What Are RIKEN's AI Science Hub AGIS and the "Rikyu" Supercomputer? Japan's National Science-AI Strategy, Explained

This is Hamamoto from TIMEWELL.

On June 19, 2026, RIKEN announced that it had decided to name its supercomputer dedicated to scientific research "Rikyu." As news, it may look modest. Yet trace the substance and you find a machine that sits at the core of a national strategy: Japan's determination to grow "AI for science" domestically. The development is led by RIKEN's hub called AGIS, and behind it lie a partnership with the U.S. Argonne National Laboratory and a concrete budget of 2.8 billion yen in the FY2025 supplementary budget. In this article, I lay out what RIKEN's AGIS and the "Rikyu" supercomputer really are, the TRIP-AGIS project and its Japan-U.S. partnership, the relationship to Fugaku and Fugaku NEXT, and MEXT's national AI for Science strategy — using only primary sources from RIKEN and the government.

This piece is a spin-off that takes only Japan's moves from the guide to what AI for Science is and digs into them. If you also want the global frontier and the AlphaFold story, reading that one first will give you the full picture. Let me start with three key points.

• Rikyu is a supercomputer dedicated to science, developed and operated by RIKEN's AI hub AGIS. Built from 400 computing nodes carrying NVIDIA's latest chips, it is scheduled to begin operations in July 2026 and works with "Fugaku" to grow AI foundation models for science.
• At its core is TRIP-AGIS, which partners closely with Argonne National Laboratory, the U.S. hub for AI for Science. It is allocated 2.5 billion yen in the FY2026 initial budget and 2.8 billion yen in the FY2025 supplementary budget.
• As the overarching national strategy, MEXT set up an AI for Science Promotion Committee, formulated a basic strategy on March 31, 2026, and invested on the order of 114.3 billion yen in the FY2025 supplementary budget. High school inquiry-based learning is the first step by which students connect to this larger current.

What is RIKEN's AI science hub "AGIS"?

Before the machine itself, let me talk about the hub that runs it. What builds and operates Rikyu is RIKEN's "Scientific Foundation Model Development Program." RIKEN renders this in English as the Advanced General Intelligence for Science Program, with the short name AGIS¹. The name is stiff, but what it does can be put in a sentence: RIKEN is seriously building AI for science.

What AGIS aims for is to take general-purpose foundation models such as large language models as a base, then additionally train them on scientific research data — papers, experimental data, simulation results — to grow "scientific foundation models" specialized for particular fields¹. MEXT documents define a scientific foundation model as a foundation model, learned on ordinary text and images, that has been further trained and tuned on scientific research data for use in science². In other words, rather than using general-purpose AI as is, the idea is to build AI that has absorbed the vocabulary and instincts of life science for life science, or materials science for materials science.

AGIS does not stop at building a model and calling it done. It sets its sights on connecting the scientific foundation models it develops to the laboratory, autonomously running the loop of creating data and improving the model, and developing AI agents that serve as the interface with researchers². By combining lab automation with AI, it aims to speed up the very research cycle of posing a hypothesis, designing an experiment, analyzing the data, and deciding the next move. RIKEN describes the aim as a "dramatic acceleration of the research cycle" and "an expansion of the search space of scientific research"¹. The wheel that researchers once turned by hand, one at a time, begins to spin through the night driven by AI. That is the picture being drawn.

The organizational position matters too. AGIS operates as part of the Transformative Research Innovation Platform (TRIP) that RIKEN has set up, a framework connecting the institute's diverse research resources across fields³. Within it, AGIS carries the development of science AI. The program is led by Makoto Taiji, and it is structured to make the most of RIKEN's own research strengths in specific fields while also partnering with institutions strong in related areas¹. Rather than keeping everything in-house, RIKEN brings data and knowledge together with outside institutions. This "open" way of building is exactly what leads to the Japan-U.S. partnership that comes later.

The truth about the "Rikyu" supercomputer

If AGIS is the software hub, Rikyu is the beating heart of hardware that runs it. On June 19, 2026, RIKEN announced it had decided to name its supercomputer for scientific research "Rikyu"⁴. Operations are being adjusted toward a launch in July 2026, RIKEN explains⁴.

The origin of the name captures the machine's philosophy well. Rikyu carries the meaning of exploring the "ri" — the principles and laws behind natural phenomena — using AI and high-performance computing, and then "mastering" (kyu) them⁴. On top of that, the sound echoes the tea master Sen no Rikyū, and that resonance was part of the reason for the choice; the name even carries an allusion to the tea ceremony's shu-ha-ri, in which AI protects existing knowledge (shu), breaks it with new knowledge (ha), and departs into new frontiers of science (ri)⁴. For a naming by a government-affiliated research institute, it is remarkably poetic. As a name for Japanese-born science AI, I sense it was chosen with an eye to being easy to convey overseas.

The internals are concrete. Rikyu is built from 400 computing nodes carrying NVIDIA GB200 NVL4, with 1,600 GPUs of the generation called Blackwell packed inside⁴. The communication speed between nodes reaches up to 3.2 terabits per second, and it posts more than 64 petaflops in double precision (FP64) and more than 15 exaflops in the low precision (FP8) used for AI computation⁴. Lining up figures alone is hard to feel, but essentially, think of it as a computer built entirely for training generative AI and large foundation models at speed.

What should not be missed is the division of roles with the existing supercomputer "Fugaku." Rikyu excels at massively parallel AI computation, while Fugaku excels at large-scale scientific computation⁴. RIKEN expects that by having the two work together, they will greatly contribute to the development and use of advanced scientific foundation models⁴. Train AI models on Rikyu, run physics simulations on Fugaku, and move data back and forth between them. It is a design that joins AI computation and conventional scientific computation to reach research that neither alone can touch. It looks like the story of building one flashy machine, but it is really a stance of "connecting AI computation and traditional scientific computation on a domestic base."

Beyond Fugaku, its successor "Fugaku NEXT" is waiting in the wings. I will treat the full picture of this computing base in the next section, but the point to hold here is that Rikyu is not a one-off device — it is properly embedded within the flow of Japan's supercomputing strategy.

TRIP-AGIS and the partnership with Argonne National Laboratory

What gives concrete form to Rikyu and AGIS, from the angle of budget and international partnership, is the effort called "TRIP-AGIS." MEXT budget documents clearly set out an item, "development and sharing of AI foundation models for scientific research (TRIP-AGIS)"², and this forms the backbone that supports AGIS's activity as a national project. Under the TRIP headquarters, it develops scientific foundation models and opens their use widely to industry and academia. The name derives from the English "Artificial General Intelligence for Science of Transformative Research Innovation Platform," the documents note².

The international partnership is what matters here. TRIP-AGIS is explicitly described as proceeding with development in deep partnership with Argonne National Laboratory, the U.S. hub for AI for Science². Their relationship is no passing fancy. RIKEN and Argonne National Laboratory signed a memorandum of understanding on AI for Science on April 5, 2024³. Its purpose is to play a central role within the AI for Science framework established between the Japanese and U.S. governments, and it reaches into mutual use of scientific and technical research information, research datasets, and research computing resources, as well as exchanges of researchers and doctoral students and jointly held seminars³. Data, computing resources, and people, brought together across borders. You can clearly see the shape of Japan choosing to compete in the world of science AI not alone, but paired with a top U.S. institution.

The budget backing is verifiable too. TRIP-AGIS is allocated 2.5 billion yen in the FY2026 initial budget proposal (also 2.5 billion yen the previous year) and 2.8 billion yen in the FY2025 supplementary budget². And the partnership between RIKEN and Argonne National Laboratory is noted in the documents as being carried out within this TRIP-AGIS budget². Rather than carving out a separate line for international partnership, they have built the Japan-U.S. collaboration into the very cost of developing scientific foundation models. It reads as a sign that the partnership is not decoration but woven into the practical work of development.

The fields TRIP-AGIS targets first are concrete as well. It is set to begin with the life and medical sciences (for example, searching for drug candidates, predicting how cells respond to stimuli, and predicting adaptation to disease) and the materials and physical-property sciences (proposing material structures that realize a function and ways to fabricate them)². Rather than taking on every field at once, it grows scientific foundation models starting from areas where Japan is strong and data has accumulated, then extends the templates it gains to other fields. I think it is a sound order.

I dig deeper into how such long-running AI agents reshape research in the article on AI-driven research. If you want the background to the "AI that connects to the lab and runs autonomously" that TRIP-AGIS aims for, please read that alongside this.

Fugaku, Fugaku NEXT, and the full picture of the computing base

Look only at Rikyu and you understand it as a single point. In reality, Japan's AI for Science is designed as a "plane" that bundles multiple computing bases. Grasp this and Rikyu's position comes sharply into focus.

MEXT documents cite the development of Fugaku NEXT, HPCI systems, and the like as the computing base indispensable for developing AI foundation models for science². Fugaku NEXT is the new flagship system that will succeed the current Fugaku. On January 27, 2026, four parties — RIKEN, Argonne National Laboratory, Fujitsu, and NVIDIA — announced they would cooperate on advancing frontier HPC (high-performance computing) and AI⁵. Within this framework, RIKEN is described as advancing the development of Fugaku NEXT through international collaboration with Fujitsu and NVIDIA⁵. It is a broad cooperation, bringing together the HPC technology cultivated with Fugaku and NVIDIA's advanced AI and HPC technology, spanning everything from next-generation architecture and system software to scientific applications, and even AI-driven robotic self-driving labs and the integration of quantum and supercomputing⁵.

To organize it: Fugaku is the current mainstay, carrying large-scale scientific computation. Rikyu is a dedicated machine specialized for AI computation, carrying the development of scientific foundation models from July 2026 and working with Fugaku. And Fugaku NEXT is the successor flagship, under development with an eye further ahead. These connect vertically, forming the layer of computing base that underpins AI for Science from below.

Investment in the computing resources themselves is concrete. MEXT allocated 7.6 billion yen in the FY2025 supplementary budget for "developing the environment of the computing base indispensable for AI for Science"². This rests on the recognition that HPCI, the framework sharing the computing resources of 14 institutions nationwide, is already stretched thin, and that strategically bolstering computing resources is urgent for advancing AI for Science². This 7.6 billion yen is expected to develop, in roughly two to three cases, computing resources on the order of about 500 GPUs each, at roughly 4 to 5 exaflops class in AI-performance terms². It goes beyond the story of a single machine, Rikyu, to a nationwide effort to thicken the infrastructure for AI computation. The seriousness toward hardware is visible from both the amounts and the number of cases.

Honestly, I regard this computing base as the lifeline of Japan's AI for Science. However fine the ideas or strategy, if the computing resources to train the models fall short, it becomes a pie in the sky. Lining up Rikyu, Fugaku, and Fugaku NEXT as a domestic lineage, and layering on partnerships with Argonne National Laboratory and NVIDIA. This groundwork, I believe, will decide whether Japan's science AI can hold its own in the world.

MEXT's national AI for Science strategy

RIKEN's Rikyu and TRIP-AGIS are not running on their own. Above them sits the framework of a national strategy laid down by MEXT. Let me finish with an overview of the policy under which each device and project moves.

First, the structure. MEXT set up an AI for Science Promotion Committee and, drawing on its discussions, formulated a "Basic Strategy for Promoting AI for Science" on March 31, 2026⁶. The strategy clearly sets out a direction of growing scientific foundation models domestically, centered on Japan's strong fields such as life science and materials. A line in the national document even reads that "maintaining domestic AI research and development capability is extremely important from a security standpoint"², betraying a sense of urgency about not leaving AI for science to other countries. Partner deeply with the United States, yet keep the core development capability at home. It is a delicate and realistic act of steering, balancing partnership and self-reliance.

Look, too, at the overall budget. For the innovation of scientific research through AI for Science, 114.3 billion yen was allocated in the FY2025 supplementary budget, or 152.7 billion yen including related costs². The FY2026 initial budget proposal also secures 19.3 billion yen². Within this 114.3 billion yen supplementary budget, 37.0 billion yen goes to the core "Program for Scientific Research Innovation through AI for Science" and 7.6 billion yen to the "development of the computing base indispensable for AI for Science" noted above². TRIP-AGIS's 2.8 billion yen is part of this larger current. The machine, the model development, the computing base, and the strategy have begun to point in the same direction. That this much has taken shape in just a few years was, honestly, faster than I expected.

There are also mechanisms to widen the base. The innovation program includes a "project type" (32.0 billion yen), which invests about 2 billion yen per project in priority fields that are Japan's winning cards, and a "challenge type" (5.0 billion yen), which distributes about 5 million yen each to roughly 1,000 projects across every field². This challenge type is the open-call program known as "SPReAD," and its first call, covering fields as broad as the humanities and social sciences, was held from April 17 to May 18, 2026⁷. Invest big in priority fields, while spreading small grants widely to cultivate the base of researchers. It is a two-track design for lifting research capability from the bottom up.

This whole set of moves is positioned as driving the improvement in research capability sought by the 7th Science, Technology and Innovation Basic Plan, the overarching framework of national science and technology policy². The hardware of Rikyu, the model development of TRIP-AGIS, and the software of MEXT's strategy and budget. All of them fitting inside one picture is the shape of Japan's AI for Science as of 2026.

From the classroom to the national strategy

So far this has been a story of researchers and the state. So how does it connect to the high school classroom? I believe inquiry-based learning is the closest, most accessible entry point to this national strategy.

The reason is that the work of AI for Science and the process of inquiry-based learning are strikingly alike. Japan's Courses of Study embed in the goals of the Period for Inquiry-Based Cross-Disciplinary Study a process of "finding a question from your own relationship with real society and real life, setting your own task, gathering information, organizing and analyzing, and summarizing and expressing." Pose a question, gather information, analyze, express. Structurally, this is the same as the work of the AI agent that TRIP-AGIS is trying to connect to the lab. Only the scale and the tools differ; the skeleton is the same. That is exactly why an inquiry class can become a place to experience "AI for Science in miniature." I write about the concrete ways to combine inquiry and generative AI, down to the steps, in the practical article on high school inquiry-based learning × generative AI.

That said, get the order wrong and it all falls apart. Let AI take over the question itself, and inquiry turns into "having AI produce the answer and cleaning it up." MEXT's N-E.X.T. High School Vision, too, sets out a two-stage approach: not shutting AI out, but first cultivating abilities AI cannot replace, then mastering the use of AI. I have organized this thinking in detail in the guide to the N-E.X.T. High School Vision and in the article that digs into "abilities AI cannot replace". Rather than ending the story of Rikyu, a frontier computing base, at "how impressive," I want to deliver it to students paired with the principle that the initiative to pose questions stays in human hands. That is the dividing line that turns inquiry from mere lookup work into an entry point to AI for Science.

At TIMEWELL, too, through WARP for Schools, our program for schools and educational institutions, we help design inquiry-based learning that uses generative AI and provide hands-on support in which students build products with their own hands. We have been involved in developing more than 500 people and have worked on a partnership project with the Tokyo Metropolitan Government (WARP ENTRE). Our representative, Hamamoto, is also involved in supporting education at Musashino University's Faculty of Entrepreneurship. The story of Rikyu and scientific foundation models may look like an event at a distant institute, but it is also the story of the tools today's high schoolers will touch a few years from now. If you work at a school and want to think together about how to connect the frontier of science and AI to classroom inquiry, please take a look.

Summary

• Rikyu is a supercomputer dedicated to science, developed and operated by RIKEN's AI hub AGIS. Built from 400 computing nodes carrying NVIDIA GB200, it is scheduled to begin operations in July 2026 and works with Fugaku to grow scientific foundation models.
• The name means to "master" (kyu) the "ri" (principles) — a coinage that echoes Sen no Rikyū and the tea ceremony's shu-ha-ri, chosen with an eye to being easy to convey overseas as Japanese-born science AI.
• The core project, TRIP-AGIS, partners deeply with the U.S. Argonne National Laboratory and is allocated 2.5 billion yen in the FY2026 initial budget and 2.8 billion yen in the FY2025 supplementary budget. The Japan-U.S. partnership is carried out within this budget.
• Fugaku, Rikyu, and Fugaku NEXT form a lineage of computing bases, with a further 7.6 billion yen for developing the computing base. As the overarching strategy, MEXT formulated a basic strategy on March 31, 2026, and invested on the order of 114.3 billion yen in the supplementary budget.
• High school inquiry-based learning shares the same skeleton — "pose a question and verify it" — as this national strategy. A design in which students never let go of the role of posing the question turns inquiry into an entry point to the frontier.

Japan's science AI is, before we knew it, no longer a story of "someday." July 2026, when Rikyu begins to run, should be a symbolic milestone. What question will you pose in today's inquiry class? That small choice, I believe, becomes the first step for students living in an age when a protein's shape appears in minutes.

---

References

Related articles

• What Is AI for Science? How AI Is Changing Scientific Research, and Japan's National Strategy, Explained Clearly
• What Is AI-Driven Research? How Long-Running Agents Are Reshaping Research
• What Is the N-E.X.T. High School Vision? A Clear Guide to MEXT's High School Reform
• What Are "Abilities AI Cannot Replace"? How High School AI Education Will Change
• High School Inquiry-Based Learning × Generative AI: Practical Steps to Empower Students

Footnotes

1. RIKEN, "Scientific Foundation Model Development Program (AGIS)" ↩ ↩² ↩³ ↩⁴

2. MEXT, "On the FY2025 Supplementary Budget and FY2026 Initial Budget Proposal for AI for Science" (AI for Science Promotion Committee, February 9, 2026) ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸ ↩⁹ ↩¹⁰ ↩¹¹ ↩¹² ↩¹³ ↩¹⁴ ↩¹⁵ ↩¹⁶ ↩¹⁷ ↩¹⁸

3. RIKEN, "RIKEN and Argonne National Laboratory sign a memorandum of understanding on AI for Science" (April 11, 2024) ↩ ↩² ↩³

4. RIKEN, "Name of the AI for Science Development Supercomputer Decided as 'Rikyu'" (June 19, 2026) ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸

5. RIKEN, "RIKEN, Argonne National Laboratory, Fujitsu, and NVIDIA to cooperate on advancing frontier HPC/AI" (January 27, 2026) ↩ ↩² ↩³

6. MEXT, "Basic Strategy for Promoting AI for Science" (March 31, 2026) ↩

7. MEXT, "Program to Create Emerging Challenge Research through AI for Science (SPReAD)" special site ↩