Make money just like Meituan Delivery, another business in the autonomous driving sector.

In the past two years, a seemingly niche track in the autonomous driving industry has suddenly gained remarkable momentum.

Momenta, Horizon Robotics, Dyna Technology... These leading enterprises that develop assisted driving solutions for passenger vehicles have all extended their business to autonomous delivery. WeRide, which laid out the track earlier, has even spun off its Robovan business independently to seek separate financing at a valuation of 400 million US dollars.

From the perspective of the entire autonomous driving landscape, this trend is somewhat confusing. After all, from a pure sales point of view, this is far from a lucrative business.

First, look at the market size.

The China Federation of Logistics & Purchasing predicts that by 2030, the annual production and sales volume of autonomous delivery vehicles is expected to reach 860,000 units, and the total cumulative ownership of the industry will exceed 2 million units.

In terms of current market concentration — as of Q1 2026, Jiusi Intelligence (including Cainiao) has an ownership of over 25,000 units, accounting for 52.3% of the market share; Neolithic has 17,000 units, accounting for 36.2% of the market share. The two together occupy nearly 90% of the market. The total share of Jiusi Intelligence, Neolithic and White Rhino exceeds 95%.

Even if market competition intensifies and the market pattern becomes more decentralized in the future, with 5 to 6 leading enterprises sharing the total upper limit of 2 million units, each enterprise will only have an ownership of 300,000 to 400,000 units, which is very limited in scale, even less than the sales volume of a single passenger vehicle customer.

Next, look at the profit per vehicle.

Autonomous delivery vehicles are experiencing a fierce price war. Jiusi Intelligence has pushed the bare vehicle price of its E6 series down to 19,800 yuan, and the current available selling price range of Z5 is about 49,800 to 79,800 yuan; the preferential price of Neolithic X3 is about 115,500 yuan, which includes lifetime free use of FSD.

According to the research report of Soochow Securities, autonomous logistics vehicle manufacturers need to amortize R&D investment and operating expenses through large-scale production, and it is expected that break-even can be achieved when the sales volume exceeds 5,000 units.

Even if we optimistically assume that the hardware profit per vehicle can reach 2,000 yuan (a figure that has exceeded the per-vehicle profit level of many passenger vehicle intelligent driving solutions), the upper limit of the 2 million-unit market means that the total hardware profit pool of the entire track is only about 4 billion yuan. The annual profit of leading passenger vehicle enterprises in China often reaches tens of billions of yuan, and the annual profit scale of the entire automobile manufacturing industry is at the level of hundreds of billions of yuan. The 4 billion yuan is almost negligible, and it has to be shared by all participants.

Judging only from the "vehicle selling" business, autonomous delivery is indeed not worth such a big mobilization of leading autonomous driving companies.

The accounting of assisted driving,

is far more lucrative than autonomous delivery

Leading autonomous driving companies are not incapable of doing arithmetic. In their main battlefield, the figures are much more optimistic.

Assisted driving track:

Momenta: The prospectus shows that from 2023 to 2025, Momenta's operating revenue increased from 743 million yuan to 2.413 billion yuan, with an average annual compound growth rate of over 80%. Licensing service revenue increased from 23 million yuan in 2023 to 968 million yuan in 2025, a surge of about 42 times in three years.

Then look at the Robotaxi line:

Waymo completed a $16 billion financing in February 2026, with its valuation rising to $1260 billion. It has about 500,000 paid travel orders per week, with a target of reaching 1 million orders per week by the end of the year;

Tesla's Cybercab has reduced its per-vehicle cost to $23,000-$25,000, and its mass production operation cost is only $0.21-$0.25 per mile, close to 20% of that of traditional online car-hailing services;

Pony.ai recorded a Robotaxi business revenue of 116 million yuan in 2025, a year-on-year increase of 129%; its Robotaxi revenue in the first quarter of 2026 reached 59.12 million yuan, a year-on-year surge of 395.4%. More importantly, Pony.ai has achieved positive per-vehicle profit in Guangzhou and Shenzhen, among which the average net income per vehicle per day in Shenzhen reached 394 yuan.

Whether it is selling intelligent driving solutions or operating Robotaxi, the business model of passenger vehicles is far more "lucrative" than autonomous delivery.

On the one hand, the licensing fee per vehicle is several thousand yuan, and customers are original equipment manufacturers with shipments of millions of units. On the other hand, the autonomous delivery track has an upper limit of 2 million units in market size and a total profit of only 4 billion yuan. From any financial perspective, this is a multiple-choice question that does not require hesitation.

So why do leading players still squeeze into this "small pond" of autonomous delivery?

To understand this "collective impulse", we must first see through another business — the food delivery war that burned 150 billion yuan from 2025 to early 2026.

Food delivery does not make money by "selling meals",

the same goes for autonomous delivery

From Q2 2025 to Q1 2026, a hundred-billion-yuan level money-burning war of food delivery broke out in China's Internet industry. Meituan, Alibaba and JD poured huge capital into the market frantically. It is estimated that the total subsidy exceeded 150 billion yuan, which pushed the average daily order volume of the industry from the original 80-90 million units to a peak of more than 200 million units.

After the subsidy tide receded, the data of Q1 2026 showed a strangely reasonable trend: Meituan's average daily food delivery order volume stabilized at about 65 million units, but the customer unit price was only around 30 yuan. More notably, 75% of the new orders are concentrated in the price range below 15 yuan. The average daily order volume of Taobao Flash Delivery and JD Food Delivery is about 50 million and 9 million respectively, and their customer unit prices also remain at a low level.

Looking only at these figures, this is simply a business that "gets poorer as you operate it".

It is reported that in June 2026, Meituan's profit per order was about 0.1 yuan. Earning a dime per order sounds like a bad joke. But let's do the math: 65 million orders multiplied by 0.1 yuan equals 6.5 million yuan of net profit per day. That's nearly 200 million yuan a month, and more than 2 billion yuan a year. And this is only the food delivery business segment.

This is the core of "low unit price and large base": the operating cost of the platform is relatively fixed, the more orders there are, the lower the marginal cost. Small profit per order relies on scale to overwhelm costs.

Moreover, food delivery is not a profit center, it is a traffic valve. Food delivery is one of the highest-frequency daily demands, which can firmly retain users and divert traffic to high-margin businesses such as hotels and tourism. This "high-frequency hitting low-frequency" strategy is the real profit model of food delivery platforms.

Autonomous delivery follows exactly the same path, and the industry is undergoing a business model shift from "selling vehicles" to "selling transportation capacity".

Taking Jiusi and Neolithic, two specialized enterprises in this field, as examples, we learned from the sales staff of both sides that:

Jiusi Intelligence adopts a subscription system. After customers purchase the vehicles, they must buy subscription services to use them. The monthly subscription fee for the Z5 model is about 1,900 yuan, and that for the Z8 model is about 2,400 yuan. The first year is free, and the fee will be charged from the second year. The subscription package includes autonomous driving software, operation support and route planning services. A rough calculation shows that the annual subscription fee for one vehicle is 20,000 to 30,000 yuan. Based on a 5-year life cycle, the operating revenue will contribute 100,000 to 150,000 yuan, far exceeding the profit margin of the hardware itself.

Neolithic has a different strategy. The service fee of autonomous driving software is directly included in the vehicle price, which is a lifetime buyout. The continuous revenue comes from the annual traffic fee and insurance fee of about 2,000 yuan, and there is a 5-year free policy at present. Logically, the hardware profit is obtained in advance, and then lower service fees are used to lock in customers, which is essentially to seize market share.

The common point of the two models is: hardware is only the ticket, and operation is the gold mine.

The value of this "operation" logic is fully reflected in the cost reduction data. Taking the end-connection scenario of express delivery as an example, the average monthly comprehensive cost of autonomous delivery vehicles is about 4,200 yuan, a decrease of about 40% compared with the 6,800 yuan of traditional manual delivery. In the urban distribution scenario, the monthly cost of autonomous urban distribution vehicles is about 2,583 yuan, while that of manned electric trucks is about 8,233 yuan, and the cost reduction rate reaches 69%. The case of SF Express is more convincing: according to the calculation of Huayuan Securities, assuming that it puts 8,000 autonomous vehicles into operation by the end of 2025, the monthly cost reduction per vehicle can reach 2,186 yuan, and the annual operating cost can be reduced by 189 million yuan.

These cost reduction data are exactly the confidence for autonomous delivery enterprises to charge operating service fees to customers. Part of the saved cost is given back to customers, and part is converted into the platform's own operating revenue.

The profit of autonomous delivery is indeed thin, but it has a large base, high frequency and strong sustainability. Just like Meituan's food delivery earns only a few cents per order, it can support tens of millions of orders a day. Although the per-vehicle operating revenue of autonomous delivery is meager, it can bring stable and predictable cash flow to enterprises. The scale is not large, but it is characterized by steady and long-term returns.

Technology reuse:

Low migration cost

Autonomous delivery has an additional attraction for leading autonomous driving companies.

It is widely believed in the industry that on the premise that the existing technology stack can be reused, the marginal cost of entering the autonomous delivery track is not high.

Shen Shaojie, CEO of Dyna Technology, once revealed that "the commercial vehicle model can be directly migrated from passenger vehicles, with extremely low engineering cost"; Yu Qian, CEO of Qingzhou Zhihang, also expressed a similar view: L2 and L4 essentially share the same AI model base. Although the product forms are different, their requirements for safety, comfort and predictability are completely consistent.

In an exchange with the technical director of a general-purpose autonomous driving company, the person in charge also gave a similar judgment: "If you have a good command of autonomous driving technology for passenger vehicles, you can make logistics vehicles in closed parks work, which is a dimensionality reduction. If you make Robotaxi work, low-speed park logistics delivery is a subset of it." In his view, all autonomous driving companies must eventually become general-purpose intelligent bases, "there is no way back but to transform into the fields of trunk logistics, autonomous vehicles and robots."

This means that enterprises such as Momenta, Horizon Robotics and Dyna Technology, which have invested billions of R&D funds in the field of high-level intelligent driving for passenger vehicles, can directly migrate their underlying perception, planning and control algorithms to a large extent to the autonomous delivery scenario. The same set of base model only needs different scenario adaptation, and the migration cost is far lower than that of developing from scratch.

This judgment is supported by data. Pony.ai's technology licensing and application service revenue reached 229 million yuan in 2025, and the annual delivery volume of its autonomous driving domain controller increased by more than 5 times compared with 2024, and its customers "mostly come from industries such as robots and low-speed unmanned delivery logistics". This shows that the autonomous driving technology capability for passenger vehicles is spilling over to the low-speed delivery scenario.

For enterprises that already have a mature L2+ technology stack, entering the autonomous delivery track is almost a "convenient" move — there is no need to rebuild the R&D team, no need to re-accumulate data, and only need to carry out adaptive optimization for low-speed scenarios. This input-output ratio is extremely cost-effective in business.

In addition, autonomous delivery can dilute the high R&D cost of autonomous driving to a certain extent.

The R&D of autonomous driving is a "money-burning" protracted war. Taking Momenta as an example, its cumulative R&D investment in the past three years is as high as 4.66 billion yuan, and its R&D expenditure in 2025 alone is 1.869 billion yuan, accounting for 77.5% of its revenue. Cao Xudong, CEO of Momenta, once said that serving 10 million mass-produced vehicles is a key indicator. When the total number of vehicles served reaches 10 million, the huge R&D investment will be fully diluted, and the profit space of the enterprise will be significantly expanded.

The incremental 300,000 to 400,000 autonomous delivery vehicles mentioned above cannot be compared with the passenger vehicle market, but it is an additional boost that hardly requires extra R&D investment, can effectively increase the total installed volume, and accelerate the approach to the inflection point of scale effect.

The "parking lot business"

of autonomous driving companies

If we want to find a position for autonomous delivery in the entire autonomous driving landscape, the most appropriate analogy may be: the supporting parking lot of a shopping mall.

Building a large shopping mall requires huge investment and is the core asset. The parking lot is not the main part. It is planned incidentally when building the shopping mall, with almost no extra cost. But once the shopping mall is put into operation, the parking fee will become a continuous net income, which is a stable, predictable and almost zero-risk net cash flow.

Assisted driving solutions and Robotaxi are the "anchor stores" in the shopping mall, while autonomous delivery is the "parking lot business". It does not need to rebuild the technical system, nor does it need to form a R&D team from scratch, but it can open up a channel for continuous cash flow outside the main business. The profit per vehicle is thin, but small gains add up to a large amount; the market scale is limited, but it is stable and sustainable.

This is an additional business with a small scale but almost guaranteed profit.

This article is from the WeChat official account "Circular Intelligent Driving", the author is Circular Intelligent Driving, and it is published with authorization from 36Kr.