Zibra AI is a technology company building a GPU-native data layer, with its core being compression technology for 3D and 4D data, aiming to keep simulation, VFX, and physical AI datasets compact, streamable, and directly usable by GPU applications.
In terms of use, Zibra positions itself as "the data layer between physical reality and the virtual world." It believes the next generation of computing is becoming spatial: AI is learning to understand the physical world, while digital experiences are becoming richer, more dynamic, and more simulation-dependent. 3D and 4D datasets are 1,000 to 10,000 times larger than traditional media, which makes storage, transmission, and processing a fundamental infrastructure challenge. Compression therefore becomes the key layer that makes spatial data scalable in practice. Smaller data means faster workflows, lower infrastructure costs, and entirely new real-time use cases.
Its main capabilities revolve around its compression technology. The officially listed research topics include: random-access compression for unstructured meshes, using compressed page layouts so that spatial regions, fields, and time steps can be addressed independently without decompressing the entire dataset; streaming 4D simulation data from object storage, using a progressive loading strategy to provide the first required time step before the rest of the time-varying dataset has finished transferring; and training point-based physical models on compressed data, reading point attributes from a full mesh dataset without fetching topology, thereby transforming data loading throughput.
The products are divided into three directions that share the same compression core. ZibraXYZ targets physical reality, providing a data layer that enables large-scale simulation and sensor datasets to be stored, moved, visualized, and fed into AI workflows faster, with the positioning of removing the data bottleneck for physical AI. ZibraVDB and ZibraGDS target the virtual world, helping studios and real-time creators use more complex 3D content, accelerate production pipelines, and bring simulation-quality assets into interactive experiences that were previously impractical.
In terms of usage, the company states that these products are currently provided to studios and engineering teams through licensing. The page has a "Talk to our engineers" entry point. Regarding specific installation commands, integration steps, supported platforms, and pricing, the materials do not yet provide details.
Applicable scenarios include: storage, transmission, and training data supply for large-scale simulation and sensor data in physical AI and industrial computing; compression and playback of high-quality volumetric effects and animated geometry in film, television, and game production; and simulation-grade 3D content that requires real-time interaction. The company says it is trusted by leading global enterprises and mentions participation in industry events, receiving the VDS Growth Stage award, SIGGRAPH Real-Time Live! live demonstrations and an Epic MegaGrant, and it was also shortlisted among 50,000 startups worldwide in The Pitch by Deel. The core team comes from AMD, Ubisoft, Nokia, Gameloft, Blackwood Games, Pingle, and Arcadiium, working on rendering, simulation, and large-scale data infrastructure.
In terms of limitations, the materials do not specify the exact billing model, pricing, free quota, supported engine versions, operating systems, or hardware requirements, nor do they provide detailed usage tutorials. This information is not yet available.
