Skild AI Tops $100 Million in Annualized Revenue 10 Months After Commercial Rollout

Skild AI said Sept. 9, 2026, that its annual recurring revenue (ARR) topped $100 million. The company said it reached the milestone 10 months after its first commercial rollout and now has more than 60 paying customers. Applications include material handling, delivery, inspection, security, cooking, warehousing, factories and data centers. The company said mobility applications account for 10% of revenue and its Fetch solution accounts for 4%.

Skild AI's announcement puts a measure on the shift toward judging robotics by deployment and revenue, rather than demonstrations, with ARR of $100 million and more than 60 paying customers. Its use of NVIDIA AI infrastructure to train S1, followed by deployment in varied settings such as factories, warehouses, data centers and commercial kitchens, highlights the need for both computing infrastructure and adaptation to specific operating environments.

What is new in Skild AI's announcement is its emphasis on revenue and the breadth of deployment, rather than research results. The figures — $100 million in ARR, 10 months and more than 60 customers — suggest deployment and operations have moved to the center of the business, beyond the prototype stage. The company broke out mobility applications at 10% of revenue and its Fetch solution at 4%, indicating that revenue is not concentrated in a single application, though it remains to be seen which uses will become major growth engines. A related article from this newspaper, “Nvidia-linked General Robotics says AI wrote 80% of its code, cutting robot training time by 99%”, focused on using AI to accelerate robot development. Skild AI, by contrast, emphasizes how robots adapt to changing environments after deployment. The company's description of S1 — that it can learn from a single video and use new demonstrations as prompts without updating its weights — suggests a design aimed at responding to frequent process changes in factories and warehouses. From an industry perspective, the announcement suggests the focus is not only on the performance of the robots themselves, but also on whether computing infrastructure, field data and the effort required for redeployment can be addressed together. The use of NVIDIA AI infrastructure for training underscores the close relationship between computing resources and model development. The examples of high-precision assembly with Foxconn, wire-harness manufacturing with Sumitomo Wiring Systems, and commercial kitchens with Mitsui & Co. show how requirements can vary by process even when using the same model. The central question, then, is how far a general-purpose model can be adapted to specific work settings; the next issue will be aligning speed, precision and update methods across deployments. It remains unclear how many of these deployments will move to mass production and permanent operation, and how widely S1's update frequency and training data will expand. The announcement gave the number of paying customers and revenue breakdowns, but not specific figures for deployed units, customer retention or profitability in each application. The next points to watch are how differences among deployments can be standardized and how much variation the system can handle without retraining.

The company said it is deploying Skild Brain on dual-arm manipulators with NVIDIA and Foxconn for high-precision assembly of NVIDIA Blackwell systems. It said the task changes with each product cycle and previously required reprogramming every robot on the line. At Sumitomo Wiring Systems, the company said it aims to automate processes in wire-harness manufacturing that have been considered “impossible to automate.” With Mitsui & Co., it said it is conducting a trial deployment of general-purpose robots equipped with S1 for commercial kitchens serving Japan's supply chain of 1.4 million meals a day. The company said S1 can learn from a single video example, allowing an operator to record a new demonstration and provide it as a prompt, without updating the model's weights.

The announcement includes Sumitomo Wiring Systems and Mitsui & Co., putting the focus in Japan on wire-harness manufacturing and commercial kitchens — areas where processes can be standardized but exceptions still arise. If S1 performs as described, learning from a single video and adapting without weight updates, the conditions for deploying it in processes with frequent changes will be an issue for Japanese manufacturers and food-service companies.