What is Skild AI? The $14bn robot-brain maker's funding, S1 model and $100m revenue run rate explained

Skild AI makes one product, a robot brain, and in less than three years it has become the most richly funded company of its kind. The Pittsburgh start-up, founded in 2023 by roboticists Deepak Pathak and Abhinav Gupta, sells the Skild Brain, a foundation model meant to drive quadrupeds, humanoids, robot arms and mobile manipulators alike. It raised $1.4 billion in January at a valuation of more than $14 billion and said on September 10 that it had passed $100 million in annual recurring revenue.

Those two numbers sit uneasily together. By MEKAWIRE's arithmetic, the January valuation is more than 140 times the revenue run rate Skild reported eight months later. This explainer sets out what Skild actually sells, where its money and revenue come from, how it compares with Physical Intelligence and other model makers, and what would have to be true for the price to make sense.

A model that is supposed to run any robot

Skild described the Skild Brain in July 2025 as a two-layer system: a low-frequency policy that decides what to do and a high-frequency policy that turns that into joint commands. It is pre-trained on large-scale simulation and internet video of people handling objects, then post-trained on teleoperation and on data from paying deployments. The company calls this omni-bodied: the same model, adapted by fine-tuning, controls bodies it was not designed around.

The second product step came on August 18 with S1, which Skild calls an in-context learner. An operator records one video of a new task and gives it to the robot as a prompt; the model's weights are not retrained. On long tasks of four to eight minutes that the model had not seen in training, Skild reports a 66% per-step success rate against 9% for a conventional language-prompted vision-language-action model, both at 100,000 hours of pre-training. It also says one in-context demonstration is worth about 380 fine-tuning examples, which took 50 to 100 hours of teleoperation to collect. On tasks the models had seen, the language-prompted policy did better at small scale, 53% against 43% at 1,000 hours. These are the company's internal benchmarks, not an independent test.

That bridge from research to sales is where Skild's story differs from most robot-model companies.

Where the revenue comes from

In its September post the founders said Skild had more than 60 paying customers ten months after its first commercial deployment, across moving goods, deliveries, site inspection, security, food preparation, warehouses, factories and data centres. About 90% of revenue comes from manipulation and 10% from mobility, of which 4% is autonomous mobile robots inherited from Zebra Technologies. It had recognised $50 million since deployments began. Named projects include dual-arm assembly of NVIDIA Blackwell systems with Foxconn, wire-harness work at Sumitomo Wiring Systems and kitchen pilots with Mitsui & Co., whose supply chains serve 1.4 million meals a day in Japan.

The numbers imply a business that is still accelerating. Recognised revenue averaged $5.0 million a month over ten months, while a $100 million run rate equals $8.3 million a month. Skild had said in January that 2025 revenue went from zero to about $30 million within a few months.

The Zebra deal shows how Skild buys distribution cheaply. Zebra's quarterly filing shows the sale of its robotics unit, the former Fetch Robotics, closed on March 27 for $20 million, made up of $9 million in cash, about $2 million in escrow and $9 million of Skild equity. Zebra had paid $301 million for Fetch in 2021, so the price was 6.6% of that. Skild also lists a March partnership with ABB Robotics and Universal Robots to put the Skild Brain on their arms.

How the money compares

In MEKAWIRE's verified records, Skild's $1.4 billion Series C is 48.6% of all officially disclosed money raised by robot foundation-model and embodied-AI developers in the twelve months to October 18, across six rounds totalling $2.88 billion. Its valuation rose 9.3 times in 554 days, from $1.5 billion at the Series A of July 2024. Its closest rival by profile, Physical Intelligence, has $1.0 billion of funding in our records and a $5.6 billion valuation that rests on outlet reports rather than a company announcement. Figure AI, a humanoid maker that builds its own Helix model, carries a disclosed $39 billion valuation but sells hardware and software together.