The problem
The best AI models think in English and train on compute budgets only superpowers can afford — so Japanese companies get mediocre Japanese comprehension and a strategic dependence on foreign labs for their most sensitive workflows.
How it works
Sakana AI, founded in Tokyo in 2023 by former Google researchers David Ha, Llion Jones, and Ren Ito, builds generative-AI foundation models optimized for the Japanese language that work well with small datasets — efficiency over brute-force scale — and ships them to enterprises and developers.
Pain points
Japanese-language errors in critical business text, GPU costs that lock out domestic players, and data-governance anxiety about sending workloads to overseas models.
Business model
Licensing models to enterprises, strategic partnerships (including Japan’s financial heavyweights), and research contracts — with published methods acting as distribution for adoption.
Challenges
US frontier labs outspend Sakana many times over, so it must keep proving efficiency beats scale; research fame has to convert into recurring enterprise revenue; and talent wars for top ML researchers cut both ways.
Funding
- Raised: ¥20B (~$135M) Series B led by MUFG with Khosla Ventures, Macquarie Capital, NEA, Lux Capital, and In-Q-Tel; ~$379M total (TechCrunch, Nov 2025).
- Valuation: $2.65B post-money ($2.5B pre-money), per CEO David Ha via TechCrunch.
Latest — November 2025
On 17 November 2025 Sakana AI closed its ¥20 billion Series B at a $2.65 billion post-money valuation — one of Japan’s largest AI fundraises — to keep building Japan-focused models, with MUFG leading alongside global venture players.