After a $1.4 Billion Valuation, Skild AI Doesn't Want to Just Be the Brain

“

Next entrance is the body?

On September 15, a paper appeared on arXiv

"BRIDGE:An Open-Source Humanoid Platform via Morphology-Control Co-Design for Physical AI", which means in Chinese

"An Open-Source Humanoid Platform via Morphology-Control Co-Design for Physical AI"

”，

Skild AI's two founders participated in the creation, and the paper highlights an open-source humanoid platform that is 88 centimeters tall and weighs 13 kilograms.

A company that sells "brains" in embodied intelligence released an open-source "body", and the underlying meaning is intriguing.

01.

What is Skild betting on?

Looking at Skild AI's capital curve, the rhythm is very steep.

Founded in 2023, the B round valuation was $4.7 billion in November 2024. The B round valuation was $4.7 billion in June 2025. The C round valuation was $14 billion in January 2026, led by SoftBank and NVIDIA, with a $14 billion funding round.

From $4.7 billion to $14 billion, less than a year, tripled. Cumulative funding exceeds $2 billion.

Why would SoftBank and NVIDIA be willing to bet on this? SoftBank's "brain" + "body" strategy is essentially pouring AI capabilities into the physical world. Skild's Skild Brain is the closest product to a general-purpose brain in this strategy.

What Skild AI originally did was

an omni-bodied general-purpose robot brain, a model-driven four-legged, humanoid, desktop robotic arm, mobile robotic arm, and almost all robot bodies, without the need for reprogramming for each task.

In March 2026, Sk,ild AI signed collaborations with ABB Robotics, Universal Robots, and NVIDIA, deploying Skild Brain into industrial robotic arm production lines. The same year also saw deployment on Foxconn production lines.

The common point of these deployments is that Skild does not sell a complete robot, but instead provides a brain for others' robots.

The brain is installed in ABB's robotic arms, Universal Robots' collaborative arms, and Foxconn's production lines, earning ARR based on deployment.

The body manufacturers provide the body, and Skild provides the brain, charging for the brain. This is Skild's business model: selling the brain, not the body.

Then why does it still need to make its own body?

After reading the paper, we found that

this 88-centimeter open-source robot is not an official product, but Skild AI's attempt to set a reference interface between the brain and the body for the entire industry.

Skild AI believes that

the interface between the brain and the body may be more valuable than the brain itself.

02.

Why would a brain company touch the body?

Because companies that make brains are essentially very dependent on physical carriers.

Skild Brain needs to be deployed across different robots, which requires the body to have a unified interface.

However, today, hundreds of robot companies, each robot manufacturer has different motors, different joints, and different control frequencies. Every time the brain connects to a new body, it needs to go through a round of adaptation.

ABB's robotic arm is compatible with one set, Universal Robots' collaborative arm is compatible with one set, and Foxconn's production line is compatible with one set. The adaptation cost is getting higher, making the universality of the brain harder to achieve.

Skild makes BRIDGE, not for its own use, but to provide the industry with an open-source reference body. This allows the $14 billion brain to have a place to run, ultimately reducing the adaptation cost from one set per manufacturer to one reference set.

The size of BRIDGE is closer to desktop-level rather than industrial-level.

BRIDGE's hardware design, control strategies, and training scripts are all open-source. Its height is 88 cm, and its weight is 13 kg, half the size of the Yu Shi G1 (1.5 m, 35 kg), and much smaller than the YouBian X (1.7 m). It's designed to be a robot that can be placed on a lab desk, with the core purpose of serving developers and helping algorithm teams quickly run through processes.

In short, it's equivalent to Skild AI setting a minimum viable body for the entire Physical AI industry. You don't need to build a 1.7-meter humanoid robot,

You just need an 88 cm, open-source, and compatible with Skild Brain reference platform, which can lower the deployment threshold of the brain.

Skild AI hopes that any team that gets this open-source robot can replicate the deployment process of Sk,ild Brain in their own laboratory.

To achieve this goal,

the core of the paper actually discusses the architecture design of morphology-control synergy.

Because in traditional approaches, the hardware team first draws out the body, and the algorithm team then applies control strategies on top of it, with both lines working separately, and finally during integration, the joint angle limitations of the body restrict the action space of the algorithm, while the algorithm's response delay forces the hardware to re-adjust parameters.

BRIDGE combines both lines into a single design cycle.

The joint configuration of the body follows the algorithm's action requirements, and the algorithm's training data follows the body's actual response.

Iterating until convergence, the body becomes the algorithm's reference implementation, and the algorithm becomes the body's soul specification. This means that the next robot manufacturer's competitive advantage is not more flexible joints, but the ability of the brain and body to co-evolve within the same cycle.

Interestingly, we found that Skild AI often uses Yu Shi's H1 as the carrier for the brain. This year, Yu Shi released the UnifoLM-WMA-0, a cross-body world model, which is actually very similar to what Skild AI has done. The UnifoLM-WMA-0, this cross-robotic-body open-source world model, has a very broad adaptation range, but

Yu Shi's approach is more like first making the body into a factual universal open standard, and then doing the brain well.

Similarly, in March 2025, the Zhiyun Institute released RoboOS at the Zhongguancun Forum, which is a cross-body size-brain collaboration framework, with an open-source embodied brain RoboBrain, capable of achieving lightweight deployment across scenarios and multi-tasks.

Zh iyun RoboOS solves the same problem, cross-body collaboration, but its main approach is to open-source a framework. However, these paths are essentially converging in the end.

03.

What does a reference body mean?

An analogy to the chip industry. Intel doesn't make money by selling development boards. The significance of development boards is that the entire industry's software is debugged, tested, and optimized on the Intel platform.

When the software ecosystem runs entirely on the Intel platform, switching to another platform would require rewriting the drivers.

Skild BRIDGE bets on the same logic. When all industry's robot brains have been adapted on BRIDGE, switching to another reference body would require redoing the adaptation. The adaptation cost is reduced, but the premise is that you use BRIDGE as the reference body.

Previously, body manufacturers sold a robot for a one-time fee. In the future, they will sell a robot for a one-time fee and also charge the brain company for adaptation fees. The moat of body manufacturers is being re-priced by brain companies' interfaces.

Companies that sell brains make bodies, not entering the body track, but reducing the cost of deploying brains. With the adaptation cost reduced, the ARR of all brain companies will accelerate. The one who defines the interface first will collect the money from the followers later.

But here is a hidden premise: the premise of this path is that the simulation ecosystem needs to be established first.

The simulation ecosystem in China is still fragmented, with Isaac Sim, MuJoCo, and Gazebo each occupying a corner, without a factual standard. Skild, due to its investors, has clearly chosen Isaac Sim.

Skild's S1 model runs entirely in NVIDIA Isaac Sim. On September 22, 2026, it released a demo, self-play training to kick a ball, and then transfer the skill to the real world, which is indeed very smooth. However, this system undoubtedly reinforces NVIDIA's fact standard in the simulation layer once again.

Because the significance of the reference platform is not in how many units it sells, but in how many people use it as a benchmark.

In the future, the boundary between the brain and body will eventually be broken, so

whoever defines this interface first will define the standard for the next generation of the robot industry.

END

Industrial robot companies

E斯顿 Automation

E夫特 Robotics

Fao Robotics

Yuexiang Robotics

Jieka Robotics

Songling Robotics

Luoshi Robotics

Atongmu Robotics

Jizhi Jia

Hikvision Robotics

Yifei Technology

Aritel Robotics

Service and Specialized Robotics Companies

Yijiahe

Jingpin Tieshang

Qiteng Robotics

Shihe Robotics

Pudu Robotics

Shirode Robotics

Kuma Technology MAMMOTION

Humanoid Robotics Companies

UBTech Robotics

Universal Robots

Xingde Depth

Xingdong Jiayuan

Weijing Robotics

Zhujidong Power

Leju Robotics

Daxiang Robotics

Magic Atom

Zhongqing Robotics

Paxini Perception

Cyborg Robotics

Digital China

Fuliye Intelligent

Tianlian Robotics

Kepler Humanoid Robotics

Lingbao CASBOT

Qingbao Robotics

Zhejiang Humanoid Robotics Innovation Center

Dongyi Technology

Zhishi Technology

PNDbotics

Zhuoyide Robotics

Qinglang Intelligent

Galileo GALILEO

Songyan Power

Tianji Smart

Wo'an Robot

LiGong Huahui

Accelerated Evolution

Embodied Intelligence Enterprise

Cross-Dimensional Intelligence

Galaxy Universal

Qianxun Smart

Lingxin Qiaohou

Ruierman Smart

Weiyi Manufacturing

Promote Technology

Zhongke Silicon Era

Shu Tutu Technology

Lingqiao Smart

Xingchen Smart

Qiongche Smart

Fangzhou Infinite

| iFLYTEK |

Beijing Humanoid Robot Innovation Center

National and Local Co-built Humanoid Robot Innovation Center

Daimeng Robot

Shibit Robot

Xinghai Tu

Yuequan Biomimetic

Zeroth Power Robot

Zhongke ShenGu

Zhi Fang Square

Daka Robot

Haocun Technology

Ju Shi Smart

Xynova Xi Nuo Future

Feixi Technology

Future Power

Boden Smart

Qianjue Technology

Lingsheng Technology

Ji Cui Manufacturing

Xin Baite Technology

Chen Huen Line Technology

Dexmal Original Power

Youliqi

Self-variable

Ruiyan Zhikong Lingqiao Hand

Qiwu Technology

RoboScience Machine Science

Zhongke Fifth Era

Critical Point

Danghong Technology

Qiaojie Shu Wu

Vbot Weita Power

Tashan Technology

Ju Brain Pan Shi

Youai Zhizhe Robot

Zhi Xing Jian

Aimiao Robot

Lu Ming Robot

Quantum Smart

Ruoyu Technology

Tiehu Robot

Zhi Wuji

Medical Robot Enterprise

Yuanhua Smart

Tianzhi Hang

Sizhe Rui Intelligent Medical

Jingfeng Medical

Tuo Dao Medical

Zhenyida

Shuru® Robot

Luosen Bot

Shui Mu Dongfang

｜

Kangnuo Sistem

Di Shi Medical

Upstream Industry Chain Enterprise

Lvde Harmonic Wave

Yinshi Robot

Kunwei Technology

Maita Smart

Qingtong Vision

Benmo Technology

Lan Dian Touch Control

| Xin Jingcheng Sensor

BrainCo Qiang Brain Technology

Yuli Instrument

Ji Ya Jingji

Sikan Technology

Shen Yuan Sheng

Fei Pu Navigation Technology

Inks

Juxia Intelligent Drive

Lingyun Guang Yuanke Vision

Xuanji Power

Yiyou Technology

| Ruiyuan Precision |

Lingzhu Era

HIT Huawei Ke

Xinghui Sensor

Lingdi Technology

Quan Zhi Bo

CubeMars Robot Power

Wanglong Robot Elevator

Hangkai Micro Electronics

Wutong Sensing Control