AI Jobs in Japan in 2026: Sakana AI, Preferred Networks, and the Tokyo Research Cluster

Japan's AI Moment, Honestly Evaluated

For a long time, Japan was a strange gap on the global AI map. The country has world-class robotics research, strong industrial AI applied work, and one of the largest deep-learning labs in Asia (Preferred Networks). But it consistently produced less in the way of frontier LLM research than its raw R&D spend would suggest, and it lost most of its top AI researchers to US labs through the 2010s.

2024 and 2025 shifted that picture. Sakana AI launched in Tokyo with the explicit goal of building a non-US, non-Chinese frontier-AI lab. The Japanese government committed multi-billion-yen funding to AI infrastructure. Major Japanese corporates (Toyota, Sony, Rakuten, NTT) significantly expanded their AI hiring. And for the first time in two decades, the salary numbers for senior AI engineers in Tokyo started looking competitive with their San Francisco peers on a take-home basis.

The market is still smaller than London or Paris. But for international AI researchers and engineers, Japan in 2026 belongs on the shortlist in a way it didn't in 2022.

Related reading: How to Get a Tech Job in Japan in 2026: Tokyo, Visas, and the English-Friendly Employers · Japan Highly Skilled Professional Visa Guide 2026: Points, J-Skip, and 1-Year PR Track · Japan Tech Salary Guide 2026: Software Engineering, ML, and Product Pay in Tokyo.

The Tokyo AI Cluster in 2026

Sakana AI

Founded in mid-2023 by David Ha (formerly Google Brain and Stability AI) and Llion Jones (one of the eight co-authors of the original Transformer paper at Google). Tokyo-headquartered. Built and released the Evolutionary Model Merge work in 2024 and continues publishing frontier-AI research at a pace closer to top US labs than to any other Japanese AI org. Hires globally with English-first culture. Compensation is opaque but reported to track top US lab levels.

If you want to do frontier LLM research from Asia and you're not based in China, Sakana AI is the most credible option in 2026.

Preferred Networks (PFN)

Japan's flagship deep-learning lab since 2014. Originally built Chainer (the deep-learning framework that pioneered many design ideas later adopted by PyTorch). Today PFN's work spans:

• AI hardware (MN-Core series of deep-learning accelerator chips)
• Industrial AI with Toyota and other Japanese manufacturers
• Pharma research (the PFN-Mirai collaboration with major pharmaceutical companies)
• Open-source models (PFN-LLM family)

~600 employees, headquartered in Otemachi, Tokyo. Hires research scientists and engineers; PhD preferred but not required. Salary range for senior researchers: ¥15M – ¥30M base + bonus.

Sony AI

Distributed lab with offices in Tokyo, Switzerland (Zurich), and US. Strong on game AI (the Gran Turismo Sophy reinforcement-learning work was a high-profile output), image sensing and computer vision, AI ethics, and the increasingly visible work on AI for entertainment and content creation. Hires across all three offices; Tokyo office runs in English at the research level.

Rakuten Institute of Technology

Rakuten's central research arm. Applied focus: e-commerce recommendation, fraud detection, conversational AI for customer service, search relevance. Less "frontier research" than Sakana or PFN, more "applied ML at scale on a real e-commerce platform." Strong career trajectory for engineers who want production impact alongside research opportunity.

NTT Communication Science Labs

One of Japan's longest-running corporate research labs. Speech recognition, NLP, human-computer interaction, machine learning theory. Academic in feel; the right place for ML researchers who want corporate stability with publication freedom.

Big Tech Japan AI Teams

• Google Tokyo — Google Research office covers applied AI/ML and increasingly some DeepMind APAC collaborations
• Meta Japan AI — small but growing, focused on multilingual NLP and content integrity
• Microsoft Research Japan — significantly expanded since 2023; applied AI for Copilot, multimodal research
• Amazon AWS Japan ML — applied ML for AWS AI services and customer engineering
• ByteDance Japan AI — recommendation systems and content understanding for TikTok Japan

Government-Backed AI Research

• RIKEN AIP (Center for Advanced Intelligence Project) — Japan's premier public AI research centre. Strong on machine learning theory, statistical AI, and AI for science. Academic-tier salaries (¥6M – ¥12M) but world-class research environment
• AIST (National Institute of Advanced Industrial Science and Technology) — applied AI research for industrial applications
• The University of Tokyo (UTokyo) AI Center — university-affiliated research with growing industry partnerships

What Japanese AI Roles Pay in 2026

Total compensation ranges, gross annual JPY:

• Junior ML Engineer (0–2 yrs, English-friendly company): ¥7M – ¥10M
• Mid ML Engineer (3–5 yrs): ¥10M – ¥16M
• Senior ML Engineer (5–8 yrs): ¥16M – ¥25M
• Staff ML Engineer / Lead Researcher: ¥25M – ¥40M
• AI Research Scientist (PhD, mid): ¥12M – ¥22M
• Senior AI Research Scientist: ¥22M – ¥35M
• Sakana AI / top-tier reported compensation: notably above ¥30M base with equity component (specific numbers not disclosed publicly; informed estimates suggest some senior roles approach ¥50M+ total)
• Big Tech Japan senior+ AI roles: ¥30M – ¥55M base + equity

One data point worth noting: Sakana AI's compensation packages for senior researchers in 2025 closed the gap with mid-tier US AI labs to within roughly 20% on a take-home, tax-adjusted basis. That's a major shift from 2022, when the gap was 60%+.

Where Japanese AI Hiring Is Different

• Industry-academia partnerships matter more than in US/EU. Researchers often hold joint positions with universities while working at corporate labs. PFN researchers regularly co-publish with University of Tokyo
• Long-term project commitment is valued over rapid pivots. A researcher who spent 4 years on one problem at PFN is seen positively; the same trajectory in US labs sometimes reads as "stuck"
• Hardware ML is a real specialisation. PFN's MN-Core work, Sony's edge AI chips, NTT's optical computing research mean Japan has more credible chip-software co-design work than most other markets
• Multilingual NLP focus is heavy on Japanese-specific challenges. Models trained on Japanese-specific data, working through Japanese tokenisation problems, supporting Japanese honorifics in conversational AI — all are real specialisations

How to Apply, Practically

• Direct careers sites for Sakana AI, Preferred Networks, Mercari ML, Rakuten Institute of Technology — all run their own pipelines with English application processes
• LinkedIn Japan with AI/ML filters — recruiter activity is strong, especially since 2024
• Wantedly — Japan-specific platform for tech startups including AI-focused
• BizReach — premium platform that recruits senior tech (paid for candidates above a salary threshold)
• NeurIPS / ICML / ICLR poster sessions — Japanese AI labs increasingly recruit at top ML conferences; if you're presenting, expect outreach
• University recruiting through UTokyo, Kyoto University, Tokyo Tech — for academic-track positions

The Contrarian AI Career Move

Most ambitious AI engineers and researchers in 2026 default to "join a US or EU frontier lab." That's a defensible move and the natural choice for many. But here's a contrarian observation: Sakana AI is the only credible non-US, non-Chinese frontier-AI lab being built right now. The first 50 engineers and researchers at Sakana will have outsized impact on whether a third pole in the AI race actually emerges.

If you believe diversifying the AI research landscape matters (geopolitically, scientifically, or just because the US/Chinese lab ecosystem has gotten less appealing to work in over the past three years), Sakana is the most interesting career bet in 2026 outside the US/China frontier labs. The compensation is good. The team quality is high. The work is at the frontier. Pick the bet that matches the impact you want to have over the next decade, not the one that pays best today.