Sakura Internet's GPU Cloud Selected to Power Sakana AI's Trillion-Parameter Model

Osaka-based Sakura Internet's bare-metal GPU cloud service selected as compute infrastructure for Sakana AI's 1-trillion parameter AI agent model development under Japan's national GENIAC program.

OSAKA — Sakura Internet’s bare-metal GPU cloud service Kōkaryoku PHY will provide compute infrastructure for Sakana AI’s development of a 1-trillion parameter AI model under GENIAC Phase 4, a national AI development program led by the Ministry of Economy, Trade and Industry and NEDO.

Sakana AI will build an agent-native foundation model specialized for Japanese institutional practices, customs, and language operations. The company cited existing reliability from prior use of Kōkaryoku PHY as a key factor in the selection. Sakura Internet launched the service in January 2024 and added NVIDIA Blackwell B200 GPU plans in August 2025.

“Large-scale AI development requires a trustworthy compute foundation. We are very reassured to have Sakura Internet’s support for this project,” said Takuya Akiba, chief scientist at Sakana AI.

Why it matters: An Osaka-headquartered cloud provider winning compute infrastructure for a Ministry-backed, trillion-parameter model project shows growing regional capacity to support national-scale AI development outside Tokyo’s traditional dominance. Sakura Internet is headquartered at Grand Green Osaka North Building.