SoftBank is building a physical AI model to autonomously control robots

SoftBank is working on a next-generation "physical AI" model that can control robots and machines without human input. The project, backed by major Japanese firms, aims to bring AI directly into real-world industrial use by 2030.

In Short

• SoftBank and partners plan a 1 trillion-parameter AI model for autonomous machine control
• The AI aims to control robots autonomously
• The project is backed by companies like NEC, Honda, Sony and major Japanese banks

In a move that shows how quickly AI is moving beyond chatbots and into the real world, SoftBank is now working on something far more ambitious, an AI system designed to control machines and robots on its own. The idea is not just to make AI "think," but to make it "act" in factories, industries, and potentially everyday environments.

According to a report by Nikkei Asia, SoftBank has set up a new company in Japan with backing from several major players across industries. Names like NEC and Honda Motor are part of this effort, along with other large corporations including Sony Group and leading banks such as MUFG Bank. The project is being positioned as a large public-private collaboration, and even the Japanese government may step in with support.

At the centre of this initiative is what is being called "physical AI." Unlike traditional AI systems that mostly deal with text, images, or conversations, this model is being designed to directly control robots and machines. The aim is to create a foundation model — similar in concept to what powers tools like ChatGPT, but one that can understand and respond to real-world environments.

The company plans to build a massive AI model by the end of this decade, targeting around 1 trillion parameters. In simple terms, that would place it among the most advanced AI systems developed in Japan. More importantly, the system is expected to handle multiple types of data, including images, video, audio, and text, allowing it to make more accurate and context-aware decisions when interacting with machines.

To bring this vision closer to reality, SoftBank is also collaborating with Preferred Networks, a known player in AI development. Together, they aim to create a system that can eventually work alongside humans in manufacturing environments. The goal is to enable smoother coordination between people and machines by around fiscal 2030.

Another key part of the plan involves building an ecosystem around this AI model. Companies involved in the project are exploring the idea of forming a consortium that would encourage adoption across industries. This means feeding the model with industry-specific data — from finance to automotive to materials — so that it can be tailored for real-world use cases.

There is also a strong strategic angle behind this move. Right now, the global AI race is largely led by the US and China, with companies like OpenAI, Anthropic, and Alibaba Group building powerful foundation models. Many Japanese firms currently rely on these foreign platforms. However, as AI starts handling sensitive industrial data, concerns around data security and leakage are growing.

SoftBank’s approach directly addresses this issue. Alongside the AI model, the company is planning to build a dedicated data centre in Japan. A former LCD panel factory in Sakai, Osaka — acquired from Sharp — will be converted into a high-performance computing hub. This facility will use advanced GPUs to process data domestically, reducing the need to send sensitive information overseas.

This local infrastructure is expected to play a big role, especially for industries dealing with confidential operational data like machine tools and manufacturing systems. By keeping everything within Japan, the project aims to offer both performance and security — two factors that are becoming equally important in the AI space.

- Ends