With $1.4B Raised, Skild AI Builds a General Embodied Brain Across Tasks and Hardware

AI typically evolves from specialized systems to general-purpose models, and then into high-value vertical applications.

In software AI, general multimodal models have already replaced task-specific systems like facial recognition and speech transcription, and can be fine-tuned for domains such as healthcare, law, and customer service.

Physical AI, however, has lagged behind. Traditional robots operate only in narrow environments and fail to generalize, making truly practical humanoid robots still a distant goal.

Skild AI aims to change this with Skild Brain, an “omni-bodied” robotic foundation model designed to control different robot forms and perform diverse tasks using the same brain.

In 2024, Skild AI raised US$300 million in Series A and US$1.4 billion in Series B, led by SoftBank, with participation from NVIDIA and Jeff Bezos, valuing the company at over US$14 billion.

Skild Brain uses vision-based, end-to-end motion control and a hierarchical architecture that enables robots to adapt in real time without predefined maps or rules. Key innovations include training across tens of thousands of robot morphologies and applying in-context learning to physical AI, allowing robots to recover from failures or extreme hardware changes.

Already deployed in security, logistics, data centers, and manufacturing, Skild targets consumer home robots as its long-term goal. As embodied AI crosses the generalization threshold, its real-world value could grow exponentially.