The OECD.AI Policy Navigator

Our policy navigator is a living repository from more than 80 jurisdictions and organisations. Use the filters to browse initiatives and find what you are looking for.

AI Bridging Cloud Infrastructure 3.0 (ABCI 3.0)


Added by:   OECD analyst
Added on:   17 Jul 2026
Updated by:   OECD analyst
Updated on:   17 Jul 2026

ABCI 3.0 is Japan's national AI computing infrastructure that provides access to supercomputing resources for developing sovereign AI capabilities. Operated by AIST and funded through Japan's METI, it delivers 6.2 exaflops of computing power using 6,128 NVIDIA H200 GPUs. The infrastructure supports industry, academia, and government in developing large language models and advancing generative AI research, with priority given to national research institutes, universities, and startups.

Name in original language

AI橋渡しクラウド基盤 3.0 (ABCI 3.0)

Initiative overview

ABCI 3.0 represents Japan's strategic response to growing global AI competition and the need for sovereign AI capabilities. As the third generation of ABCI infrastructure—first launched in August 2018—the system became fully operational in January 2025, providing unprecedented computing resources for AI development within Japan. In June 2025, ABCI 3.0 officially proved its status as Japan’s premier AI supercomputer by ranking 4th globally and 1st in Japan on the HPL-MxP benchmark, which measures mixed-precision AI computing performance. It also secured 15th place in the world (2nd in Japan) on the TOP500 list, reflecting its immense raw computational power.

The initiative addresses Japan's critical need to reduce dependence on foreign AI technologies while developing foundation models with deep understanding of Japanese language and culture. Built by Hewlett Packard Enterprise using NVIDIA's H200 Tensor Core GPUs connected via Quantum-2 InfiniBand networking, the system delivers peak performance of 6.22 exaflops in half precision and 3.0 exaflops in single precision—representing a 7 to 13 times improvement over the previous ABCI 2.0 system. As of May 2025, more than 60 R&D projects involving foundation models, LLMs, and multimodal AI were already active under its "Accelerated Use" program.

ABCI3.0 funded through METI's Economic Security Fund with an initial investment of ¥36 billion ($232 million). The infrastructure prioritizes resource allocation for public and strategic R&D purposes, particularly supporting national institutions, universities, and startups.

ABCI 3.0's mission has evolved from providing general AI infrastructure to focusing on cutting-edge challenges in sovereign AI development and supporting the domestic AI startup ecosystem. The system provides relaxed usage limits compared to previous versions, including increased node allocations and extended reserved durations, enabling larger-scale foundation model development. Looking ahead, ABCI 3.0 is a cornerstone for advancing "Physical AI"—integrating real-world data like 3D point clouds and sound into multimodal foundation models—marking the next frontier in Japan's strategy to enhance industrial competitiveness through sovereign AI technology.

Other relevant details

Results, outcomes and impacts: Since its original launch in 2018, ABCI has served approximately 3,000 users from industry, academia, and government. The infrastructure has enabled the successful development of Japanese large language models including PLaMo and Swallow LLM, which demonstrate enhanced Japanese language capabilities compared to foreign models. Through the LLM Building Support Program launched in August 2023, ABCI has facilitated development of state-of-the-art Japanese language models. Research teams at Tokyo Institute of Technology and AIST used ABCI to develop the Swallow LLM series, achieving superior performance on Japanese language tasks through continual pre-training. Challenges and lessons learned: As the demand for generative AI in industry, academia, and government in Japan grew rapidly, ABCI could not provide enough resources to meet such user demands in a timely manner. ABCI's role has shifted to focus on challenging research and development based on national demands, such as sovereign AI, and encouraging AI startups. ABCI started full-scale operation in August 2018, was upgraded to ABCI 2.0 in May 2021, and was upgraded to ABCI 3.0 in January 2025. Building on experience managing ABCI over the past several years, AIST upgraded to ABCI 3.0 in collaboration with NVIDIA and HPE to develop computing infrastructure that will advance research and development capabilities for generative AI in Japan. By sharing and transferring experience with domestic cloud providers, AIST aims to help them become more competitive in the AI datacenter market.