Preferred Networks
Japan's deepest physical-AI lab — Toyota, Fanuc, and Mitsubishi on the cap table.
preferred.jp ↗Preferred Networks is Japan's most important AI company that almost nobody outside Japan knows about. The thesis is unusual: don't chase LLMs, don't compete with US consumer AI, instead build the deepest physical-AI platform for industrial Japan — and the payoff is a cap table with Toyota, Fanuc, Mitsubishi, and the Japanese government, plus a leadership role in Japan's national physical-AI strategy. For anyone investing in industrial AI or robotics, Preferred Networks is the single most important non-US company to understand.
What's going for them.
- 01The deepest Japanese AI lab — 10+ years of continuous physical-AI research (robotics, materials, manufacturing) before the generative AI wave made the category fashionable. Unique technical depth in a category US labs largely don't cover.
- 02Toyota and Fanuc on the cap table as strategic partners — the two largest manufacturing/robotics companies in Japan are structurally invested in Preferred Networks' success. That access is irreplaceable for physical-AI work.
- 03In-house MN-Core accelerator chips — Preferred designed and deployed its own AI training silicon, one of the few non-US teams to build custom chips at meaningful scale.
- 04Participating in Japan's national physical-AI consortium (announced April 2026 with SoftBank, NEC, Honda, Sony) that's building a trillion-parameter model for industrial applications — Preferred is one of the two technical leads.
- 05Long history of frontier research in molecular simulation, materials design, and robotics that predates the current AI hype cycle — the quality of output (and the international research reputation) is higher than headline visibility suggests.
What they built
Preferred Networks builds AI systems for physical applications — manufacturing robotics, autonomous driving, materials simulation, drug discovery, and scientific computing. The product portfolio includes the MN-Core AI accelerator (custom silicon), deep-learning platforms deployed with industrial partners, and a growing business in generative AI for industrial applications. The company operates as both a research lab and a commercial partner to Japan’s industrial base, with a particularly deep relationship with Toyota on autonomous driving and factory automation.
How they got here
Toru Nishikawa and Daisuke Okanohara founded Preferred Networks in 2014, spinning out of an earlier NLP-focused company (Preferred Infrastructure). The thesis was prescient: deep learning would matter far more for physical systems — robots, factories, cars — than for chatbots. The first five years were spent building out partnerships with Japan’s industrial giants. Toyota’s strategic investment in 2017 ($100M) made Preferred one of the few AI companies with direct access to large-scale physical data and deployment infrastructure.
The last five years have seen the company expand in three directions: custom silicon (MN-Core), applied AI for materials and drug discovery (Mitsui Chemicals partnerships), and increasingly, generative foundation models tuned for industrial workflows. The most recent strategic development — Japan’s April 2026 national physical-AI consortium announcement — positions Preferred as one of the two technical anchors (alongside SoftBank) for a sovereign-AI effort specifically focused on physical applications.
What’s ahead
Three things will define Preferred’s next few years. First, physical AI foundation models: the Japanese consortium’s trillion-parameter physical-AI model is a direct bet that industrial AI will need its own dedicated models, not general-purpose LLMs repurposed. If the bet pays off, Preferred is the primary beneficiary. Second, commercial scale: Preferred has been disciplined about not chasing LLM revenue, but the enterprise physical-AI category is finally large enough to support standalone revenue scale. Third, international expansion: the company has been almost entirely Japan-domestic; that’s starting to change with materials-AI collaborations in Europe and the US.
Why it matters
Preferred Networks is the reference case for what a non-US, non-consumer AI lab looks like when it focuses on a specific thesis (physical AI) and executes for a decade. For founders building in robotics, industrial automation, or materials science, the Preferred trajectory is proof that patient, research-driven, partnership-heavy AI can produce a multi-billion-dollar business without a viral consumer moment. For investors, Preferred is the single cleanest way to get exposure to Japan’s industrial AI transition.
Founder interview coming soon.
We'll be sitting down with the founders and operators of the companies we profile — on fundraising, product decisions, and what they're building next. If you're part of the Preferred Networks team and want to share a perspective, get in touch.
Thinking about fundraising or M&A?
Amafi Advisory works with AI companies on strategic, fundraising, M&A, and technical advisory. Even if you're just exploring.