AI Technology Observations | 15 October 2026, 07:00

This observation covers technical facts published and verified around the period from 7:15 am on 14 October to 7:15 am on 15 October 2026 in Sydney. Two items are included.

Hesai formally launches the Kosmo 3D world data engine

Hesai Group formally launched Kosmo on 14 October, moving its previously demonstrated spatial-capture concept into a product system combining the Kosmo One spatial camera and Kosmo World cloud platform. Kosmo One integrates high-resolution lidar, a one-inch image sensor and an inertial measurement unit. The cloud platform can produce 3D Gaussian Splatting, mesh and depth data in one reconstruction, use generative processing to fill unscanned areas, and apply semantic segmentation to identify objects. Users are given access to raw capture data and 3D assets, and Hesai lists integrations with NVIDIA Isaac Sim, Isaac Lab, Unreal Engine, RenderMan and Houdini.

Hesai says one person can typically scan an indoor space of about 100 square metres in roughly 15 minutes, while cloud model generation can be up to about five times as fast as comparable 3DGS products under similar hardware, dataset and quality requirements. The company also reports a 99 per cent task-success rate for a Sharpa robot deployed at a Dairy Queen store in Shanghai. These speed, accuracy and deployment results are developer-reported. Reuters independently confirmed the launch, sensor combination, agreements with Chinese robotics companies and the store deployment, but did not independently reproduce the performance figures. Hesai states that 8K is the capture maximum while user-exported files are limited to 6K, and that scan time varies with layout, area and capture requirements. Public materials do not disclose hardware pricing, cloud-service pricing, delivery regions or independent evaluation results.

Google provides limited model and compute access to European researchers

Google DeepMind and Google Cloud announced on 14 October that up to 200 European scientific researchers will receive direct access to their frontier models through the European Union's RAISE initiative, together with a total of US$3 million in compute credits. The available tools include AlphaFold 3, WeatherNext, AlphaGenome, AlphaEvolve and AlphaEarth Foundations. Whereas those tools had previously been released separately or placed in specific services, the new fact is a bounded research program that combines model access with cloud-compute credits. It is not a general release to the public or to every European research institution.

Google's notice uses both present-tense provision and pledge language, while Reuters independently confirmed the participant ceiling, total credit amount and tool list. The available material does not provide an application, selection or activation timetable, nor does it specify per-researcher credit allocation, API permissions, data-processing terms or the functions enabled for each model. The program therefore cannot yet be described as 200 researchers already receiving service.

Sources

Hesai Group, “Building the Data Infrastructure for Physical AI: Hesai Launches Kosmo 3D World Data Engine”, 14 October 2026, https://www.hesaitech.com/9974-2/

Reuters, “China's Hesai launches spatial camera to boost AI-powered robotics adoption”, 14 October 2026, https://www.reuters.com/technology/chinas-hesai-launches-spatial-camera-boost-ai-powered-robotics-adoption-2026-10-14/

Google, “We’re powering European scientific discovery with advanced AI compute and frontier models”, 14 October 2026, https://blog.google/innovation-and-ai/models-and-research/google-deepmind/were-powering-european-scientific-discovery-with-advanced-ai-compute-and-frontier-models/

Reuters, “Google DeepMind, Google Cloud to allow European researchers access to frontier models”, 14 October 2026, https://www.reuters.com/technology/google-deepmind-google-cloud-allow-european-researchers-access-frontier-models-2026-10-14/