Mitsubishi Heavy Industries (MHI) and Preferred Networks (PFN) signed a capital and business alliance agreement on August 31, 2026, with MHI investing 10 billion yen in PFN through a third-party share allotment. The deal locks in a long-term cooperative framework to build AI systems for mission-critical machinery in social infrastructure and national security, where reliability and rapid response are non-negotiable.
An initial business alliance between the two companies was announced in June 2026. This agreement formalizes the financial commitment and accelerates joint development, pending final procedures including shareholder approval at PFN.
The technical problem they are solving
Deploying AI in power grids, defense systems, and aging infrastructure creates demands that consumer-grade models rarely face. These domains require deep integration with physical equipment, control systems, and field operations. The AI must be understandable and controllable by engineers on site, not just performant in a lab. MHI and PFN argue that maintaining domestic control over the full technology stack - from chips to operational data - is critical for Japan's ability to use these systems independently of external supply chains or geopolitical shifts.
MHI brings decades of operational data and engineering expertise across aerospace, defense, energy, and industrial machinery. PFN contributes a vertically integrated AI stack: the MN-Core series of AI processors, the PFCP cloud platform, and the PLaMo generative AI and LLM foundation model. The alliance aims to combine these assets so that intelligent machinery can be developed and deployed as a single, coherent system rather than a patchwork of third-party components.
What the leadership said
Eisaku Ito, President and CEO of MHI, tied the deal to the company's "ITO" (Innovative Total Optimization) strategy. "PFN's commitment to implementing domestically developed AI technologies across hardware and software deeply resonates with the spirit of craftsmanship that flows through MHI," Ito said. "Deploying AI in mission-critical domains is one of the top-priority issues for strengthening societal resilience."
Daisuke Okanohara, CEO of Preferred Networks, emphasized the practical barriers that make these deployments hard. "To make AI practical on the ground, we must go beyond model performance - we must address field-specific constraints, ensure reliability, real-time responsiveness and continuous operability, while optimizing hardware and software as a whole," Okanohara said. He noted that the joint research had already aligned both companies on this perspective before the capital commitment was made.
The scope of the partnership
The two companies will focus joint R&D on machinery and systems that need high reliability and rapid responsiveness. Target sectors include social infrastructure and national security applications. PFN's full-stack approach - developing its own AI chips, computing infrastructure, and foundation models - is central to the partnership's technical strategy. MHI's role is to supply the domain-specific engineering knowledge, simulation capabilities, and operational field data that generic AI models lack.
The alliance agreement was finalized on August 31, 2026, and the public announcement followed on September 16. The 10 billion yen investment is structured as a third-party allotment of new shares issued by PFN, with MHI as the underwriter.
Why this matters for IT and development professionals
For developers and engineers, this deal signals a growing market for AI systems that run in constrained, high-stakes environments - not just cloud APIs and chatbots. The emphasis on vertically integrated hardware-software stacks, on-device AI chips, and domain-specific foundation models points to technical requirements that differ sharply from mainstream web-based AI. Skills in AI for software developers will increasingly need to span embedded systems, real-time control logic, and hardware-aware model optimization. The MHI-PFN alliance is a concrete example of where that demand is heading: toward professionals who can bridge pure machine learning and the physical machinery it controls.
Your membership also unlocks:
