Physicl Secures $10M Seed to Bridge Simulation and Robotics with High-Fidelity 3D Worlds

Physicl, a new venture from serial entrepreneur Alex de Vigan, has closed a $10 million seed round led by Inovia Capital. The startup emerges from the team behind nfinite, which spent a decade building a platform that turned messy product data into photorealistic 3D assets for e-commerce. When generative AI began displacing that business, de Vigan recognized a harder problem where the same pipeline could shine: teaching robots to understand and manipulate the physical world.

Why simulation fidelity matters for robotics

Training robots on video alone yields brittle policies. Real-world teleoperation data is expensive and slow to collect. High-fidelity simulation offers a scalable alternative, but only if the virtual worlds obey real physics — correct geometry, material properties, lighting, and contact dynamics. Physicl's Studio product launches with 10,000 curated 3D environments and 100,000 assets, letting engineers generate targeted synthetic datasets for edge cases such as object detection in clutter or under variable shadows.

Early adopter SpAItial, a world-model builder, used Physicl data to train its latest Echo-3 system, which reconstructs 3D simulations from 2D imagery. The collaboration underscores a growing stack: foundation model builders need massive, physically grounded synthetic corpora, and they need tooling to generate it continuously from raw sensor streams — images, LiDAR, video — at trillion-token scale.

Why it matters for GPU / AI infrastructure: Physicl's workflow — ingesting heterogeneous real-world captures, reconstructing high-fidelity 3D scenes, and rendering billions of training frames — is fundamentally a GPU-bound pipeline. Each stage (NeRF/3D Gaussian reconstruction, physics simulation, differentiable rendering, dataset generation) scales with compute. As robotics foundation models grow, demand for on-demand GPU clusters that can burst through petabyte-scale synthetic data jobs will accelerate. AiGpu's infrastructure is positioned to serve exactly this workload profile.