 operational mileage reaching 10 million kilometers. Operations have expanded beyond the Ordos model to other regions, with partnerships involving over 30 customers, 8 OEMs, and more than 30 ecosystem partners. But the difficulty increases from here. When running a few or dozens of trucks, companies can deploy a large engineering team to monitor a single route. However, at the scale of hundreds or thousands of trucks, issues like vehicle dispatch, maintenance, insurance, accident handling, parts supply, and even driver training all become significant costs. If each new route requires deploying a massive new team, scaling up could paradoxically make it harder to turn a profit.

Thus, Kalpower's current approach is not simply selling an autonomous driving system, but replicating the operational framework already validated in Qipanjing. This is also one reason the company chose not to manufacture trucks themselves. Take Shaanxi Automobile as an example: the division of labor is clear—the OEM handles chassis, drive-by-wire systems, and vehicle manufacturing, while Kalpower focuses on L4 autonomous driving software and operational systems. A dedicated intelligent driving production line has been built with an annual design capacity of 100,000 vehicles, and after-sales support leverages Shaanxi Automobile's 1,041 service stations nationwide, supplemented by cloud-based fault warnings and specialized repair tools. Behind this division of labor lies a pragmatic financial calculation. China does not lack heavy truck manufacturing capacity, and building a new vehicle assembly and nationwide after-sales system from scratch would require massive investment. By leveraging mature OEM manufacturing and service networks, autonomous driving companies can concentrate limited resources on algorithms and operations.

Kalpower's expansion path is framed as "1+N+X": starting with the proven energy transport scenario, then replicating to more regions and freight segments. Its current focus remains on Inner Mongolia and the northern energy belt, including Shaanxi, Shanxi, and Gansu. This expansion strategy is not aggressive, but it aligns with the current reality of unmanned freight—prioritizing areas with stable routes, consistent cargo supply, and sufficient transport demand.

As Technology Commercializes, the Challenge Shifts from "Trucks" to "Systems"

During the discussions, a compelling debate emerged: Should unmanned freight be "technology-driven" or "operations-driven"? Looking at real-world deployment, this may be a false dichotomy. Autonomous driving companies certainly need models, algorithms, and data, but transport companies will not pay for model parameters alone. A researcher from Tsinghua University's School of Vehicle and Mobility noted that data scale is foundational for autonomous driving deployment, but simply accumulating data does not equate to improved model capabilities; filtering high-value data could significantly reduce training costs. Kalpower's logic leans toward "finding data from operations"—solving the problems that most directly impact transport efficiency, safety, and cost first.

This is where the value of mining zones and bulk freight scenarios lies. They may not attract the public attention of robotaxis, but they offer stable cargo sources, fixed routes, and high vehicle utilization. If technology can generate clear returns here, the business loop may close earlier than on open urban roads. However, this does not mean autonomous trucks have entered a phase of unimpeded expansion. First, there is safety and compliance. Kalpower claims to have obtained multiple L4 autonomous truck commercial operation licenses and cross-city commercial pilot qualifications, with no safety liability accidents since operations began. But at the industry level, variations in road access, operational rules, and accident liability determination across different regions still directly affect the speed of cross-regional replication for autonomous trucks.

Second, human roles remain necessary. The current platooning model is not about simply "removing" drivers, but reallocating manpower. Kalpower transitions some traditional drivers into pilot or remote monitoring roles, while managing operational risks through online inspections and fatigue monitoring. The company believes this transition can reduce the intensity of long-haul driving and extend some drivers' career spans. Further down the line lies the fully cab-less transport robot. According to Kalpower's roadmap, cab-less models are currently aimed mainly at enclosed or semi-enclosed scenarios such as industrial parks, steel, papermaking, and agricultural supplies, with expansion to broader public roads awaiting more mature policies and operational conditions. This essentially delineates the current commercial boundary for autonomous trucks.

For commercial vehicles, technology ultimately comes back to the most basic accounting: driving safer, at lower cost, and generating sustainable profits. Whoever can extend that calculation from a single route to a nationwide network will be the one to truly cross the hardest threshold of unmanned freight commercialization.