Google backs Sakana AI to push Gemini models deeper into Japan

This deal is a big tell about how Google wants to win outside its home turf. Instead of just dropping Gemini APIs into Japan and hoping enterprises bite, it’s wiring itself directly into one of the country’s most technically ambitious AI startups. Sakana AI is already valued in the multi‑billion‑dollar range and is explicitly focused on alternatives to standard Transformer architectures and agentic systems. Giving that team privileged access to Gemini and Gemma effectively turns them into a high‑leverage R&D outpost for Google in a strategically important, cash‑rich market.

From a race‑to‑AGI perspective, this is a classic “ecosystem” move. Google gets more real‑world feedback loops for its frontier models in regulated sectors like finance and government, where Sakana is already winning contracts, while Sakana gets compute, models, and brand power it could never match alone. That tightens Google DeepMind’s grip on a regional champion and makes it harder for OpenAI or domestic challengers to displace Gemini in Japan’s enterprise stack. It also increases the odds that novel ideas from Sakana’s research on evolutionary model composition and self‑improving agents feed back into Google’s next‑gen architectures.