Physical AI News (Japan)

Physical AI, robotics and compute — reported from Japan.

日本語
· Deployment & production· 日本語で読む

RoboSense sells 595,600 LiDAR units in July-September 2026, connects to NVIDIA HSB

RoboSense reported sales of 595,600 LiDAR units in July-September 2026, a quarterly record. Sales for the first three quarters totaled 1,314,800 units, exceeding its full-year 2025 total of 912,000. The company said 194 vehicle models had secured sourcing nominations and 81 were in the mass-production delivery stage. It also said it supplies more than 3,400 robotics companies. In September, its E2 and EM4 connected to NVIDIA Holoscan Sensor Bridge, enabling low-latency transmission of 3D sensing data.

What happened

RoboSense (速腾聚创, 2498.HK) said Oct. 9, 2026, that it sold 595,600 LiDAR units in July-September 2026. Sales rose 220.9% from a year earlier and 53.1% from the previous quarter, setting a quarterly record. Sales for the first three quarters totaled 1,314,800 units, already exceeding the 912,000 units sold in all of 2025.

Why it matters

In its Oct. 9, 2026, announcement, RoboSense reported sales and partnerships across both automotive and robotics markets. Its announcement of a connection to NVIDIA Holoscan Sensor Bridge shows LiDAR beginning to be treated not as a standalone component, but as a part expected to connect to an edge AI computing platform.

Analysis

The focus is not just sales growth but the simultaneous expansion of two business pillars: automotive ADAS and robotics. Of the 595,600 units sold in the quarter, 450,400 were for ADAS and 145,200 were for robotics and other uses. The robotics and other segment also grew 309.0% year over year. This suggests RoboSense's growth is supported not only by automotive mass production but also by the expansion of 3D sensing into robotics. A related article by this newspaper, Jabil discusses progress in humanoid development and mass production, report says, examined the shift from prototyping to design and manufacturing geared toward mass production. RoboSense's figures show that this shift is already advancing on the component side. The combination of sourcing nominations for 194 vehicle models, mass-production deliveries for 81 models and mass-production partnerships with 36 companies indicates that sensing modules have moved from development components to vehicle-installed parts. However, the company did not disclose which models adopted which specifications. The connection to NVIDIA HSB is more than a product addition: it links the input and output paths for sensing data to NVIDIA's edge AI computing platform. This signals a shift in emphasis from competition over sensor performance alone to competition among systems spanning sensors, computing platforms and multisensor fusion. RoboSense, which has participated in the Jetson, DRIVE and Omniverse ecosystems, has now extended that work to real-time data connections through HSB. For robotics, data acquisition and training will be key issues; for automotive applications, mass-production qualification will be central. Questions remain. The information disclosed does not show which configurations among the 194 vehicle models have actually entered mass production, the data latency or integration method after the HSB connection, the recipients of E2 mass-production deliveries, or which applications are driving revenue among the more than 3,400 robotics customers. See the original source for details.

Implications for Japan

The 36 automakers and Tier 1 suppliers include BYD, Geely, Volkswagen, SAIC Motor, FAW, Xiaomi, Great Wall Motor and Toyota. Mass-production qualification for automotive LiDAR could therefore affect vehicle and component procurement planning, including among Japanese companies. If configurations such as the HSB connection, in which sensing data flows into NVIDIA's computing platform, become more widespread, connectivity with in-vehicle software and surrounding systems—not just the sensor itself—could become a competitive requirement.

Sources

Translated from the Japanese original: /news/robosense-2026-nen-7-9-gatsu-no-lidar-hanbai-59-56-man-dai-n-7d3cf0. The Japanese version is authoritative if they differ.