SKILD AI

Skild AI came out of Carnegie Mellon, one of the best robotics schools on earth. Deepak Pathak and Abhinav Gupta are both professors there who spent years researching how robots learn.

Their frustration was that most robots are hand-coded for a single job. Move them or change the task and they fail.

So in 2023 they left the lab to build what they call a general-purpose brain for robots, trained on huge amounts of data the way language models are trained on text. They named the model Skild Brain.

Skild does not build robots. It builds the software that makes them smart, then licenses that intelligence to the companies that make the hardware.

Think of it as an operating system for machines with arms and legs. A manufacturer builds the body.

Skild's model provides the brain that lets it see, grab, walk, and adapt. The money comes from partnering with robotics and industrial firms rather than selling a physical product itself.

The core product is Skild Brain, a foundation model built to control many kinds of robots. The company has shown it running quadruped robot dogs that can walk over rough ground, robotic arms that grasp unfamiliar objects, and humanoid platforms.

The pitch is that the same underlying model handles all of them. Instead of a new program per robot, one brain adapts to whatever body it is placed in.

Skild's strategy is to be the one model that works everywhere. The founders argue that a brain trained on many robots and many tasks will beat brains built for a single machine.

To do that they need scale, which means a lot of money and a lot of data. So they raised a giant Series A fast and used their academic reputations to attract top researchers.

The bet is that generality wins, the same way one large AI model beat thousands of narrow ones in language.

Robots are much harder than text. A language model that makes a mistake writes a bad sentence.

A robot that makes a mistake drops a box or hurts someone. Skild has to prove its brain is reliable in the messy physical world, not just in demos.

It is also early, with no shipping product and heavyweight competition from Nvidia, Google DeepMind, and Physical Intelligence, all chasing the same general robot brain. Raising money is the easy part.

Making the model actually work is not.

Skild's $300 million Series A in 2024 was led by Lightspeed Venture Partners and Coatue, with SoftBank, Menlo Ventures, Sequoia, and Jeff Bezos also in. The round valued the year-old company at $1.5 billion, which is a lot for a startup with no product yet.

Investors were betting on the founders' Carnegie Mellon pedigree and the idea that robotics is where AI goes next. Reports in 2025 suggested SoftBank was pushing to pour in even more.