Robots "feed" Hesai, is the next stop for lidar the era of embodied intelligence?

The “Unconventional” Future of LiDAR.

Author | Wang Xiaojuan

Editor | Zhou Zhiyu

March 24, Hesai Technology released its unaudited financial results for Q4 2025 and the full year.

Amid years of skepticism about “burning money for scale,” Hesai boosted confidence in the LiDAR industry with a record-breaking annual revenue of 3.03 billion RMB (up 45.8% year-over-year) and its first full-year GAAP profit.

In 2025, when debates over pure vision versus LiDAR in autonomous driving technology intensified, the release of this performance is thought-provoking.

On one hand, LiDAR is accelerating its adoption with the penetration of intelligent driving; on the other, in a market where automakers are fiercely cutting costs, LiDAR remains an expensive hardware that is most easily optimized out.

Is LiDAR an essential component of the intelligent driving era, or a luxury during the transitional phase of technology? As the penetration rate of intelligent driving increases, the market is providing an answer; meanwhile, LiDAR hardware seems to have found more intelligent entities beyond cars, opening new markets for itself.

01 Profitability at Scale

What’s most noteworthy about Hesai’s financial report isn’t just revenue growth, but the change in its profit.

For hardware manufacturers, “diseconomies of scale” has long been a challenge for all LiDAR vendors. In recent years, high R&D costs, BOM expenses, and the pressure from automakers to reduce prices have trapped many companies in a “sell more, lose more” dilemma.

In 2023, Hesai reported a loss of 840 million RMB, and TSTC (Tengsheng Juchuang) lost 760 million RMB, plunging the industry into a “loss-making for reputation” predicament. However, Hesai’s 2025 financials prove that this business can be profitable.

Hesai became the world’s first LiDAR company to achieve full-year GAAP profitability, thanks to its fundamental technological restructuring. Traditional LiDAR relies on numerous discrete components, making assembly complex and costly.

Hesai’s in-house chip development for transceivers and processing units (ASIC) not only significantly reduces physical size but also brings core component costs onto the Moore’s Law trajectory.

According to the company, through continuous innovation in chip technology, Hesai has reduced LiDAR costs by 99.5% over eight years. By November 2025, its self-developed chips had delivered a total of 185 million units, ranking first globally. In 2025, Hesai launched its self-developed main control chip “Fermi C500,” completing the last piece of its full-stack self-research.

This ability to transform “precision optical instruments” into “semiconductor-standardized products” is key to maintaining gross margins in price wars. Despite ASP (average selling price) continuing to decline, Hesai’s overall gross margin remained stable at 41.8% in 2025.

Meanwhile, industry segmentation is accelerating. Hesai’s annual shipments surged, further squeezing out small and medium-sized competitors.

In the current supply chain system, automakers have very low tolerance for supplier errors. Once a leading company surpasses the million-unit production threshold, its advantages in bargaining power, yield control, and delivery stability will only grow. This means that second- and third-tier LiDAR companies that have yet to achieve self-sustaining revenue will face survival issues rather than growth challenges by 2026.

02 The Ongoing Route Battle

Contrasting Hesai’s strong performance is the polarization of attitudes toward LiDAR in the automotive industry in 2025.

While shipment volumes are increasing, this is largely due to the rising penetration of LiDAR in mainstream models priced between 150k and 200k RMB, driven by the proliferation of intelligent driving features. For example, in 2025, LiDAR penetration in new energy passenger vehicles reached 17%, crossing the critical “chasm” of 16% for the first time (a threshold considered by the book “Diffusion of Innovations” as the point where a new technology can truly “break through” and become widespread). This indicates that the growth is driven more by market penetration than by absolute demand from automakers.

Currently, the fierce competition in China’s new energy vehicle market has extended to every component. Under immense pressure to cut costs, LiDAR priced at hundreds of dollars per unit remains a prime target.

Three years ago, a LiDAR costing 5,000–6,000 RMB now has plummeted to 1,500–3,000 RMB, a drop of over 70%. Leading manufacturers are further reducing prices toward the $200 (about 1,400 RMB) range, aiming for application in vehicles priced around 150k RMB.

Wall Street Insights learned that some automakers, including XPeng, which have long focused on intelligent driving as a core selling point, are now deploying “light” or even no-LiDAR autonomous driving solutions by enhancing pure vision algorithms and onboard computing power. Since 2024, models like XPeng MONA M03, P7+, and the new G6/G9 have all eliminated LiDAR.

XPeng’s autonomous driving director Yuan Tingting told CarNewsChina last year: “Removing LiDAR is a clear choice.” The logic is that new AI systems based on large language models trained on massive data cannot effectively incorporate LiDAR data.

He Xiaopeng even predicted: “The choice between the two paths might be an issue now, but by 2027, it might not be.” Meanwhile, companies like Huawei, Li Auto, and NIO still use multi-sensor fusion, considering LiDAR an indispensable feature for flagship models.

However, many still see LiDAR as a “safety fallback,” and public perception is shifting. Two years ago, roof-mounted “horns” symbolized high-end and technological sophistication. Today, equipping a vehicle with LiDAR is about ensuring “safety redundancy” that cannot afford to fail.

Huawei’s executive director Yu Chengdong explained in a 2025 science popularization video: cameras struggle in backlit, nighttime, or foggy conditions; millimeter-wave radar has limited recognition accuracy; but LiDAR can generate 3D point clouds by emitting laser beams, providing sufficiently high recognition precision.

His conclusion was that only integrated perception solutions combining three hardware types can provide comprehensive safety for assisted driving. Hesai CEO Li Yifan likened LiDAR to an “invisible safety airbag,” emphasizing its shift from a functional component to a safety component.

Analyst Zhang Chenghang from Chuanchuang Securities recently predicted that as the trend toward equal rights in intelligent driving develops, LiDAR may reach a scale-up inflection point. The global onboard LiDAR market is expected to reach $9 billion by 2030 and $14.8 billion by 2035.

Currently, automakers still choose whether to equip LiDAR based on their technological maturity and cost considerations. The ongoing debate in the LiDAR field has yet to be settled.

03 The Endgame of AI Leap

Looking further ahead, considering current autonomous driving trends, the long-term outlook for the LiDAR industry faces a contradiction: the ultimate leap in intelligent driving technology may actually reduce reliance on high-precision physical sensors like LiDAR.

2025 is the year when end-to-end autonomous driving models are expected to be widely deployed. When systems no longer depend on manually written rule-based code but are trained directly on massive video data to realize seamless perception–decision–control loops, the demand for 3D high-precision point cloud data will diminish.

If visual AI becomes sufficiently powerful to infer 3D space and depth from 2D images like humans do, the value of LiDAR will be diluted.

The logic behind Tesla and XPeng’s shift toward pure vision is precisely this—high-computing chips enable processing of vast amounts of image data, while LiDAR point clouds become difficult to incorporate into end-to-end models.

Therefore, as autonomous driving algorithms mature, dependence on LiDAR may weaken.

Even if industry consensus dictates that LiDAR must be retained for that 0.01% safety margin, its future growth logic will change. With product homogenization, LiDAR could become a standardized industrial product for volume sales, with diminishing hardware premium margins for manufacturers.

However, policy incentives for L3 autonomous driving are creating new variables for LiDAR.

In December 2025, the Ministry of Industry and Information Technology announced the first batch of L3 autonomous vehicle access permits, with Changan Deep Blue SL03 and BAIC Arcfox Alpha S (L3 version) successfully approved for pilot testing. This shift of responsibility from drivers to automakers makes LiDAR an essential rather than optional component, with vehicle installations potentially increasing from one to three or six units, opening new opportunities for the industry.

For leading companies like Hesai, the key to breaking through may lie outside the vehicle. In 2025, Hesai’s laser radar deliveries for robotics reached about 240k units, a 425.8% increase year-over-year. This growth far exceeds that of onboard applications. TSTC’s robotics business also performed well, with a gross margin of 45%, much higher than the 17.4% for vehicle-mounted products.

It is reported that Yushu Robotics’ entire lineup is equipped with Hesai’s JT series LiDARs, which were showcased on the 2026 Spring Festival Gala, with over 200k units delivered; Hesai also signed a supply contract for 150k JT radars with QuMi Ecology, setting a record in the consumer robotics field.

With the development of general artificial intelligence, embodied intelligent robots and driverless logistics vehicles are on the cusp of explosive growth, and LiDAR will have broad application prospects. Morgan Stanley predicts that by 2050, global demand for robotic LiDAR will increase nearly 300 times compared to 2025.

Unlike cars on structured roads, robots operating in complex, unstructured 3D environments require more precise spatial perception. This non-automotive market is becoming the next blue ocean for LiDAR capacity.

Hesai Technology’s strong 2025 financial report demonstrates its victory in the first half of the hardware elimination race—building cost advantages through chip self-research and scale effects to dilute fixed costs, and achieving profitability ahead of competitors in the LiDAR race.

But in today’s era of rapid AI model advancement, the industry’s ultimate opponent is no longer just peer competitors but the continuously evolving “pure vision AI.”

Hardware cycles will ultimately depend on the pace of software development.

Risk warning and disclaimer

Market risks exist; investments should be cautious. This article does not constitute personal investment advice and does not consider individual users’ specific investment goals, financial situations, or needs. Users should consider whether any opinions, views, or conclusions in this article are suitable for their particular circumstances. Invest at your own risk.