Skild AI: Building a Universal Brain for Robots or “Omni Bodied” foundation model

Skild AI: Building a Universal Brain for Robots

Skild AI is a Pittsburgh based robotics company founded in 2023 that is developing a general purpose artificial intelligence system designed to power almost any robot. Rather than building specialized software for individual machines or tasks, the company is creating what it calls an “omni bodied” foundation model, a single shared intelligence that can adapt to different robot bodies and environments with minimal or no additional training.

Founders and Origins

The company was started by Deepak Pathak (CEO) and Abhinav Gupta (President), both former Carnegie Mellon University faculty members and researchers who previously worked on robotics and AI at Meta. Their background in self supervised learning and adaptive robotics shaped Skild’s core approach: treating physical intelligence more like large language models treat language, with a heavy emphasis on scale, generalization, and continuous improvement from real world use.

The Skild Brain and S1 Model

At the heart of Skild’s technology is the Skild Brain, powered by its flagship foundation model known as S1. The system is designed to work across a wide range of robot morphologies, including industrial arms, mobile manipulators, quadrupeds, and humanoids, without needing prior knowledge of the specific hardware.

A notable capability of S1 is in context learning for long horizon tasks. Instead of requiring extensive task specific fine tuning or thousands of demonstrations, the model can watch a single video of a human performing a task and then attempt to execute it. Demonstrated examples include multi step activities lasting up to around 10 minutes, such as potting a plant, brewing pour over coffee, making pancakes, or assembling kits. According to the company, performance on previously unseen tasks reaches roughly 66 percent success per step in tests, compared with much lower rates for comparable systems, and a single video demonstration is estimated to be roughly equivalent to hundreds of traditional training examples.

The model is trained using a combination of large scale simulation, human video data, and real robot experience. Deployed robots feed data back into the shared model, creating a flywheel in which each new installation improves performance for subsequent ones.

Commercial Progress

Skild does not manufacture robots. It supplies the intelligence layer that runs on existing hardware. By September 2026 the company reported reaching a $100 million annual revenue run rate just 10 months after its first commercial deployments. Its software was running on hundreds of robots across more than 60 customers in areas including manufacturing, logistics, inspection, security, and food preparation.

Notable deployments and partnerships include work with NVIDIA and Foxconn on assembly tasks for advanced computing systems, as well as collaborations with major robot makers such as ABB Robotics and Universal Robots.

Funding and Valuation

Skild has raised more than $2 billion in total funding. A $300 million Series A in 2024 valued the company at $1.5 billion. In January 2026 it closed a $1.4 billion Series C led by SoftBank that pushed its valuation above $14 billion. Other participants have included NVIDIA’s NVentures, Jeff Bezos (through Bezos Expeditions), Sequoia Capital, Lightspeed, Coatue, and various strategic investors.

Why It Matters

Robotics has long been constrained by data scarcity and the high cost of custom programming for every new task or machine. Skild’s approach aims to address both problems by creating a general foundation model that generalizes across hardware and improves through deployment. If successful at scale, this model of a shared “brain” for robots could accelerate the shift from single purpose industrial automation toward more flexible, adaptable physical AI systems.

The company is still early in its commercial journey, and long term reliability across complex real world environments remains an open challenge for the entire field. Yet its rapid revenue growth, high profile partnerships, and technical claims around single video task learning have made it one of the most closely watched players in physical AI.