Japan's ABCI 3.0 Supercomputer Powers AI Innovation with 6.22 Exaflops
Image Credit: Takashi Miyazaki | Splash
Japan’s National Institute of Advanced Industrial Science and Technology (AIST) has been operating ABCI 3.0, a supercomputer for artificial intelligence research, since January 20, 2025. Located in Kashiwa, Chiba Prefecture, the system supports researchers, businesses, and institutions, with a focus on generative AI technologies, such as those used in chatbots and image generation.
Hardware Specifications
ABCI 3.0 includes 6,128 NVIDIA H200 GPUs, processors designed for AI computations. AIST reports the system achieves 6.22 exaflops in half-precision (16-bit) calculations, approximately twice the 3.0 exaflops in single-precision (32-bit), as half-precision allows more calculations per cycle. An exaflop represents a quintillion calculations per second, supporting complex AI models. The system is 7 to 13 times faster than its predecessor, ABCI 2.0, according to AIST and NVIDIA data.
[Read More: How AI is Revolutionizing Sub-100-Gram Drones Under Japan's Strict Drone Laws]
Storage Capacity
The supercomputer has 75 petabytes of storage, combining shared file systems and cloud storage. AIST indicates this capacity is more than double that of ABCI 2.0, with read/write speeds also over twice as fast. This storage supports generative AI, which requires large datasets for training models to produce text, images, or videos. The all-flash storage system enables efficient data processing.
[Read More: Japanese AI Pioneers Overlooked: The Untold Story Behind Neural Network Innovations]
Support for AI Research and Training
ABCI 3.0 is an open platform, used by over 1,000 users, including universities and startups, as of March 2025. It facilitates research on multimodal AI models that integrate text, images, and video, with applications in fields like healthcare and education. The system also supports AI training programs, providing access to tools for developing expertise, and evaluates AI models for performance and reliability. Japan’s AI efforts include collaborations with companies like SoftBank and NVIDIA.
[Read More: UTAR PhD Student Wins Cisco AI Hackathon with Anti-Procrastination AI Platform]
Transition from Previous System
Funded by a ¥36 billion (approximately US$232 million) investment from Japan’s Ministry of Economy, Trade, and Industry, ABCI 3.0 replaces ABCI 2.0, operational since 2018. Ryousei Takano, deputy leader at AIST, oversaw the transition, ensuring ABCI 2.0’s operations continued with minimal disruption during the January 2025 launch.
[Read More: Fujitsu Unveils Multi-AI Agent Security Tech to Combat Emerging Cyber Threats]
International and Environmental Considerations
ABCI 3.0 contributes to Japan’s AI objectives, aligning with NVIDIA CEO Jensen Huang’s remarks on the need for AI infrastructure, made during a 2023 visit to Japan. In March 2025, AIST signed an agreement with a European AI institute to share resources, supporting international research. The system incorporates an energy-efficient design with an upgraded cooling system, including free cooling technology, to support high-density computing.
[Read More: AI Emits 50% More Carbon to the Atmosphere?]
Source: arXiv, Nvidia Blog, Futunn, Asahi
TheDayAfterAI News
We are a leading AI-focused digital news platform, combining AI-generated reporting with human editorial oversight. By aggregating and synthesizing the latest developments in AI — spanning innovation, technology, ethics, policy and business — we deliver timely, accurate and thought-provoking content.