Physicl Launches to Build the Data Foundation for Physical AI

SAN JOSE, CA – March 17, 2026 – A new company named Physicl emerged from stealth today at NVIDIA's GTC conference, unveiling an ambitious plan to build the foundational data layer for the next era of artificial intelligence. The startup, launched by veterans from the 3D digital twin company Nfinite, is introducing a data infrastructure platform purpose-built to create and scale the high-fidelity, simulation-ready 3D data required to train robots and other forms of "Physical AI."

As the AI industry pushes beyond language and image generation towards systems that can perceive, understand, and interact with the physical world, a significant roadblock has emerged. Training these embodied systems requires a new class of data that is not only vast but also spatially consistent and grounded in the laws of physics. Physicl aims to be the primary provider of this critical resource.

"Every major advance in AI has required a new data layer," said Alex de Vigan, CEO of Physicl, in the company's launch announcement. "For Physical AI, that missing layer is structured, spatially consistent, physics-aware data that models can actually learn from. Physicl exists to build that foundation — enabling robots and world models to understand space, simulate environments, and ultimately operate reliably in the real world."

The Data Bottleneck in Embodied AI

The challenge facing developers of robotics and autonomous systems is immense. While large language models were trained on the vast expanse of text and images on the internet, no such readily available dataset exists for physical interaction. Collecting real-world data for a robot to learn tasks like grasping objects or navigating a warehouse is painfully slow, prohibitively expensive, and fails to capture the long tail of rare but critical edge cases. A robot trained only on what it has seen in one specific environment will struggle when faced with novel situations.

This has led many leading AI labs and technology companies to build their own bespoke 3D data and simulation environments in-house. It is a costly, time-consuming process that requires specialized teams and diverts resources from core model development. This "data bottleneck" is widely seen as a key factor limiting the pace of innovation in embodied AI.

Physicl's platform is designed to break this bottleneck. The company offers a continuous, production-grade pipeline for what it calls "world-ready data." This involves three core infrastructure layers: a data normalization engine that converts various visual inputs into structured 3D representations, a physics-aware augmentation system to generate large-scale variations for robust training, and simulation-ready data pipelines that can be fed directly into training environments.

From Digital Twins to World Models

The team behind Physicl is not new to the challenges of digitizing the physical world. CEO Alex de Vigan previously led Nfinite, a company that carved out a niche by creating one of the largest pipelines of high-fidelity 3D digital twins, primarily for the e-commerce and retail sectors. Nfinite's technology allowed brands to create photorealistic 3D versions of their products, transforming online shopping experiences.

This experience provided the team with deep expertise in creating scalable 3D asset pipelines. A key strategic move during this period was a collaboration with Getty Images, announced earlier this year, to convert parts of Getty's massive 2D image library into physically-contextualized 3D scenes. The partnership was explicitly aimed at creating responsibly sourced, high-quality data to train spatially-aware AI systems, foreshadowing the mission now fully realized with Physicl.

The launch of Physicl represents a strategic pivot from commercial visualization to a more fundamental technological challenge. By leveraging nearly a decade of experience in 3D data generation, the team is now applying its expertise to the far more complex domain of embodied intelligence, where data must encode not just appearance but also geometry, material properties, and physical behavior.

The Data Layer for NVIDIA's Physical AI Ecosystem

Physicl's launch at NVIDIA GTC is no coincidence. The company is positioning itself as an essential component within NVIDIA's rapidly expanding ecosystem for Physical AI. As NVIDIA builds the "compute and simulation infrastructure" with platforms like Omniverse, Isaac Sim, and Isaac Lab, Physicl aims to be the dedicated "data layer" that fuels them.

The integration appears to be deep and strategic. Physicl is offering Omniverse-ready assets in the OpenUSD format, allowing developers to plug structured 3D environments directly into their workflows. Its environments are optimized for robotics training in Isaac Sim and Isaac Lab, supporting tasks from manipulation to navigation. Furthermore, the company's data is being structured to be compatible with NVIDIA's Cosmos, a set of world foundation models designed for generating synthetic data.

This alignment makes Physicl a critical enabler for the thousands of developers building on NVIDIA's stack. Instead of spending months creating a single simulated environment, a robotics company could theoretically source a continuous stream of diverse, physically accurate, and IP-safe environments from Physicl, much like a creative agency licenses stock photos from Getty Images or an AI company uses Scale AI for data labeling. The company launches with a library of millions of simulation-ready 3D assets and environments, providing the initial scale needed to attract developers.

Powering the Next Wave of Intelligent Systems

Physicl's platform is designed to support three converging areas of AI development. The most immediate is robotics, where simulation is essential for training embodied systems to perform tasks in the real world. The second is "World Models," a burgeoning field of AI research focused on creating models that can reason about and predict outcomes in a physical space. Finally, the platform aims to ground Vision-Language Models (VLMs) in physically coherent data, allowing them to understand and describe scenes with greater accuracy.

The company's launch materials state that its platform is already being used by teams at major technology organizations, including Meta, DeepMind, World Labs, and Getty Images. While specific use cases from Meta and DeepMind remain under wraps, which is common for a company just exiting stealth, the connections to others are clearer. The relationship with Getty Images is a continuation of the work started at Nfinite. World Labs, which uses NVIDIA Isaac Sim to validate its own generative world models, represents a natural user for Physicl's simulation-ready data within the NVIDIA ecosystem.

By focusing on this specialized, yet critical, piece of the AI puzzle, Physicl is making a significant bet that the future of AI is not just in digital spaces but in the physical world. If that bet pays off, providing the foundational data for this transition could make Physicl an indispensable part of the industry's infrastructure. Developers and researchers can apply for early beta access via the company's website, signaling the start of what could be a new phase in the development of truly world-aware AI.