Hesai Technology Unveils Kosmo 3D World Data Engine, 8K Spatial Camera and Cloud Platform

Hesai Technology (NASDAQ: HSAI; HKEX: 2525) on Oct. 14, 2026, announced Kosmo, a 3D world data engine, and launched Kosmo One, described as the world’s first 8K ultra-high-definition AI spatial camera, and the Kosmo World cloud platform. At the launch event, it announced strategic partnerships with Galbot and Yuanli Lingji and presented use cases for Pixar Animation Studios and Amap. Hesai is positioning the products as a new line focused on a real-world data platform for physical AI.

Hesai is seeking to repurpose measurement technology developed for LiDAR to supply 3D data for robot training and world models. The company said in its presentation materials that it can scan 100 square meters in about 15 minutes and model it at about five times the speed of comparable products. Whether it can reduce the cost of digitizing real-world environments will be a key question.

Kosmo’s innovation lies not in a standalone sensor or imaging device, but in a three-layer system that integrates capture, reconstruction and delivery. Kosmo One measures physical spaces, Kosmo World converts them into 3DGS, mesh and depth formats, and KSC performs semantic segmentation. The company is emphasizing not a “device that takes pictures” but a “data platform on which physical AI can be trained.” The offering extends Hesai’s LiDAR business, but shifts the focus from point clouds and distance measurements to sim-ready data for robot training and world models. In our related coverage, Counterpoint Forecasts a Sharp Increase in DRAM Demand for Humanoid Robots by 2030 noted the potential for memory demand to grow on the training and inference side of humanoid robotics. This time, Hesai is seeking to control the input side: the generation of high-quality 3D data from the real world. Alongside the edge computer for robotics in Asmeck Begins Offering the AIE-KT78/68 Edge AI Computer for Robots, Compatible With NVIDIA Jetson Thor and the computing platform in Ubitium Raises $14 Million to Develop a General-Purpose Processor for Physical AI, the move suggests that competition in physical AI is expanding beyond computing systems to data-generation devices. From an industry perspective, Hesai’s connection to existing development platforms such as NVIDIA Isaac Sim, Unreal Engine and RenderMan is significant. Rather than locking users into a proprietary format, the design feeds captured real-world environments into downstream simulation and production workflows, potentially making it a starting point for supply chains spanning robotics, video production and location-based services. Examples involving Amap’s “Flying Street View” and Pixar RenderMan show how the same 3D data can be used for training, production and display. However, what has been disclosed so far is limited to the product launch and examples; actual mass-production shipment volumes, the share of recurring revenue and commercial terms with partners remain unclear. The next areas to watch are Kosmo One shipment volumes, continued use of Kosmo World, the extent to which data generated by KSC is standardized, and whether partnerships with Galbot and Yuanli Lingji move to implementation. In particular, the degree to which the data is reproducible for robot training and transferable across different machines and environments could determine whether Kosmo remains a one-off demonstration or becomes a platform business.

Kosmo One integrates high-resolution LiDAR, a 1-inch image sensor and an IMU, synchronizing the capture of geometry, color, texture and camera position at the microsecond level. It uses 3D Gaussian Splatting for reconstruction and claims millimeter-level geometric accuracy. One operator can scan an indoor area of about 100 square meters in about 15 minutes, and Hesai says Kosmo World creates 3D models from the captured data at about five times the speed of comparable products. Kosmo World generates 3DGS, mesh and depth formats from a single reconstruction, and fills in unscanned areas, reconstructs materials and textures, and performs object-level semantic segmentation. Hesai also cited integration with NVIDIA Isaac Sim, Isaac Lab, Unreal Engine, RenderMan and Houdini. It outlined two business models: offering hardware and software together, and providing high-quality spatial data generation services for world models and physical AI.

For Japanese companies working with 3D data in robotics, autonomous driving and video production, an integrated system for capturing and reconstructing real-world environments could become a competitive factor. Hesai’s stated integrations with NVIDIA Isaac Sim, Unreal Engine and RenderMan are likely to increase the importance of data platforms that work with existing development environments.