[Edaily Reporter Shin Yeong-bin ] Global AI companies that have been competing in the chatbot and Agent AI spaces are now expanding their frontiers into robotics. While OpenAI is directly building its own robot development and production organization, companies such as Google DeepMind, Anthropic, and DeepSeek are also moving to secure a foothold in the “physical AI” market.

According to the information technology (IT) industry on the 20th, OpenAI is currently recruiting for approximately 26 robotics-related positions. This is more than double the number of openings—around 10—in the first half of the year. In particular, the company’s plan to build the “body” of a robot and move toward actual production has taken shape.

From Motors and Gears to the Supply Chain… OpenAI Takes Direct Control of the Robot’s “Body”

First

,

the trend in the development of actuators—which function as the

robot’s

“joints”—is particularly noticeable. OpenAI is hiring actuator design engineers, actuator gear design engineers, and motor and actuator manufacturing engineers.

Actuators are core components that convert electrical energy into motion to move a robot’s joints, such as its arms and legs. OpenAI aims to develop its own actuators that integrate motors, gears, sensors, and thermal management structures.

The company also plans to establish a supply chain organization. In a job posting for a Robotics Parts Procurement Manager, OpenAI described itself as a “growing hardware company.” Its current strategy is to build a supply chain capable of handling everything from research and development (R&D) through new product introduction (NPI) to future mass production.

The supply chain will include precision-machined parts, gears, electronic components, and printed circuit boards, as well as contract manufacturers. The job description also specifies the task of transitioning business partners from suppliers suitable for prototypes to those capable of mass production.

Anthropic Focuses on Connectivity Standards, Google on ‘Robot Brains’… Robotics Race Heats Up in the AI Industry

Although they are all aiming for physical AI, their paths differ. While OpenAI seeks to build the ‘body’ of robots directly, Anthropic is focusing on standards that connect AI and machines, and Google DeepMind is targeting a ‘brain’ that can be shared across multiple robots.

Last month, Anthropic unveiled the “Model Hardware Standard (MHS),” which connects AI agents to physical equipment. This standard enables AI to identify and operate robots or experimental equipment from different manufacturers in a consistent manner.

Anthropic announced that Doosan Robotics(454910)is testing MHS for automated quality inspection and task coordination among multiple robots. Universal Robots also plans to add MHS support to its robot platform.

Google DeepMind is focusing on developing a “brain” that can be universally applied to various robots. Last July, it unveiled its latest Vision, Language, and Action (VLA) model, “Gemini Robotics 2.” This model converts camera and language inputs into actual robotic movements, enabling control of everything from robotic arms to humanoids.

DeepMind stated that it is conducting joint research with companies such as Aptronic, Boston Dynamics, and AgileRobot, and that over 100 companies and robotics startups are testing Gemini Robotics.

Last month, China’s DeepSeek invested 140.8 million yuan (approximately 29.2 billion won) in the initial public offering (IPO) of Unitree, a Chinese humanoid and quadruped robot manufacturer. The two companies are also jointly developing an AI model for humanoids, with DeepSeek handling the AI “brain” and Unitree responsible for robot hardware and control technology.

France’s Mistral AI is also making moves. Following its acquisition of Austria’s Emmi AI—which possesses physical and industrial AI technologies—in May, it unveiled its first robotics model in July. The company is targeting the market where AI actually operates equipment in industrial settings such as factories and logistics warehouses.

William Rojas, Director of Strategic Intelligence Research at market research firm GlobalData, analyzed in a recent report on physical AI: “Intelligence no longer remains confined to the digital realm,” adding, “We are shifting from machines that simply followed predetermined commands to systems that perceive, reason, and act on their own.”