Pittsburgh-based Skild AI disclosed on September 10, 2026 that its robot brain business is now generating revenue at a $100 million annualized run rate — a milestone hit just ten months after its first commercial deployment — as it debuted the S1 foundation model that lets any robot learn a new task from a single video demonstration.
The commercial ramp
Skild's software is running on hundreds of robots across more than 60 customers, up from eight earlier this year. Together with NVIDIA and Foxconn, Skild is deploying its Skild Brain on dual-arm manipulators used to assemble NVIDIA Blackwell systems — an unusually reflexive use of physical AI, with the model helping build the very silicon it runs on.
Inside the S1 model
The company's new S1 foundation model was built end-to-end on NVIDIA infrastructure, spanning synthetic data generation with Cosmos, large-scale training on Blackwell GPUs, simulation in Isaac, and real-world deployment on Jetson Thor. Skild says S1 hits 96% success on tasks it has seen and 66% on unseen tasks, and that a single demonstration effectively replaces roughly 380 post-training examples — a rare glimpse into concrete data efficiency numbers for a physical-AI foundation model.
The SoftBank-backed thesis
SoftBank led Skild's $1.4 billion Series C in January 2026 at a valuation above $14 billion, betting the company can build a single "omni-bodied brain" that generalizes across humanoids, quadrupeds, mobile manipulators and industrial arms. That thesis is increasingly the shape of the market: Kinetix AI, Physical Intelligence and Field AI are all raising nine-figure rounds to build robot brains rather than complete robots.
What comes next
Skild plans to expand S1 to more form factors, more industrial partners and a public API that lets integrators plug the model into their own robot fleets. With XDOF pushing teleoperation datasets and NVIDIA cementing itself as the training and inference backbone, the physical-AI stack is starting to look a lot like the cloud stack that preceded it.
Reporting based on coverage from NVIDIA's blog, Unite.AI and The Next Web.