Skild AI's S1 model learns new robot tasks from one video instead of hours of teleoperation. The company says it is deploying with Foxconn, working toward deployment at Sumitomo Wiring Systems, and piloting kitchen robots with Mitsui. Worth watching for physical AI and industrial automation, but not a product individuals can buy.
Skild AI, founded 2023 in Pittsburgh by ex-Carnegie Mellon professors Deepak Pathak and Abhinav Gupta, builds S1, an in-context-learning robotics foundation model that executes unseen tasks up to ten minutes long from a single video demonstration. S1 succeeded the original Skild Brain in August 2026 and the company reports paying customers including Foxconn (deploying) and Mitsui (piloting).
Founded: 2023 · HQ: Pittsburgh, PA, USA · Team: 100-250 · CEO: Deepak Pathak · Funding: $2.2B total raised across seed (2023, $14.5M), Series A (July 2024, $300M, led by Lightspeed and Coatue), Series B (summer 2025, ~$500M, led by SoftBank), and Series C (January 2026, $1.4B, led by SoftBank) · Valuation: $14B+ (Series C, January 2026)
About Skild AI
Skild AI was founded in early 2023 by Deepak Pathak and Abhinav Gupta, two computer science professors who left Carnegie Mellon University to pursue a shared thesis: that the robotics industry did not need hundreds of task-specific AI models but a single general-purpose brain capable of driving any robot through any physical challenge. Pathak, who serves as CEO, spent years at CMU studying curiosity-driven learning and self-supervised robot control. Gupta, who serves as President, contributed expertise in large-scale robot learning and simulation-to-real transfer. The pair earned the Best Robotic System Award at the Conference on Robot Learning in 2022 for work on adaptive sim-to-real training. The company is incorporated as a C-corporation with its principal office in Pittsburgh, Pennsylvania, and additional engineering hubs in Pasadena, California and, since February 2026, Bengaluru, India.
The company's flagship product today is S1, a robotics foundation model introduced in August 2026 as the current generation of the earlier Skild Brain platform. Where the original Skild Brain was trained to be omni-bodied, controlling any robot type without prior knowledge of its exact mechanical structure, S1 adds in-context learning: shown a single video demonstration of a task, seen or unseen, short or long, it executes without any fine-tuning of its weights. Skild AI reports S1 completing previously unseen long-horizon tasks, such as potting a plant or brewing pour-over coffee, up to ten minutes long from one demonstration, and cites an internal benchmark in which its in-context approach reached a 66% success rate on unseen tasks versus 9% for a comparably-trained language-prompted policy. S1 is built on NVIDIA AI infrastructure and pre-trains on human demonstration video, teleoperation, and procedurally generated physics simulation, the same three data sources used for the original Skild Brain.
Skild AI unveiled the original Skild Brain publicly in July 2025, releasing footage of humanoid robots, quadruped dogs, and tabletop manipulation arms all running under a single model. In April 2026, the company acquired Zebra Technologies' Robotics Automation Business, the former Fetch Robotics unit, including its Symmetry Fulfillment warehouse orchestration platform, making Skild the operator of an end-to-end physical automation stack: humanoids for pick-and-place tasks, quadrupeds for facility inspection, robotic arms for packing, and autonomous mobile robots for material transport, coordinated by a single software layer. In March 2026 Skild also announced deployment partnerships with ABB Robotics and Universal Robots to bring its models to existing industrial arm hardware, alongside its ongoing work with NVIDIA and Foxconn on factory-floor deployments.
Skild AI's business model is B2B licensing: robot manufacturers, logistics operators, and industrial enterprises pay to run Skild's models on their physical systems; the company does not sell robots directly. On September 10, 2026, Pathak and Gupta announced the company had crossed $100 million in annualized recurring revenue, with $50 million already recognized, ten months after its first commercial deployment and across more than 60 paying customers. Manipulation tasks such as food preparation, security inspection, and warehouse and factory work account for roughly 90% of that revenue; mobility accounts for about 10%, of which autonomous mobile robots from the Zebra/Fetch acquisition alone are 4%. The company's own September post names three flagship efforts, each at a different stage: dual-arm manipulators assembling NVIDIA Blackwell systems with NVIDIA and Foxconn (described as being deployed), wire-harness manufacturing at Sumitomo Wiring Systems (described as work toward deploying S1), and general-purpose kitchen robots piloted with Mitsui & Co., whose supply chains serve 1.4 million meals a day across Japan.
Total funding across four rounds comes to approximately $2.2 billion. The company started with a $14.5 million seed round in 2023. In July 2024, it announced a $300 million Series A led by Lightspeed Venture Partners and Coatue, with participation from SoftBank, Bezos Expeditions, Sequoia Capital, and General Catalyst, at a post-money valuation of $1.5 billion. By summer 2025, Skild closed an approximately $500 million Series B led by SoftBank, with NVIDIA (via NVentures), Samsung, and LG also investing, at a valuation of roughly $4.5 billion. In January 2026, SoftBank led a third consecutive round, a $1.4 billion Series C, with new participation from Macquarie Capital, Disruptive, and 1789 Capital, alongside returning investors Felicis, Coatue, Lightspeed, and Sequoia, bringing the valuation to over $14 billion, the largest robotics AI funding round on record at the time. No later round has been disclosed as of October 2026.
Deepak Pathak serves as CEO, responsible for product direction and investor relations. Abhinav Gupta holds the title of President, overseeing research and technical operations. Both founders remain active researchers and are based primarily in Pittsburgh. Third-party workforce trackers put headcount at roughly 100-250 as of mid-2026 (Revelio Labs reported approximately 125 employees in March 2026, up from 47 a year earlier), still lean relative to the $2.2 billion raised, reflecting a strategy centered on high-density research and engineering hires. The company maintains offices in Pittsburgh (primary, robotics and AI research), Pasadena, California (West Coast engineering), and, since February 2026, Bengaluru, India, its first expansion outside the US.
Skild AI frames its goal as AGI for the physical world: an artificial brain general enough to operate any robot body on any task, rather than the narrow, task-specific systems that dominate robotics AI today. The company's research lineage runs from LocoFormer, an in-context locomotion model released in September 2025, through S1's in-context manipulation results in August 2026, to Physical Self-Play, a post-training technique announced September 23, 2026, in which the company reports S1 learning extremely dexterous and dynamic tasks, including soccer, through self-play rather than human demonstration alone. The core research bet across all three is that scaling cross-embodiment data, including human video, teleoperation, and physics simulation, produces a foundation model that generalizes to new robot bodies and tasks without task-specific fine-tuning, distinct from the prior dominant approach in robotics AI that required a separate dataset and model per robot body, task, and environment.
In the robotics AI foundation model space, Physical Intelligence is the nearest rival, having raised over $900 million and focuses on precision manipulation tasks for tabletop and upper-body humanoid robots. Skild's own S1 research post names Generalist AI and a 2026 RoboTTT paper as concurrent approaches to in-context manipulation, though Skild claims the only published results on unseen tasks longer than ten minutes. Figure AI, backed by OpenAI and Microsoft, targets humanoid robots specifically in automotive assembly with deep OEM integration at BMW and other manufacturers, a narrower but deeper approach. Where Skild AI still faces pressure: employee-count estimates of roughly 100-250 remain small relative to the competitive intensity in this space and to a $14 billion valuation, third-party headcount figures are unaudited and vary by source, and the company names only three customers (Foxconn, Sumitomo, Mitsui) out of more than 60, so how concentrated its revenue is cannot be checked from outside.
Skild AI is a US-incorporated company based in Pittsburgh. Its investors include foreign firms such as SoftBank (Japan), Samsung (Korea), LG (Korea), and Macquarie (Australia), and this review found no public statement from the company on government or export-control reviews. The company's February 2026 expansion into Bengaluru, India adds a second jurisdiction for data and export-control considerations alongside its Pittsburgh and Pasadena operations. Enterprise customers based in the EU remain subject to GDPR for any personal data Skild processes on their behalf. No EU AI Act classification has been publicly disclosed, and no FedRAMP authorization has been announced as of October 2026.
The company's near-term priorities, per its own September 2026 disclosures, are scaling S1's in-context learning and physical-self-play research, deepening the NVIDIA/Foxconn, Sumitomo, and Mitsui deployments, and converting its 60-customer base into a broader enterprise pipeline across warehouse, manufacturing, food service, and construction sectors. Having crossed $100 million in annualized recurring revenue just ten months after its first commercial deployment, Skild AI has moved from a pre-revenue research narrative to a scaling-commercial one faster than most robotics AI peers, though the company has not disclosed a profitability timeline or IPO date. The next 12 months will test whether the deployment-data flywheel the founders describe, specialist deployments feeding data back into the general S1 model, keeps compounding as fast as the ARR growth suggests. See the directory of AI companies for other robotics and physical AI vendors on Hokai.
Mission
Building a single, general-purpose artificial brain that can control any robot for any task.
Products
- S1 (Robotics foundation model, in-context learning (B2B license)): https://www.skild.ai/blogs/s1
- Symmetry Fulfillment (Warehouse orchestration platform (acquired from Zebra Technologies/Fetch Robotics, April 2026)): https://www.skild.ai/blogs
Links
Frequently Asked Questions
How much funding has Skild AI raised, and what is it worth in 2026?
Skild AI has raised approximately $2.2 billion across four funding rounds: a $14.5 million seed round, a $300 million Series A (July 2024, led by Lightspeed and Coatue), a roughly $500 million Series B (summer 2025, led by SoftBank), and a $1.4 billion Series C (January 2026, led by SoftBank with Macquarie Capital, Disruptive, and 1789 Capital joining). The Series C valued the company at over $14 billion, the largest robotics AI round on record at the time.
What products does Skild AI sell in 2026?
Skild AI's core product is S1, an in-context-learning robotics foundation model introduced in August 2026 that a customer can teach a new task with a single video demonstration rather than a fine-tuning run; it is licensed B2B to robot manufacturers and industrial operators, with no public pricing disclosed. Through its April 2026 acquisition of Zebra Technologies' Robotics Automation Business, Skild also sells Symmetry Fulfillment, a warehouse orchestration platform that coordinates multiple robot types on a single floor. No consumer product or public API exists.
Does Skild AI have SOC 2, GDPR, or other compliance certifications?
No HIPAA attestation, ISO 27001 certificate, or SOC 2 report is listed on Skild AI's site as of October 2026, typical for a robotics vendor still building out its first enterprise compliance reviews. Any personal data Skild handles on behalf of EU customers still falls under GDPR, though the core product, robot motor control, does not inherently collect personal data. Customers evaluating Skild for regulated deployments should request a security questionnaire directly from its enterprise sales team.
Who competes with Skild AI in robotics foundation models?
Physical Intelligence, which has raised over $900 million, is Skild's closest publicly known rival, focused on precision manipulation for tabletop and humanoid robots rather than Skild's cross-embodiment approach. Figure AI, backed by OpenAI and Microsoft, takes a narrower, deeper approach: humanoid robots purpose-built for automotive assembly lines at BMW and other OEMs. Skild's own research also names Generalist AI as a concurrent in-context-learning approach, though Generalist is not yet listed on Hokai.
How does an enterprise start working with Skild AI?
Skild AI does not sell to individual consumers, and its site shows no self-serve signup, price list or developer API as of October 2026, so an enterprise has to approach the company directly. Its September 2026 post describes Foxconn as a deployment, Sumitomo Wiring Systems as work toward deployment, and Mitsui & Co. as a pilot.
Top Alternatives
- Figure AI: Pick Skild for a generalist brain across form factors; pick Figure for humanoid vertical integration.
- Physical Intelligence: Pick Skild for a cross-embodiment brain already in commercial deployment; pick Physical Intelligence for deeper tabletop and upper-body manipulation research.