- Web-based viewer streams billion-point 3D environments across devices
- Company says technology could help address AI’s lack of spatial awareness
Chinese spatial computing company Manycore Tech (群核科技) has open-sourced a browser-based 3D viewer capable of streaming city-scale virtual environments without requiring users to install software, as competition intensifies around infrastructure for immersive internet applications and embodied AI.
The company, one of the so-called “Hangzhou Six Dragons” startups, said its Aholo Viewer allows users to explore 3D scenes containing more than 1 billion Gaussian points directly through browsers on smartphones, PCs and VR headsets.
The scale is roughly equivalent to digitizing the entire West Lake, a world-famous tourist attraction, in Hangzhou.
Gaussian points are tiny data units containing color and spatial information that together form photorealistic 3D scenes.
The launch reflects growing efforts by Chinese technology firms to push 3D content consumption into the mainstream, turning immersive virtual environments into experiences that can be accessed as easily as short-form videos through a web link.
Potential applications include digital tourism previews, virtual film production and game development using reconstructed real-world environments, according to the company.
Manycore also said its Aholo spatial intelligence platform now offers open APIs for spatial reconstruction, cloud rendering and AI-generated 3D models, aiming to lower barriers to creating and distributing 3D content.
3D Gaussian Splatting
The underlying technology is based on 3D Gaussian Splatting, a rendering technique that converts real-world objects and scenes into navigable three-dimensional models.
Aholo Viewer uses a streamable level-of-detail, or LoD, architecture that allows large-scale 3D content to be transmitted across devices in a way similar to video streaming.
Industry observers have increasingly viewed the number of Gaussian points a system can load, and the speed at which it renders them, as a benchmark for the practicality of 3D browsers.
In April, Standard professor Fei-Fei Li’s open-source Spark 2.0 engine demonstrated streaming of more than 100 million Gaussian points.
Manycore said Aholo Viewer can process up to 1 billion points while improving rendering speed, smoothness and memory efficiency.
The company argued that the implications extend beyond entertainment or media, positioning 3D internet infrastructure as a way to address a major limitation in AI systems: incomplete spatial understanding of the physical world.
As more real-world 3D data becomes available through platforms such as Aholo Viewer, the company said the datasets could help close training gaps for embodied AI systems.
“For 30 years, the internet has largely transmitted information through two-dimensional formats like text, images and video,” Huang Xiaohuang, Manycore Tech chairman and co-founder, said in a statement.
“A room is reduced to a handful of photos and a product becomes a flat image. We hope open 3D infrastructure such as Aholo Viewer can help move the internet into the 3D era,” he added.
