← all stories other 1 sources · 1h ago

Sakana AI CEO David Ha Outlines a Japan-Specific AI Strategy

Sakana AI is betting that a business-to-business model, built around routing between models and human-in-the-loop decision-making, is the viable path for a Japanese AI startup rather than trying to out-build US frontier labs.

Reporting from 1 sources: GIGAZINE.

Sakana AI CEO David Ha Outlines a Japan-Specific AI Strategy

Sakana AI CEO David Ha discussed the company's Japan-focused approach on the Disrupting Japan podcast. Instead of competing with US giants on consumer chatbots, Sakana AI partners with major Japanese firms on task-specific workflows, such as loan document processing at Mitsubishi UFJ Bank. Ha stressed routing tasks to appropriate models as a path to AI sovereignty.

David Ha, CEO of Tokyo-based Sakana AI, laid out the company's strategy on the Disrupting Japan podcast. The startup, founded in 2023 by Ha, Llion Jones, and Ren Ito, does not aim to release a consumer chatbot that competes with US giants. Instead, Sakana AI partners with large Japanese companies on AI workflows for specific tasks. Ha described a project with Mitsubishi UFJ Bank where AI supports drafting and organizing evidence for loan decisions, but does not make the final call. A separate initiative with the SMBC Group uses multiple AI agents to handle information gathering, analysis, and document creation for corporate proposals, with humans reviewing the output.

Ha emphasized routing as a core concept: different AI models have different strengths, and a system should direct simple tasks to cheap, fast models and harder analysis to more capable ones. He argued that routing also reduces reliance on US and Chinese frontier models, giving Japan more control over its AI infrastructure. Sakana Chat, released to general users in March 2026, includes the Namazu series of large language models tuned for Japanese use and a web search function.

Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.

Sources