From Fugaku to FugakuNEXT: NVIDIA and RIKEN Jointly Build a New Foundation for Japan's AI and Quantum Computing

As the global supercomputing race enters the deep water where AI and quantum computing intertwine, Japan has chosen a pragmatic path with strategic depth: tightly coupling a national research institution with the world's leading AI computing power provider.

At SC25, NVIDIA announced a collaboration with Japan's RIKEN to integrate the GB200 NVL4 platform into two new supercomputers—one geared toward AI for science, the other focused on quantum computing research. On the surface, this is a routine computing power procurement; at a deeper level, it is a landmark event in Japan's computing infrastructure shift from "CPU-centric HPC" to "GPU-accelerated AI-quantum-HPC convergence."

Two systems, one strategic anchor

The first system RIKEN is about to deploy features 1,600 NVIDIA Blackwell GPUs, based on the GB200 NVL4 platform and interconnected via Quantum-X800 InfiniBand. Its target scenarios cover life sciences, materials science, climate prediction, manufacturing, and laboratory automation—almost a one-to-one match with the priority areas of Japan's national science, technology, and innovation strategy.

The second system has a stronger exploratory character: 540 Blackwell GPUs are dedicated to quantum algorithms, hybrid simulation, and quantum-classical computing methods. Quantum hardware is still in the noisy intermediate-scale quantum (NISQ) stage, and the synergy between classical computing power and quantum processors is a key path to unlocking practical value. By using the GPU cluster as "escort" computing power for quantum computing, RIKEN is in effect building a hybrid computing framework in advance for future quantum advantage.

The two systems will go into operation in the spring of 2026, exactly connecting with the development window for the next-generation flagship supercomputer FugakuNEXT. It is worth emphasizing that the two systems will also serve as "proxy machines" for FugakuNEXT—used for co-design at the hardware, software, and application levels. This means Japan is not putting all its eggs in one basket; rather, while waiting for the MONAKA-X CPU to mature after 2027, it is first accumulating application ecosystem and system integration experience on the NVIDIA GPU platform.

FugakuNEXT: More than just doubling performance

As the successor to the current top-tier supercomputer Fugaku, FugakuNEXT is jointly designed by Fujitsu and NVIDIA and is planned to be operational by 2030. Its uniqueness lies in adopting Fujitsu's proprietary MONAKA-X CPU and using NVIDIA's proposed NVLink Fusion technology to achieve high-bandwidth direct connection between CPU and GPU. This is not simply "GPU acceleration," but a deep integration of two computing architectures—the CPU handles logic control and traditional computing, the GPU takes on large-scale parallel workloads, and the two form a unified memory semantics through a low-latency, high-bandwidth channel.NVIDIA claims that FugakuNEXT will achieve a 100-fold improvement in application performance over existing CPU systems, and plans to integrate production-grade quantum computers. The strategic implication implicit in this goal is that Japan is no longer pursuing single-peak computing power, but rather emphasizing "application performance" and "heterogeneous integration." As the global supercomputing race shifts toward energy efficiency and actual scientific output, FugakuNEXT's design philosophy happens to keep pace with this shift.

For Fujitsu, cooperation with NVIDIA can both reduce the market risk of its self-developed CPU and leverage the CUDA ecosystem to lower software migration costs. From the perspective of Japan's semiconductor strategy, this "CPU+GPU" combination is also a pragmatic marriage of domestic design capability with the global advanced accelerator supply chain.

Software Ecosystem: The Underestimated Key Move

Beyond hardware, NVIDIA and RIKEN are advancing floating-point simulation software that uses Tensor Core GPUs to accelerate traditional scientific computing. This technology means that 64-bit floating-point operations originally dependent on CPUs can be partially offloaded to GPUs, allowing large-scale HPC applications to truly harness GPUs' AI computing power. For Japan's vast amount of legacy scientific computing code, this is key to lowering the barrier to migration.

Meanwhile, RIKEN plans to adopt the CUDA-X library suite—containing more than 400 optimized libraries, microservices, and tools covering fields such as physics simulation, signal processing, and graph analytics. This means Japanese researchers do not need to develop underlying parallel algorithms from scratch; instead, they can directly call upon globally validated high-performance modules. This model of "leveraging an ecosystem," in contrast to Japan's past preference for a proprietary software stack, demonstrates a more pragmatic attitude toward open innovation.

Long-Term Impact on Japan's Technology Industry

From an industry perspective, RIKEN's deployment sends several clear signals.

First, Japan is treating AI as infrastructure for scientific discovery, rather than as an independent commercial application. The implementation of AI for science will accelerate the R&D cycles of new materials, new drugs, and climate models, and this capability will ultimately translate into competitive advantages in manufacturing and biomedicine.

Second, quantum computing is being incorporated into the national computing system. Through GPU simulation and hybrid algorithm research, Japan can cultivate quantum computing talent and prepare for genuine quantum hardware in the future. This "learn to walk before you run" strategy avoids prematurely locking in uncertain technological paths.

Third, the partnership with NVIDIA is also part of building Japan's computing sovereignty. In areas such as semiconductor equipment, Japan has already shown a willingness to control its supply chain; in supercomputing, leveraging the world's strongest AI accelerator to compensate for its own shortcomings in GPU design is a realistic choice. Although FugakuNEXT still insists on a self-developed CPU, it no longer avoids external GPUs—this marks a new paradigm for Japan's technology industry, shifting from "full-stack autonomy" to "autonomy at key nodes plus open collaboration."Of course, this approach also faces challenges: NVIDIA GPU supply is affected by geopolitical factors, and Japan needs to ensure stable access to advanced chips; at the same time, reliance on the CUDA ecosystem may pose long-term lock-in risks. But judging by the current pace, Japan clearly prioritizes speed over absolute independence.

Japan's Position in the Global Race

While the United States, Europe, and China are all building AI computing clusters, Japan has chosen a rare "dual-track" approach: on one hand, it is rapidly rolling out AI and quantum research platforms with NVIDIA GPUs; on the other, it is jointly developing the next-generation heterogeneous supercomputer with Fujitsu. This strategy secures near-term competitiveness while preserving room for autonomous evolution in the future. RIKEN, as a national research hub, is becoming a bridge connecting the global AI ecosystem with Japan's industrial competitiveness.

Two new systems will come online in spring 2026; by 2030, FugakuNEXT may redefine the performance benchmark for supercomputers. In the meantime, whether Japan can turn computing power into tangible breakthroughs in research and industry will determine whether its revival story can be upgraded from a "hardware narrative" to an "outcome narrative." And at least for now, the RIKEN–NVIDIA collaboration has put Japan at the forefront of AI and quantum convergent computing.