Sakana AI Series B Announcement (sakana.ai)

Tokyo-based Sakana AI announced a ¥20 billion (≈$135M) Series B — bringing cumulative funding to about ¥52 billion and a post-money valuation near ¥400 billion — from investors including MUFG, Khosla Ventures, NEA, Lux, Macquarie, Mouro, MPower, Shikoku Electric’s STNet and In-Q-Tel. The raise will accelerate R&D, deepen applied deployments with large Japanese enterprises (finance partners MUFG and Daiwa; expanding into defense, intelligence and manufacturing), and fund strategic investments and hires to scale a Japan-focused “Sovereign AI” ecosystem that emphasizes sustainable, profitable AI adoption rather than compute-heavy foundation-model races. Technically, Sakana AI is positioning itself as a counterpoint to the current compute-centric trend by advancing resource-efficient paradigms: self-evolving architectures (Darwin Gödel Machine, ShinkaEvolve), Evolutionary Model Merge and AB-MCTS for fusing and orchestrating existing models, the Continuous Thought Machine (a non‑transformer temporal architecture), an AI Scientist multi-agent research system, and energy-efficient edge LMs. Their focus on post‑training (optimizing high‑quality base models for local cultural, domain and regulatory needs) and model-merging/self-improvement suggests practical routes to reduce training costs, enable edge deployment, and build domain-specialized systems — a significant signal for researchers and practitioners exploring sustainable, sovereign and application-specific alternatives to giant, pre-trained LLMs.