This Preferred Networks company profile covers the company's valuation, revenue, funding rounds, investors and history. Preferred Networks works in artificial intelligence, with a focus on AI models, chips and industrial AI. It was founded in 2014 and is headquartered in Tokyo, Japan. Its latest valuation on record is More than ¥300 billion (about US$2 billion) (December 2024). Every figure links to a public source, and anything the company hasn't disclosed is marked as such.
Revenue
Funding history
| Date | Round | Amount | Valuation | Investors | Source |
|---|---|---|---|---|---|
| Dec 2024 | Equity and debt round | ¥19 billion (including debt) | More than ¥300B | SBI Group (lead); Mitsubishi Corporation and others | The Bridge |
Preferred Networks timeline
Key dates from the sources on this page, oldest first. Metric snapshots are listed after the funding events.
- 2014 — Preferred Networks is founded in Tokyo by Toru Nishikawa (CEO) and Daisuke Okanohara.
- Dec 2024 — Equity and debt round: ¥19 billion (including debt), valuing the company at More than ¥300B. Investors: SBI Group (lead); Mitsubishi Corporation and others.
Investors
Overview
Preferred Networks, often called PFN, is Japan's best-known artificial intelligence company. Unlike most AI startups, which focus on one layer of the technology, PFN works across the full stack: it designs its own AI chips, builds supercomputers, develops large language models and foundation models, and applies AI in industries such as manufacturing, robotics, materials science, drug discovery and transportation. It has deep partnerships with some of Japan's largest companies and is widely regarded as a national champion in AI.
Toru Nishikawa, the CEO, and Daisuke Okanohara founded PFN in Tokyo in 2014, building on Preferred Infrastructure, a search and data company they had started in 2006 while at the University of Tokyo. Both are accomplished programmers who competed in international programming contests. PFN became known early for Chainer, an open-source deep learning framework released in 2015 that introduced the "define-by-run" approach later popularised by PyTorch; PFN later moved its own development to PyTorch.
What PFN does
- MN-Core chips: custom AI processors designed for high energy efficiency; supercomputers built with MN-Core have topped the Green500 list of the most energy-efficient supercomputers. PFN is developing newer generations for AI inference and training.
- PLaMo language models: large language models trained with strong Japanese-language performance, offered to businesses and government.
- Matlantis: an AI-powered atomic-level simulation platform for materials discovery, developed with ENEOS, used by chemical and materials companies.
- Industrial AI: projects with partners such as Toyota (autonomous driving and robotics), Fanuc (factory robots) and Hitachi.
- Healthcare and life sciences: AI for drug discovery and medical applications.
Funding
PFN has raised money mainly from Japanese corporations and financial institutions rather than US-style venture funds. Toyota invested significantly in its early years, and Fanuc, Hitachi, Mizuho, NTT and others became shareholders. In December 2024 PFN raised about ¥19 billion, including debt, in a round led by SBI Group with Mitsubishi Corporation and others, at a valuation reported at more than ¥300 billion, making it Japan's highest-valued unicorn. Data providers list further loans in 2025. Reports in 2026 said PFN was considering an IPO to fund mass production of its AI chips. We use the December 2024 figure, converted at roughly US$2 billion.
Business model
PFN earns revenue from joint development projects with large companies, licensing of software and models, subscriptions to products such as Matlantis, and sales of computing services. Its chip business could become a major source of revenue if it reaches mass production. The company doesn't publish detailed financials, so we show revenue as not disclosed.
Competition
In Japanese language models, PFN competes with Sakana AI, SoftBank's SB Intuitions, NTT's tsuzumi models, and global providers such as OpenAI, Anthropic and Google, which also offer strong Japanese performance. In chips, it competes with Nvidia and AI chip startups worldwide. In industrial AI, it competes with large technology and consulting firms. PFN's advantages are its full-stack capabilities, energy-efficient hardware and long relationships with Japanese industry.
Preferred Networks at a glance
Preferred Networks is a Tokyo AI company founded in 2014 by Toru Nishikawa and Daisuke Okanohara. It develops MN-Core AI chips, PLaMo language models and the Matlantis materials simulation platform, and works with partners including Toyota and Fanuc. In December 2024 it raised about ¥19 billion led by SBI Group at a valuation reported at more than ¥300 billion.
Japan's AI strategy
Japan's government has made AI and semiconductors national priorities, subsidising chip factories, supercomputers and domestic AI model development. Concerns about relying entirely on US models for language, culture and data security have encouraged support for Japanese models. Japan's ageing population and labour shortages also create strong demand for automation and robotics, where PFN's industrial expertise is valuable. PFN is well placed to benefit from these policies, but competing with US companies' scale remains a challenge.
Why energy-efficient chips matter
AI computing consumes enormous amounts of electricity, and power availability has become a major constraint on data centre growth worldwide. Chips that perform more calculations per watt can reduce operating costs and allow more computing within the same power budget. PFN's focus on efficiency is a differentiator, although competing with Nvidia's software ecosystem and manufacturing scale is extremely difficult for any chip company.
Questions for corporate partners
- How does PLaMo perform on Japanese-language tasks compared with global models?
- What are the data security and deployment options?
- How mature is the MN-Core software stack for your workloads?
From Chainer to PyTorch
When PFN released Chainer in 2015, most deep learning frameworks required developers to define a fixed computation graph before running it. Chainer's "define-by-run" approach built the graph as the code ran, making it much easier to write and debug complex models. PyTorch, released by Facebook in 2016, adopted a similar approach and became the industry standard. In 2019 PFN announced it would move its own research to PyTorch and stop major Chainer development, contributing to PyTorch instead. The episode illustrates both PFN's technical influence and the difficulty of competing with ecosystems backed by the largest technology companies.
Robotics and automation
PFN has long worked on robotics, from industrial robots that learn to pick objects to household robots demonstrated at trade shows. Japan's shortage of workers in manufacturing, logistics and care gives these applications practical urgency, and PFN's partnerships with Fanuc and Toyota provide paths to deployment.
Timeline
- 2006: Toru Nishikawa and Daisuke Okanohara found Preferred Infrastructure.
- 2014: Preferred Networks is spun out to focus on deep learning and the internet of things.
- 2015: releases the Chainer deep learning framework; Toyota becomes a major partner.
- 2020: its MN-3 supercomputer, built on MN-Core chips, tops the Green500 energy-efficiency ranking.
- 2023–2024: releases PLaMo language models.
- December 2024: raises about ¥19 billion led by SBI Group at a valuation above ¥300 billion.
Latest developments (October 2026)
Bloomberg reported in September 2026 that Preferred Networks intends to pursue an IPO to fund mass production of its MN-Core AI chips, with chief executive Daisuke Okanohara saying samples of its newest chips would reach customers in the first half of 2027, according to a summary by FourWeekMBA. No filing or timetable had been announced. The Tokyo company raised ¥19 billion in late 2024 and ¥5 billion more in 2025, according to its newsroom.
What to watch
Watch IPO plans, MN-Core chip production, new PLaMo models, government AI projects and partnerships with Japanese industry.
Takeaways
Japanese enterprises have a domestic full-stack AI partner. Founders can see how deep corporate partnerships can fund long-term research. Investors should watch whether an IPO values PFN above its last private mark.
Sources
Every figure on this page comes from the public sources below. Where sources disagreed, we used the company's own announcement or a primary news report. Private-company data changes; check the date beside each figure.
Frequently asked questions
What does Preferred Networks do?
Preferred Networks is a artificial intelligence / ai models, chips and industrial ai company headquartered in Tokyo, Japan. It was founded in 2014.
What is Preferred Networks's valuation?
Preferred Networks was valued at More than ¥300 billion (about US$2 billion) as of December 2024, according to The Bridge. Private company valuations are set at funding rounds or share sales and can change significantly between them.
How much revenue does Preferred Networks make?
Preferred Networks is a private company and has not publicly disclosed its revenue. Figures on third-party sites are usually modelled estimates rather than numbers reported by the company.
What was Preferred Networks's latest funding round?
The most recent financing we have verified is a equity and debt round in Dec 2024: ¥19 billion (including debt), at a valuation of More than ¥300B. Source: The Bridge.
Who are Preferred Networks's investors?
Investors named in public announcements include SBI Group, Mitsubishi Corporation, Toyota, Fanuc, Hitachi, Mizuho and NTT.
Who founded Preferred Networks?
Preferred Networks was founded in 2014 by Toru Nishikawa (CEO) and Daisuke Okanohara.
Is Preferred Networks a public company?
No. Preferred Networks is a private company (private, venture backed), so its shares are not traded on a stock exchange and it isn't required to publish its financial results.
Where is Preferred Networks headquartered?
Preferred Networks is headquartered in Tokyo, Japan.