SoftBank and Partners Create Japan AI Development Firm

A consortium led by SoftBank Corp., NEC Corp., Sony Group Corp. and Honda Motor Co. has formed a new company to develop domestic, large-scale foundation models for corporate and industrial use. The venture will recruit around 100 AI engineers, give each lead member more than 10% equity, and intends to apply for government-backed funding from NEDO that totals up to 1 trillion yen over five years.

The new company will focus on developing high-performance foundation models and later expand into domain-specific variants, including models for controlling robots and factory automation. SoftBank and NEC will lead the model engineering and research effort, while Sony and Honda will concentrate on applications in consumer devices, automobiles, robotics, gaming and semiconductors. Preferred Networks is expected to contribute engineers to the buildout. Key investor and stakeholder participation includes:

• •Nippon Steel Corp. and Kobe Steel Ltd.
• •Major banks: MUFG Bank, Sumitomo Mitsui Banking Corp., Mizuho Bank
• •Other corporate minority investors negotiating stakes

Technical priorities likely to matter for practitioners: targeted scale and compute footprint for the foundation models, dataset provenance and curation for industrial domains, optimisation for on-premise or hybrid deployment to meet corporate security requirements, and integration pathways for physical AI that control actuators and robots. Expect early engineering work to focus on model architecture choices that balance parameter scale with inference latency and safety controls for physical systems.

This is a coordinated industrial response to U.S. and Chinese dominance in large-scale model development. The structure mirrors national- or industry-backed initiatives seen elsewhere that combine private-sector capability with public funding to secure domestic AI capability and supply chains. By pooling telecom, hardware, automotive and semiconductor expertise, the consortium aims to compress the time to a usable, domestically governed set of models that Japanese companies can adopt without relying solely on foreign cloud providers.

Why the funding matters: Access to NEDO resources and explicit backing from major banks and manufacturers reduces financial and deployment risk. Government-backed grants can subsidise the heavy upfront compute and talent costs required to train foundation models at scale, while corporate participants provide application pipelines and production environments for rapid productisation.

The venture faces fast-moving external competition from well-resourced U.S. and Chinese teams that have open-source contributions and cloud-scale training infrastructure. To be competitive, the new company will need to prioritise differentiated domain verticals, proprietary industrial datasets, latency-optimised architectures for edge and on-prem use, and integration with robotics control stacks that require rigorous safety and real-time guarantees.

Will the company disclose architecture goals or target parameter ranges, or publish open benchmarks that allow independent comparison? Monitor announcements about training partners, cloud or supercomputer facilities, and whether the firm chooses an open, permissive model release or a guarded commercial licensing approach. Also watch for early pilot deployments in manufacturing and automotive lines that will reveal whether the venture can operationalise 'physical AI' at scale.