Skild AI’s S1 Uses NVIDIA Tech to Teach Robots via Video

Skild AI has unveiled its S1 robotic foundation model, leveraging NVIDIA’s AI infrastructure to enable robots to learn entirely new tasks from a single video demonstration. This breakthrough, announced last week, aims to make robots significantly more adaptable across industries like manufacturing, logistics, and security without the need for extensive retraining or reprogramming.

S1’s key innovation lies in its ability to use video as a prompt for “in-context learning.” An operator records a task, such as plant potting or coffee brewing, and feeds it to the model. From there, the robot interprets the objects, intent, and sequence of actions, autonomously translating them into executable steps. This approach eliminates the need for updating model weights or conducting task-specific training runs.

According to Skild, S1 can complete complex tasks lasting up to 10 minutes, often involving dozens of sequential actions. In internal tests, the model executed tasks successfully 66% of the time—an over sevenfold improvement compared to similar AI systems, which achieved a 9% success rate.

Fast-Tracking Real-World Deployment

The S1 model builds on a broader strategic partnership with NVIDIA, incorporating technologies like NVIDIA Isaac Lab and NVIDIA Cosmos for simulation, synthetic data generation, and scalable training. Skild’s robots move from video demonstration to real-world execution in as little as 11 minutes, a dramatic reduction in setup time compared to traditional industrial robots.

Notably, the model’s adaptability has already seen commercial traction. Skild reached a $100 million annual revenue run rate within 10 months of its first deployment and now counts over 60 partnerships. One high-profile collaboration includes using Skild’s intelligence in dual-arm manipulators for Foxconn’s assembly of NVIDIA Blackwell systems—tasks requiring high precision and real-time adaptation.

A $14 Billion Robotics Powerhouse

Skild AI’s rapid rise underscores its unique position in the robotics sector. Founded in 2024, the Pittsburgh-based company has raised over $2 billion to date, including a $1.4 billion Series C round in January 2026, led by SoftBank and NVentures, NVIDIA’s investment arm. The funding round valued Skild at over $14 billion, making it one of the most valuable private companies in robotics AI.

Skild’s acquisition of Zebra Technologies’ robotics division earlier this year further signals its intent to expand into warehouse and logistics applications. While many robotics firms focus on specific hardware, Skild designs its AI as a “robot-agnostic brain,” capable of working across multiple environments and robot types.

NVIDIA’s Role in Scaling Robotics

NVIDIA’s technology plays a central role in Skild’s development pipeline. Tools like NVIDIA Omniverse and Isaac Sim provide virtual environments for training robots on diverse scenarios, while the Newton physics engine simulates real-world interactions. NVIDIA’s TensorRT optimizes the model’s inference, enabling robots to respond quickly in dynamic conditions.

This collaboration highlights NVIDIA’s broader ambitions in the growing field of physical AI. By integrating simulation, training, and deployment into a unified framework, NVIDIA accelerates the transition of adaptable robotics from research labs to industrial floors.

What’s Next?

With adaptable robotics gaining momentum, Skild AI’s S1 and its NVIDIA-powered infrastructure could redefine automation across multiple industries. While the company remains private, its rapid market penetration and partnerships with major players like Foxconn suggest it’s positioning itself as a key enabler of the next wave of industrial automation.

As robotics adoption continues to grow, Skild AI’s ability to scale its general-purpose “robot brain” may set a new benchmark for the industry. Investors and industry stakeholders will undoubtedly watch closely as the company moves deeper into real-world applications.

Image source: Shutterstock