The brief

Analyst view

• Builds frontier models for spatial intelligence: AI that perceives, generates and interacts with virtual and physical worlds.
• Marble (Nov 2025) creates interactive 3D worlds. Atlas (Sep 2026) adds camera control and temporal simulation.
• Raised $1B from AMD, NVIDIA, Fidelity and others in Feb 2026; joined AMD in Sep 2026.

Technical approach

As reported

Multimodal Architecture

Omni models for text, image, video, 3D

Models pretrained to natively process text, images, video and 3D structure without separate encoders.

Spatial Reconstruction

3D from sparse images

Reconstructs real 3D scenes from one or two images, generating camera viewpoints and 3D outputs.

Temporal Modeling

Space-time simulation for robotics

The models handle both spatial structure and temporal dynamics, enabling applications like video reframing, physics simulation and robotic arm prediction.