Skild AI

A brain for many bodies

Skild AI is a robotics company that says it was founded in 2023 with the goal of building intelligence grounded in the physical world^[1]. Its central object is not a robot but a model: the company describes its foundation model as a general-purpose brain meant to serve across diverse robot embodiments, scenarios and tasks, including manipulation, locomotion and navigation^[1].

The idea rests on a distinction familiar from software. A foundation model is trained broadly and then adapted to many downstream uses, rather than being built for one narrow job. Applied to robots, that means the same learned capability could in principle be moved between a fixed industrial arm, a mobile base or a legged machine, instead of every body needing its own bespoke software. "Embodiment" here simply means the physical form a model is running in — the shape, sensors and actuators it has to work with.

Origins and team

Skild AI says its team includes people from Meta, Tesla, Nvidia, Amazon and Google, and from Carnegie Mellon, Stanford, UC Berkeley and UIUC^[1]. The company states its long-term goal as developing artificial general intelligence rooted in the physical world^[1] — a far broader ambition than any single robot task, and one that frames the model work as a step toward machines that can handle unfamiliar situations in the physical world rather than only rehearsed ones.

The Series A

On 9 July 2024 Skild AI announced a US$300 million Series A, valuing the company at US$1.5 billion^[1][2]. According to the company, the round was led by Lightspeed Venture Partners, Coatue, SoftBank Group, and Jeff Bezos through Bezos Expeditions^[1]. Other participants it named were Felicis Ventures, Sequoia, Menlo Ventures, General Catalyst, CRV, SV Angel, Carnegie Mellon University, and the Amazon Industrial Innovation Fund and Alexa Fund^[1]. Skild AI said the capital would be used to continue scaling its model and to build its team^[1]. The investor list mixes general venture firms with corporate funds and a university, which is unusual enough to be worth noting on its own: money of that kind usually follows a claim that a technology can be applied widely, not a single product line.

From research to deployment

A post titled Building the general-purpose robotic brain appeared on 29 July 2025^[2]. On 14 January 2026 the company announced a US$1.4 billion Series C led by SoftBank, valuing it at more than US$14 billion^[2]. It said it had crossed US$100 million in annual recurring revenue ten months after its first commercial deployment^[2]. Annual recurring revenue, or ARR, is the yearly value of the subscription-like revenue a business expects to keep receiving; it is a measure of how much recurring business a company has signed up, not of profit.

On 19 March 2026 Skild AI said it had partnered with ABB Robotics and Universal Robots to deploy its omni-bodied Skild Brain across industries^[2]. ABB Robotics and Universal Robots are established makers of industrial and collaborative robot arms, so the arrangement put the company's software next to hardware that is already sold and installed in factories, rather than only in its own machines.

S1

On 18 August 2026 the company introduced S1, describing it as its flagship robotic foundation model, built from the ground up as an in-context learner^[2]. In-context learning is the ability to pick up a new task from examples or instructions supplied at the moment of use, without retraining the model's weights. For a robot, that is the difference between a system that must be reprogrammed for each new object or motion and one that can be shown what to do and adjust within the same session.

Where it sits

Read as a sequence, Skild AI's public record runs from a founding claim about physical-world intelligence, through a large early round and a stated ambition toward general intelligence, to a named model, reported commercial revenue and partnerships with industrial robot makers^[1][2]. The recurring theme is generality: one model, many bodies, many tasks. Whether that generality holds outside the settings the company reports is a question its own announcements do not settle.