Understand any stock in simple English
Enter a ticker to learn what the company does and how it makes money.
Skild AI, Inc.
Skild AI builds general-purpose foundation software — a "brain" — that lets many different kinds of robots learn and perform tasks.
Skild AI is a robotics software company. It does not build robots. Instead, it is trying to build the operating brain that goes inside them — a single, general-purpose artificial intelligence model that can control many different kinds of machines, from a warehouse arm to a legged robot, rather than one bespoke program per robot per task.
The problem it attacks is that robots have traditionally been narrow and expensive to teach. Each new robot and each new job usually meant custom code and lots of hand-guided training. Skild's bet is that a large model trained on huge amounts of robot data and video can learn general skills — grasping, moving, navigating — that transfer across bodies and tasks, so a customer can get a new robot doing useful work far faster and cheaper. Customers would be robot makers and the companies deploying robots, who pay for the software that makes their hardware actually capable. The company was founded in 2023 by Deepak Pathak and Abhinav Gupta, robotics researchers from Carnegie Mellon University, and it has drawn heavy investor interest for a young startup — a funding round reported in 2024 valued it at roughly $1.5 billion, with backers including Lightspeed and Nvidia's venture arm. In 2025 it raised again at a reported valuation near $4.5 billion, with SoftBank among the participants. The wild enthusiasm reflects a big promise and a big risk: general robot brains are still unproven at scale, and turning research demos into reliable, revenue-generating software is the hard part.
Skild makes money by licensing its software rather than selling hardware, so its fortunes ride on whether robot adoption spreads and whether customers trust one model to run many machines. It competes in a crowded race — against robot makers building their own brains in-house and against other AI labs chasing the same general-purpose prize. The whole bet is that the robot brain becomes a standard layer, like an operating system, and that Skild gets to own it.
Saved answer · last researched 10 Nov 2026
BIG PICTURE
Skild's robot brain went from 0% to 66% on tasks it had never seen, just by watching more video
- 66% Success on unseen tasks after 100k pre-training hours
- 34% Tasks it still failed
Skild tested its S1 model on jobs it had never been trained for, giving it a single video prompt and no extra practice. With about 100,000 hours of pre-training, it succeeded 66% of the time, up from roughly 0% at 1,000 hours. That jump is the whole pitch: more data makes the brain better at new work, so Skild can sell software that adapts instead of paying people to reteach each robot. Source: State of AI Report 2026, as cited by Fabian Warislohner, data from 2026
Timeline
Two Professors Start Skild
Deepak Pathak and Abhinav Gupta, two professors at Carnegie Mellon in Pittsburgh, had worked together at Meta's AI lab and kept running into the same wall. Robot videos looked amazing, but the machines could not reliably do the same job twice, and every new robot or task meant starting over. Their idea was to build a general-purpose robot brain that runs on ordinary chips, not special custom ones, and works across different robots, tasks and places.
First Test on a Real Robot
The team bought an off-the-shelf humanoid robot and loaded their software onto it. It made real progress in a single day — a promising early sign that the brain could actually drive hardware, not just simulations.
A Giant First Round
A few months later, Skild raised its first outside money: a $300 million round that valued the young company at $1.5 billion. Big investors like Lightspeed, Coatue, SoftBank and Jeff Bezos led the way, with Carnegie Mellon and Amazon funds joining in. For a company with no product on the market yet, that was a loud vote of confidence.
Skild Brain Goes Public
Another massive round followed, led by SoftBank with Nvidia and Samsung chipping in, valuing Skild at roughly $4.5 billion. At the same time, the company finally showed the world what it had been building: Skild Brain, a single foundation model trained across a huge range of simulated robot bodies. Instead of one brain per robot, here was one brain for many.
Revenue Out of Nowhere
Then something unusual for a research-heavy startup happened: the money started coming in. Skild went from zero revenue to about $30 million in a few months, selling its robot brains into security, inspection, delivery, warehouses, factories, data centers and construction sites. It was a first year of real commercial activity, and it was a fast one.
The $14 Billion Valuation
In January 2026, Skild raised a $1.4 billion round led by SoftBank, valuing it above $14 billion — roughly nine times its value in a year and a half. Total funding passed $1.8 billion, making it the largest robotics AI round ever reported. Nvidia, Bezos, Samsung, LG, Schneider Electric and Salesforce Ventures all joined, and the message was clear: the market believed the robot brain was real.
Skild Buys Zebra's Robotics Unit
Skild then bought Zebra Technologies' robotics automation division, which included warehouse software called Symmetry Fulfillment. That gave Skild the tools to manage large fleets of robots and run entire warehouses, not just power individual machines. It also handed the company a ready channel into big-scale warehouse automation, where it could plug its brain into real operations.
From Demos to Scale
Ten months after its first commercial deployment, Skild crossed $100 million in annual recurring revenue, with more than 60 paying customers and hundreds of robots in the field. Its new S1 model learned a task from a single video demonstration without retraining, and customers like Nvidia and Foxconn were using it for real assembly work. The company that started with flashy robot videos was now running actual operations at scale.