Preferred Networks
Category: AI Infrastructure
Japan's leading AI unicorn developing vertically integrated AI solutions from chips to applications, including the PLaMo family of Japanese-language LLMs, MN-Core AI processors, and industrial AI solutions. Preferred Networks was founded in 2014. The company is led by Daisuke Okanohara (岡野原 大輔). Based in 1-6-1 Otemachi, Chiyoda City, Tokyo. Team size: 201-500. Total funding raised: $308M. Latest round: Venture Round. Key investors include Toyota Motor Corporation, Fanuc, SBI Group, Development Bank of Japan, Mitsubishi Corporation, Sekisui House, Wacom, Kodansha, Mitsubishi UFJ Trust and Banking, Sumitomo Mitsui Trust Bank, TBS Innovation Partners, Toei Animation.
- Founded
- 2014
- Headquarters
- 1-6-1 Otemachi, Chiyoda City, Tokyo
- Team size
- 201-500
- Total funding
- $308M
Value proposition
Vertically integrated AI value chain — from proprietary AI chips (MN-Core series) and supercomputing infrastructure to generative AI foundation models (PLaMo) and industry-specific solutions — all developed in-house for Japanese and global enterprises.
Products and solutions
PLaMo 3.0 Prime (flagship LLM, Reasoning + Non-reasoning, 256k context, GA Jun 2026), PLaMo Lite (edge SLM), PLaMo Translate, PLaMo-fin-base, PLaMo-VL, PLaMo-SemGround (robot env. understanding, with Preferred Robotics), MN-Core series (MN-Core, MN-Core 2), MN-Core L series for generative AI inference (L1000 / L1100 / L1400, samples H1 2027, commercial by Dec 2027)
Unique value
Only company in Japan that vertically integrates the entire AI stack — from custom-designed AI semiconductors (MN-Core) and supercomputers (MN-3, #1 on Green500) to foundation models (PLaMo) and industry-specific solutions — enabling end-to-end optimization for real-world industrial AI applications.
Target customer
Large industrial enterprises in manufacturing, automotive, materials & chemicals, life sciences, finance, retail, entertainment, and public services; also government agencies and research institutions.
Industries served
Manufacturing; Automotive/Transportation; Materials & Chemicals; Life Sciences/Pharma; Entertainment/Media; Retail & Distribution; Finance/Banking; Public Services; Education; Enterprise
Technology advantage
Proprietary MN-Core AI processors with industry-leading energy efficiency (MN-3 #1 on Green500 three times); MN-Core L series (3D-stacked DRAM on logic) targeting ~50× memory bandwidth vs MN-Core 2 and up to ~10× generative-AI inference vs conventional GPUs; full-stack vertical integration chip→infra→PLaMo 3.0 (256k, Reasoning)→apps; Japanese-language LLM leadership and government adoption (Gennai); deep industrial partnerships (Toyota, Fanuc, NTT, ENEOS, MHI); strong AI hardware/software IP.
How they differentiate
Unlike most AI startups that focus on a single layer (e.g., only LLMs or only chips), PFN is unique in its full vertical integration from custom silicon to applications. Competitors like Sakana AI focus on evolutionary model merging, while PFN builds everything from scratch. Compared to AI chip startups like Groq or FuriosaAI, PFN also develops its own foundation models and has deep industrial partnerships in manufacturing, automotive, and healthcare.
Main competitors
Sakana AI (Tokyo-based AI research startup); NEC (cotomi LLM); LLM-jp Consortium (academic consortium); Groq (AI inference chips); SambaNova Systems (AI hardware/software); FuriosaAI (Korean AI chip startup)
Key partnerships
Toyota Motor Corporation (R&D + investment, Jun 2026 joint research with Toyota Frontier Research Center on MN-Core L for Physical AI), Fanuc, NTT, NVIDIA, Microsoft, Intel, Samsung Electronics (2nm fabrication), ENEOS, National Cancer Center Japan, Snowflake (PLaMo Marketplace), Amazon Web Services (Bedrock Marketplace)
Notable customers
Toyota Motor Corporation; Fanuc; ENEOS; National Cancer Center Japan; Chugai Pharmaceutical; multiple Japanese financial institutions; over 150 local governments in Japan (via QommonsAI)
Major milestones
2014: Founded, 2015: Chainer, 2017: Private supercomputer, 2018: MN-Core, 2020: MN-3 #1 Green500 (×3), 2021: Preferred Robotics, 2023: PLaMo-13B, 2024: PLaMo Prime, PFCP, MN-Core 2, 2024-12–2025-06: Latest round ~¥24B+ equity/debt (+undisclosed Jun extension), 2025-11: Okanohara CEO / Nishikawa Chairman
Growth metrics
Revenue: ¥8,486 million (~$56M) as of FY ending Jan 2021 (latest public); ~400 employees (official Jan 2026 factsheet; ~413–436 social-insurance/OpenMoney 2026); valuation >¥300B (~$2B+); CEO publicly discussing IPO within several years to fund AI chip mass production (Bloomberg Sep 2026)
Market positioning
Japan's leading privately held AI unicorn (valuation >¥300B / ~$2B+), national champion for Japanese-language LLMs (PLaMo) and industrial/Physical AI. Competing globally on energy-efficient AI chips (MN-Core / MN-Core L inference roadmap) and Japanese foundation models, with deep enterprise partnerships. As of Sep 2026, CEO publicly considering IPO within several years specifically to fund mass production of next-gen AI semiconductors (samples H1 2027; commercial target end-2027; profitability goal ~3 years).
Geographic focus
Japan (primary); expanding globally through partnerships (Samsung for chips, Snowflake for cloud deployment)
Patents and IP
MN-Core AI processor architecture patents; Chainer deep learning framework (open source, now transitioned to PyTorch); Optuna hyperparameter optimization framework (open source); PLaMo LLM family (proprietary); multiple patents in AI hardware acceleration and deep learning.
About Daisuke Okanohara (岡野原 大輔)
Co-founder of Preferred Infrastructure (2006); Ph.D. in Computer Science from University of Tokyo (2010); CTO of Preferred Networks before becoming CEO in Nov 2025; also CEO of Matlantis Corporation
Latest news about Preferred Networks
More AI Infrastructure companies
Official website: https://www.preferred.jp/en/