PFN, MC and IIJ to Establish Joint Venture Preferred Computing Infrastructure for AI Cloud Computing

Estimated reading time: 1 minute

Preferred Networks, Inc. (PFN), Mitsubishi Corporation (MC) and Internet Initiative Japan Inc. (IIJ) announced that the three companies will establish their joint venture Preferred Computing Infrastructure, Inc. (PFCI) on January, 2025. Scheduled to begin operations in early 2026, PFCI will primarily provide, operate and support customers of Preferred Computing Platform™ (PFCP™), an AI cloudcomputing service powered by PFN’s low-power AI processor in the MN-Core™series.

AI technologies, especially generative AI, are being rapidly adopted across various domains and expected to continue evolving as basic infrastructure for society and industries. Development and implementation of such AI technologies require a vast amount of computing power that can be continuously and stably expanded as the demand grows. The biggest issues resulting from the growing computing power, however, are the increasing power consumption and heat. This makes it critical for sustainable development of AI to drastically improve computing performance per watt and densification of arithmetic devices in datacenter infrastructure.

The new joint venture PFCI will build a highly efficient computing infrastructure powered by the energy-efficient MN-Core series and stably provide computing power to meet the rapidly growing demand for AI through its AI cloud computing service PFCP.

At PFCI, PFN will develop the MN-Core series and provide technical support while MC will leverage its network for business strategies and marketing. PFCI will build the computing infrastructure in the NRT12 datacenter operated by MC group company MC Digital Realty, Inc., where the joint venture will implement IIJ’s technological expertise for the maintenance and operation to ensure robust and stable service quality.

Going forward, PFCI will expand its business scope to include high-performance computing for complex modeling and simulation, and support for building customers’ own large language models, as well as provision of computing power for AI.