#USA #San Jose #Nvidia GTC #Physical AI #Physicl

Physicl Gem Lights Up NVIDIA GTC with a New Data Infrastructure Layer for Physical AI

On March 18, 2026, Physicl stepped into the spotlight at NVIDIA's GTC, unveiling its latest data infrastructure platform, crafted specifically for Physical AI and robotics. This ambitious initiative aims to solve a growing bottleneck in the AI industry as it shifts from language models and image recognition to more interactive systems capable of engaging with physical spaces.

Founded by members of the team behind Nfinite, renowned for its high-fidelity 3D digital twins, Physicl is now solely dedicated to scaling production-ready data for embodied intelligence. As the focus of artificial intelligence evolves, the necessity for physically grounded, simulation-capable 3D data has become paramount.

According to Alex de Vigan, CEO of Physicl, "Every major leap in AI has demanded a new layer of data. For Physical AI, this layer encompasses a structured and spatially coherent database grounded in physics, which allows models to genuinely learn. Physicl aims to establish this foundation so that robots and world models can comprehend their surroundings, simulate environments, and operate reliably in the real world."

Driving the Next Wave of Physical Intelligence

Physicl is setting its sights on three converging pathways:
1. Robotics: Training embodied systems to navigate and manipulate environments through physics-accurate simulations.
2. World Models: Enabling spatial reasoning and causal understanding with structured 3D environments.
3. Vision-Language Models (VLMs): Founding multimodal systems on physically coherent, simulation-capable data.

With almost a decade's worth of experience in digitalizing the physical world, Physicl enables continuous, production-ready data pipelines that elevate Physical AI beyond one-time datasets to scalable infrastructure suitable for global application. Organizations such as Meta, DeepMind, World Labs, and Getty Images are already leveraging its platform, which integrates three interconnected infrastructure layers:

• - Data Normalization: Transformation of visual inputs into structured, spatially consistent 3D representations.
• - Physical Data Augmentation: Generation of large-scale, IP-safe domain randomization based on physical realities.
• - DataSim Pipelines: Creating simulation-capable environments for AI training and facilitating the transfer from simulation to reality.

Harnessing NVIDIA Technology for the Omniverse Ecosystem

To scale Physical AI, Physicl utilizes NVIDIA’s Physical AI Stack. Key components include NVIDIA Omniverse, Isaac Sim, Isaac Lab, and Cosmos. As developers of physical AI seek production-ready 3D environments to train and evaluate their models, the current reliance on slow and costly internal data generation methods is becoming inadequate.

Physicl is poised to bridge this gap with:

• - Omniverse-Ready Assets: Simulation-ready 3D environments in OpenUSD that can be seamlessly integrated into Omniverse workflows.
• - Isaac Sim and Lab: Physically accurate environments optimized for robotic manipulation, navigation, and long-term task training.
• - Cosmos-Compatible Data: Structured 3D environments aligned to the world foundation model post-training.

Embarking with a library of millions of simulation-capable 3D assets and environments, Physicl promises a steady supply of simulation-ready 3D data, drawing parallels to how Getty Images provides licensed visual content or how Scale AI offers labeled training data.

A New Paradigm for Training Inputs

While models powered by GPUs generally require conventional training inputs, Physical AI calls for a radically different type of data. Industry leaders are increasingly stressing the necessity for the next generation of AI to utilize exponentially larger volumes of high-quality data. For systems that must interact with physical spaces, this data must encapsulate geometry, physics, and spatial relationships.

Alex states, “NVIDIA has established the computational and simulation infrastructure for Physical AI. Physicl is designed as a dedicated data layer that will fuel the next wave of AI and robotics.”

Presenting at NVIDIA GTC

Physicl showcased its platform at the NVIDIA GTC from March 16 to 19 in San Jose, California, illustrating how simulation-capable assets can be applied in NVIDIA Omniverse and Isaac Lab to expedite the development of robotics and Physical AI. Attendees can visit Booth 3307 to learn more about their offerings. Developers and researchers interested in early beta access can apply on Physicl's website.

For further information, visit Physicl's website.