FANUC to ship AI Welding Agent that reads drawings and writes robot weld programs, built on Google's Gemini Enterprise

FANUC Corp. said on Sept. 11, 2026 it will start shipping an AI Welding Agent at the end of December, a system that reads the engineering drawing of a part and generates the program a robot needs to arc weld it. The agent runs on Gemini Enterprise, Google Cloud's enterprise generative AI platform, and is FANUC's first physical AI product built with Google to carry a shipping date.

FANUC describes the product as “zero setup, zero teaching.” The agent derives welding conditions, namely current and voltage, together with the robot's motion program, from the drawing alone. FANUC did not disclose pricing, performance figures or launch customers. The company demonstrated the system at the International Welding Show at Tokyo Big Sight, which opened on Sept. 16.

How it works

According to the release, the operator captures the drawing with the camera built into the tablet teach pendant of FANUC's CRX collaborative robot, so no dedicated camera or extra device is needed. The AI interprets the part, generates welding parameters and the robot path, and the operator can run them as generated or adjust them before welding. FANUC said the agent works with any welding power source connected to its robots, which means users do not have to switch power-source brands to adopt it.

FANUC addressed data concerns directly. Customer drawings and production data are protected by Gemini Enterprise's enterprise security and are not used to train AI models for other users, the company said. Customers do not need a separate contract with Google Cloud because the service is provided under their FANUC agreement.

The target is a labour bottleneck. FANUC said skilled arc welders who can both set welding conditions and teach robots are scarce worldwide and that the shortage is delaying supply of materials and components. It named automotive and truck parts, construction materials and shipbuilding as the main target industries.

The FANUC and Google track

The launch builds on a collaboration FANUC announced with Google on May 13, which covered Gemini Enterprise and Intrinsic, Google's software platform for building and deploying AI robot applications, including its Flowstate development environment. In that release FANUC said its lineup spans collaborative robots with a 3 kg payload up to industrial robots carrying 2.3 tonnes, that its robots support ROS through open-source drivers, that it takes part in Google DeepMind's Gemini Robotics Trusted Tester Program, and that it had shipped more than 1,000 robots for physical AI applications since showing its physical AI system at the International Robot Exhibition in December 2025.

The business behind it is growing. FANUC reported consolidated net sales of ¥231,035 million for the April to June 2026 quarter, up 17.7% year on year, with robot division sales up 18.7% to ¥96,103 million, according to its July 31 filing. It raised its full-year forecast to net sales of ¥948,100 million and operating income of ¥218,000 million. On the same day as the welding agent, FANUC also announced an ultra-light portable collaborative robot aimed at automating construction sites and a high-speed, low-distortion pipe welding system, and on Sept. 30 it signed a physical AI partnership with Hitachi.

Competition

Yaskawa Electric, FANUC's closest domestic rival in arc welding robots, also exhibited at the 2026 International Welding Show, according to its event notice. On Aug. 31 it launched the MOTOMAN-HC12 collaborative robot with a 12 kg payload, a 1,410 mm reach and a newly developed torque sensor, listing arc welding among its applications, and in July it announced an agentic robot system with Google DeepMind. Kawasaki Heavy Industries' precision machinery and robot segment grew faster than both in the latest quarter, driven by semiconductor robots rather than welding. The table compares the three robot businesses.

The technical idea is simple to state. Welding drawings encode joints, weld lengths and sizes in standard symbols that a trained person reads before choosing parameters and teaching the torch path point by point. A multimodal model that can interpret those symbols lets software do the reading, and FANUC's robot software turns the result into parameters and motion. FANUC has not published accuracy data, so how often operators must correct the output remains unknown.

For the global market the significance is that the teaching step, rather than the robot, has long been the cost that keeps small-batch welding manual. A collaborative robot that a shop can set up from a drawing changes the economics for job shops and suppliers that weld many part numbers in small lots, which is where welder shortages bite hardest.

The open questions are commercial and technical. FANUC has not said what the agent costs, whether it is a one-off licence or a subscription, or whether it will run on robots other than the CRX. A cloud-based model adds a dependency on network access in plants that often restrict it. Responsibility for weld quality when an AI sets the parameters is another issue customers in safety-critical sectors such as shipbuilding will raise.

The dates to watch are the end of December 2026, when shipments are due to start, FANUC's results for the July to September quarter, and the first customer references from automotive parts makers and shipyards.