The future of Hesai lies in enabling "machine awakening".

The timing of Hesai's dual listing was extremely well - chosen.

On September 15th, Hesai Technology announced a deepened partnership with a leading US Robotaxi company, signing a lidar order worth over $40 million. It will serve as the sole lidar supplier for the company, and the order is scheduled for delivery by the end of 2026.

The next day, riding on the good news, Hesai officially listed on the Main Board of the Hong Kong Stock Exchange, becoming the first lidar company to achieve dual primary listings on both the US and Hong Kong stock markets. It raised approximately HK$4.16 billion, setting the largest IPO scale for Chinese concept stocks returning to Hong Kong in the past four years.

On that day, Hesai's pre - market share price in the US soared strongly, rising by over 6% to $30.29. After the Hong Kong stock market opened, the share price quickly climbed, with the increase exceeding 10% within two minutes, and its market value once exceeded HK$35 billion.

Recently, Goldman Sachs released a research report stating that as lidar accelerates its popularization in the Chinese market this year and begins large - scale mass production among global automakers from 2026 to 2027, it is predicted that the overseas ADAS lidar shipments will reach 3 million units by 2030, equivalent to the scale of the Chinese market in 2025. Goldman Sachs gave a "Buy" rating to Hesai's Hong Kong shares for the first time, with a target price of HK$281. The target price for its US shares was raised from $26.3 to $36, also with a "Buy" rating.

However, as of the mid - session on September 26th, Hesai's Hong Kong share price was HK$227.2, only slightly higher than the issue price of HK$212.8, and far from the target price given by Goldman Sachs.

For this company, which ranks first in global lidar shipments, a deeper question is emerging: In the current situation where more and more passenger car automakers are turning to the pure - vision route, where exactly is the future of lidar?

01

There are two routes for autonomous driving technology: one is the multi - sensor fusion route, which uses multiple sensors such as lidar and cameras to work together. It has comprehensive perception but high hardware costs. The other is the pure - vision route, which mainly relies on cameras and vision algorithms. It has low hardware costs but high requirements for algorithms.

Previously, with the significant reduction in lidar costs, the first route seemed more promising. However, recently, the pure - vision solution pioneered by Tesla is attracting more and more followers. In 2025, He Xiaopeng, the chairman of XPeng Motors, clearly stated that XPeng would fully switch to the pure - vision solution. Even the facelifted SUV G7 removed the lidar.

Similarly, some models of BYD, such as the Dolphin Intelligent Driving Edition and the Seal EV Intelligent Driving Edition in the 100,000 - yuan price range, have also abandoned LiDAR and adopted the "Tian Shen Zhi Yan C" pure - vision solution. In the market below 150,000 yuan, lidar is no longer a "must - have", and the pure - vision solution is becoming the mainstream choice.

One of the reasons for the route change is that the pure - vision route can further reduce costs.

Early on, a set of lidar systems was priced as high as 700,000 yuan, a proper high - end product. Previously, the hardware cost of Tesla's FSD system was about one - seventh of Waymo's. However, now, domestic manufacturers such as Hesai have brought down the lidar cost to rock - bottom prices. In 2025, the cost of some domestic lidars has dropped below $500 (about 3,500 yuan), and some enterprises can control the cost at the thousand - yuan level through technological optimization.

Even at such low prices, in the fierce competition, automakers pursue extreme cost control and large - scale mass production, saving costs wherever possible. After Tesla demonstrated the feasibility of the pure - vision solution, it will encourage more automakers to try.

In addition to cost, safety issues have always been a key focus in the comparison between lidar and pure - vision.

Statistically, Tesla's FSD system has about 0.15 accidents per million kilometers, while Waymo has about 1.16. In terms of accident rate, Tesla's accident rate is about one - seventh of Waymo's.

At first glance, the pure - vision route seems to be much "safer" than lidar.

However, behind the apparent gap lies a huge difference in statistical calibers: Tesla mainly reports serious accidents where the airbags deploy, and its autonomous driving mainly operates on highways and other relatively simple scenarios. In contrast, Waymo reports all accidents and operates in a "fully driverless + complex urban road network" environment.

This "fully driverless" feature is the most important indicator for L4 - level autonomous driving.

In the field of L4 - level autonomous driving, the technical path divergence between lidar and pure - vision has evolved into a deep - seated systematic difference.

Compared with the complex multi - sensor fusion route, the advantage of Tesla's pure - vision route is that it is easier to establish a data - driven closed - loop system. With camera images as the only input, the data consistency is good. The massive real - world visual data collected by the vehicle fleet can be directly used to train and optimize AI models, enabling the entire system to continuously evolve like human learning. The iteration speed is very fast, and it is gradually considered more likely to achieve human - like intelligent driving.

However, for the model to truly reach a human - like level, the model algorithm needs to be optimized, and it requires stronger computing power as support, as well as compliance with regulatory requirements and safety redundancy design.

After all, for passenger cars like Tesla, the current autonomous driving technology is more accurately described as "assistive driving", which emphasizes that human drivers must always supervise and take over when necessary. This leaves some leeway for liability determination in case of an accident.

For L4 - level Robotaxis, since the vehicle is completely controlled by the intelligent driving system, once an accident occurs, the entire liability falls on the operator. This is the biggest difference in safety between the two.

Therefore, it is not surprising that Robotaxi companies still choose Hesai while automakers are turning to the pure - vision route.

Hesai has quite a large number of Robotaxi customers. Besides the newly - signed one, it has long established partnerships with eight of the world's top ten Robotaxi companies, including Zoox, Aurora, Apollo, Didi, Pony.ai, and WeRide.

02

As a benchmark enterprise for the multi - sensor fusion route, Waymo's fifth - generation autonomous driving system (Waymo Driver) adopts the most complex perception architecture in the industry: 5 lidars (4 long - range lidars + 1 short - range blind - spot - filling lidar), 8 high - resolution cameras, and the latest 4D millimeter - wave radar array. The lidar point - cloud density of the system reaches 15 million points per second, and it can still achieve centimeter - level accuracy in environmental reconstruction under extreme conditions such as at night, in rain, and in fog.

Behind this "over - engineered" design concept is not only a profound understanding of the "zero - tolerance" requirement for L4 - level systems but also the pressure from regulations and supervision.

In the "Safety Assessment Guidelines for L4 - Level Autonomous Driving Systems" issued by the US NHTSA (National Highway Traffic Safety Administration) in 2025, "perception system redundancy" was clearly listed as a core assessment indicator. The guidelines require that L4 - level systems must have the ability to "maintain safe operation even when any single critical sensor fails".

This requirement essentially establishes the "quasi - rigid" status of lidar in L4 - level systems. Although the regulatory text does not specifically specify a particular sensor type, the unique ranging accuracy and environmental adaptability of lidar make it the optimal solution to meet the redundancy requirement.

Similarly, in the "Automotive Industry Revitalization Plan" issued by the European Commission in March 2025, this trend was further strengthened. It requires that L4 - level vehicles operating within the EU must pass the "extreme weather adaptability test", including the verification of perception ability in scenarios such as heavy rain, heavy fog, and backlighting.

Under these test conditions, lidar often has advantages that vision cannot match.

Beyond the rigid requirements in terms of safety and regulations, the different statuses of lidar in the L4 - level Robotaxi and passenger car markets essentially reflect two completely different business models and liability - bearing mechanisms.

Although Robotaxi operators like Waymo, similar to Tesla, pursue large - scale operations, the prerequisites for their scale - up are different: Since Robotaxi operators bear full legal liability and accident risks, without sufficient safety guarantees, the government will not grant road rights, insurance companies will not provide coverage, and the public will not dare to take the rides.

In this situation, their business model will fall into a vicious cycle of "not allowed on the road → unable to scale up → unable to spread costs → unable to make a profit".

In contrast, the passenger car market follows a completely different development logic. Automakers can adopt a "gradual" strategy, starting from L2 - level assistive driving and gradually improving autonomous driving capabilities through OTA updates. In this model, human drivers still bear the driving responsibility, and the consequences of system failure are relatively controllable.

Therefore, in the passenger car business model, automakers can first accumulate users. After the scale expands, they can continuously improve the system with massive user data and OTA updates.

Tesla's FSD development path perfectly illustrates this model.

In the United States (especially in California and Texas), the supervision of L2 systems is extremely lenient, only requiring "the driver to keep hands on the steering wheel", without mandatory data reporting or in - depth accident investigations.

Tesla takes advantage of this leniency to let its vehicles run the FSD Beta (test version) frequently in the real world, turning public roads into a "free testing ground". Through this, it continuously accumulates data and gradually builds the world's largest real - vehicle data collection network. Its end - to - end neural network, trained with massive real - world driving data, has approached the level of human drivers in some scenarios.

Hesai's growing closeness to Robotaxi companies is not a victory of the technical route but an inevitable choice of the business model. Lidar is chosen by Robotaxi companies not because it is "smarter" than cameras + AI algorithms, but because it is a "safety tax" and an "access ticket".

03

Hesai's decision to conduct an IPO in Hong Kong and obtain large - scale orders at this time is "right on time".

From a regulatory perspective, regulations related to intelligent driving in both China and the United States have been gradually improved. Technically, the centralized computing platforms supporting L4 - level autonomous driving are maturing. NVIDIA's next - generation in - vehicle central computing platform, NVIDIA DRIVE Thor, will be mass - produced in 2025. This platform has a maximum computing power of up to 2000 TOPS and is specifically designed for L4 - level autonomous driving.

Temporally, 2026 is a crucial time window for L4 - level autonomous driving.

However, Hesai has not put all its eggs in the L4 - level basket.

While facing the impact of the pure - vision route in the passenger car market, Hesai is actively deploying in closed and semi - closed application scenarios where lidar technology has natural advantages. Since 2025, its deployment in B - to - B markets such as industrial automation, intelligent logistics, and port operations has significantly accelerated.

A notable example is that driven by the industry trend, AGVs (Automated Guided Vehicles) and AMRs (Autonomous Mobile Robots) are becoming the core components of intelligent manufacturing. Hesai's lidar products demonstrate unique technological advantages in this field.

In 2025, the global AGV/AMR market reached a scale of $6.8 billion, and about 60% of high - end products adopted lidar as the dominant perception solution. Hesai's JT series of mini - lidars are specifically optimized for such applications, supporting millimeter - level positioning accuracy of ±10mm and capable of achieving stable SLAM (Simultaneous Localization and Mapping) functions in complex, high - dynamic scenarios such as warehouses and factories.

Similarly, scenarios such as port automation are also the most in - demand applications of lidar technology in large - scale industries. Different from the complexity of road environments, ports, as relatively closed industrial environments, provide ideal deployment conditions for lidar.

In closed industrial scenarios, lidar has several irreplaceable technological advantages over pure - vision solutions. Since the industrial environment is often accompanied by extreme conditions such as dust, water mist, and strong light, the active detection characteristic of lidar enables it to maintain stable performance under these conditions.

In the field of unmanned delivery, unmanned delivery vehicles, as a typical application in the "low - speed + closed/semi - closed" scenario, are becoming an important breakthrough point for the commercialization of lidar technology. The in - depth cooperation between Hesai and Neolix unmanned vehicles is a typical example of this trend.

The two parties signed a strategic cooperation agreement in September 2021, and the cooperation continues to this day, including the integration of Hesai32/PandarXT series LiDAR. In 2025, Neolix vehicles still rely on Hesai sensors and are in commercial operation in Shanghai and other places.

The reason why lidar can meet the needs of unmanned delivery mainly lies in the particularity of the delivery scenario: in dynamic and complex scenarios (such as intersections and crowded warehouses), visual inference is easily affected by lighting and occlusion, but lidar can reduce the error to within 1 meter.

Technically, the solid - state lidar technology, self - developed chip capabilities, and SLAM algorithm optimization experience that Hesai has accumulated in the in - vehicle lidar field are also applicable in scenarios such as AGV/AMR and unmanned delivery. In some aspects (such as positioning accuracy and stability requirements), it is even easier to achieve technological breakthroughs.

By making simultaneous efforts in multiple fields such as in - vehicle, robotics, and industry, Hesai can spread R & D costs over a larger production and sales scale, accelerate technological iteration, and further reduce manufacturing costs through supply - chain integration.

In 2025, robot and industrial applications accounted for 25% of Hesai's total revenue, and it is expected that this proportion will increase to 40% in 2026. This diversified revenue structure effectively reduces its risk of over - dependence on a single market.

Multi - scenario applications of Hesai lidar

Deep down, Hesai's multi - track layout is actually a response to market changes, aiming to build a machine - perception technology ecosystem centered around lidar and avoid putting all bets on passenger cars.

Of course, technology is interconnected. Through technology verification and iterative optimization in different application scenarios, Hesai is building a technological moat that spans multiple vertical fields. What lies behind this is that under the pressure of sharply reduced costs and significantly shrunk profit margins, lidar technology is forced to evolve from a single - application model to a diversified ecosystem.

As the perception ability brought by lidar penetrates into factories, ports, logistics, and the urban infrastructure, the competition is no longer about the right or wrong of the passenger car intelligent driving technology route. It is about defining the standards and having the say in how the next - generation intelligent agents "see" and "understand" the physical world, and more