日本語 
The next battle in artificial intelligence may not be decided by who builds the smartest model. It may be decided by who controls the machines capable of running the hardest workloads.
Japan appears to understand that shift. In May 2026, the Japan Science and Technology Agency opened applications for ARiSE, AI to Redesign Scientific Exploration, calling it a flagship initiative supporting Japan’s 2026 Basic Strategic Policy for Promoting AI for Science. At the same time, RIKEN has moved deeper into AI supercomputing and quantum-HPC infrastructure.
That makes Japan’s AI supercomputers more than a hardware story. They point to a larger strategy where AI, scientific computing and quantum technologies start working together. For enterprises, the question is no longer whether this matters. It is how quickly this computing model changes what businesses can build, simulate and discover.
Also Read: AI Identity Security: How Enterprises Are Protecting Human, Machine, and Autonomous AI Identities
Breaking Down the RIKEN-NVIDIA Supercomputing Architecture
Japan’s AI supercomputers are not being built around one machine doing everything. RIKEN is taking a dual-system approach, with each system designed around a different class of workload.
The first is RIKYU, which RIKEN officially named in June 2026. Full-scale operation was scheduled for July 2026. The system has 400 compute nodes and 1,600 NVIDIA Blackwell GPUs built around the GB200 NVL4 platform. It also uses Quantum-X800 InfiniBand networking, with interconnect speeds of up to 3.2 Tbps. RIKYU delivers more than 64.16 PFLOPS of FP64 performance and more than 15.539 EFLOPS of FP8 performance.
The numbers are impressive, but the bigger point is what they are designed to accomplish. RIKYU targets AI for Science, particularly highly parallel AI workloads. Instead of treating AI as an add-on to traditional high-performance computing, Japan’s AI supercomputers are bringing AI directly into the scientific computing stack.
The second system, ROQUO, takes a different path. It is designed as a quantum-HPC hybrid platform that connects RIKEN’s computing environment with external quantum systems. Its purpose is to improve quantum simulations, accelerate quantum algorithm development and explore workloads that are difficult for conventional HPC systems to handle alone.
That division matters. Japan’s AI supercomputers are not simply about adding more processing power. They are about matching different types of compute to different problems.
FugakuNEXT and the Future of Hybrid Quantum-Classical Computing
The next step becomes even more interesting with FugakuNEXT, the planned successor to Fugaku, targeting operations around 2030.
FugakuNEXT is being developed through collaboration between RIKEN, Fujitsu and NVIDIA. At its center is FUJITSU-MONAKA-X, an Arm-based processor designed with AI capabilities and HPC optimisation in mind. The architecture aims to bring CPUs and GPUs into much tighter cooperation through NVIDIA’s NVLink Fusion.
That sounds technical, but the business implication is straightforward. Traditional computing workloads and AI workloads do not always behave the same way. A system that can move efficiently between simulation, machine learning, real-time AI and multimodal processing has more flexibility than one built around a single workload.
Japan’s AI supercomputers are therefore moving toward a more blended computing model. FugakuNEXT is being designed around the idea that simulation, data science and AI will increasingly operate together rather than in separate silos.
The planned 1.4 nm process technology for MONAKA-X adds another layer to that ambition. Yet the real story is not the number itself. The real story is architectural flexibility.
For researchers, that could mean less friction between scientific simulation and AI-driven analysis. For enterprises, it points toward infrastructure that can support large-scale machine learning while still handling complex simulations.
That is why FugakuNEXT should not be viewed as simply ‘Fugaku 2.0.’ Japan’s AI supercomputers are evolving toward a system where different forms of computation can cooperate instead of competing for separate infrastructure.
Translating Scientific Power into Enterprise AI Solutions
This is where the conversation becomes relevant beyond research labs.
A supercomputer does not create business value merely because it is fast. The value appears when that compute allows a company to test more possibilities, simulate difficult conditions or shorten the path from an idea to a usable product.
Pharmaceutical companies, for example, can potentially use this class of infrastructure for computational modelling and scientific experimentation. The advantage is not that a machine magically discovers a drug. The advantage is that researchers can explore complex problems through larger and more demanding computational workflows.
Manufacturing presents an even clearer case. Materials science, simulation, robotics and industrial AI all depend on the ability to process large amounts of information and model complex environments. Japan’s AI supercomputers fit into that direction because the country’s industrial AI programmes are also moving toward AI-ready manufacturing data and robotics foundation models.
METI and NEDO selected 16 projects under the GENIAC computing-resource support programme in June 2026. The programme is designed to provide computing resources for AI foundation model development and strengthen Japan’s generative AI capabilities and social implementation.
METI has also backed work around multimodal foundation models for AI robots and physical AI. That matters because physical AI requires more than a chatbot sitting on a screen. It needs systems that can understand environments, process different forms of information and make decisions within complex physical settings.
Logistics companies could similarly benefit from advanced simulation and optimisation. They may not need a supercomputer sitting inside a warehouse. They need access to computing architectures capable of modelling complicated networks, testing scenarios and improving decisions before those decisions affect real operations.
The same logic applies across sectors. Japan’s AI supercomputers create the underlying computational capacity, while industrial AI programmes create pathways for businesses to use that capacity.
The more interesting shift is therefore not ‘businesses will buy supercomputers.’ Most will not. The shift is that enterprise AI infrastructure is becoming more closely connected to HPC, simulation, robotics and scientific computing.
That convergence could shorten R&D cycles, improve material discovery and make advanced simulation more accessible. The companies that benefit most will be those that understand the workload before they start buying the hardware.
The Geopolitics of AI and Why Sovereign Infrastructure Matters
The global AI race is often framed as a competition between the US, China and Japan. That framing misses the more important issue.
The real competition is increasingly about control.
Countries want access to computing power, but they also want greater control over sensitive data, intellectual property, AI models and critical infrastructure. That is why sovereign AI is becoming less about political branding and more about infrastructure choices.
Japan’s Government AI initiative provides a useful example. The FY2026 programme targets approximately 180,000 government employees and is designed to expand AI use across government.
More importantly, Japan is testing domestic foundation models within the Government AI environment. These include NTT DATA’s tsuzumi 2, Fujitsu’s Takane 32B and Preferred Networks’ PLaMo 2.0 Prime. The models are being tested through the Government Cloud, including Sakura Cloud.
The significance goes beyond government productivity. It shows Japan trying to build an AI ecosystem in which domestic models, government infrastructure and national priorities can work together.
Japan’s AI supercomputers fit into the same broader thinking. Sovereignty does not mean building every component domestically and cutting off outside technology. Japan’s approach is more pragmatic. It combines domestic research institutions and technology companies with global partners while strengthening control over the parts of the AI stack that matter strategically.
That is a more realistic model for the next phase of AI.
Preparing for the 2030 Computing Paradigm
Japan’s AI supercomputers signal a change in what powerful computing infrastructure is expected to do. RIKYU brings large-scale AI into scientific computing. ROQUO connects HPC with quantum systems. FugakuNEXT points toward a 2030 architecture where AI and conventional HPC are designed to work together from the start.
For enterprise leaders, the lesson is not to rush out and buy exotic hardware. It is to examine whether today’s infrastructure can support tomorrow’s workloads.
Data pipelines, AI platforms, simulation environments, security controls and compute capacity need to evolve together. Quantum-safe security also deserves attention where sensitive systems require long-term protection.
The bigger risk is waiting until the hardware becomes mainstream before preparing for it. By then, the competitive advantage may already belong to companies that redesigned their architecture earlier.
Japan’s AI supercomputers are worth watching for that reason. They are not just machines. They are an early signal of where enterprise computing is heading next.
If your organization is still planning AI infrastructure around today’s workloads, it may be time to ask a harder question. Is your architecture ready for the computing problems of 2030?
English