How Are Japan's DX High Schools Using Generative AI? Real Adopted-School Examples and Application Tips, Explained with Primary Sources

This is Hamamoto from TIMEWELL.

Japan's DX High School program — officially the High School DX Acceleration Project — is a MEXT (Ministry of Education, Culture, Sports, Science and Technology) initiative that subsidizes learning-environment investments at high schools that emphasize informatics and mathematics in their curriculum and strengthen cross-disciplinary, inquiry-based learning with ICT. It launched with 1,010 schools in FY2024, and 1,249 schools were adopted for FY2026. Newly adopted schools receive 10 million yen each. Many schools spend that on high-performance PCs and 3D printers, but when you read what adopted schools actually do, the differentiator is not the hardware. It is the design decision of where generative AI sits in the learning process. In this article, I use MEXT primary sources to lay out how the program works, walk through named school examples of generative AI use from MEXT's own case study collections, sort the practices into four patterns, and close with practical tips for schools preparing an application.

Three key points up front.

• DX High School covers public and private high schools with fixed subsidies: 10 million yen for new schools, 5 million yen in year two, 3 million yen in year three (more for priority categories). For FY2026, 1,249 schools were adopted (920 public, 329 private), funded by 5.2 billion yen in the FY2025 supplementary budget.
• MEXT's case study collections name real schools: a Tokyo metropolitan high school where Informatics II students build local LLMs, and a private school that pairs generative AI with 3D printers for inquiry-driven making.
• The key to an application is not an equipment list but a three-year learning blueprint. What gets tested is whether your generative AI plans connect to the program's aims: expanding Informatics II enrollment and deepening inquiry-based learning.

What the DX High School program is, and how it reached 1,249 schools

Let me start with the skeleton of the system. The program's official name is the High School DX Acceleration Project; the subsidy itself is the High School Digital Human Resource Development Support Grant. MEXT describes the purpose as fundamentally strengthening the development of talent for digital and other growth fields at the high school level, by supporting environment-building costs at schools that implement curricula emphasizing informatics and mathematics and that strengthen cross-disciplinary, inquiry-based learning using ICT¹. The backdrop is change on the university side. As universities restructure toward digital and science-mathematics fields, the policy logic runs, high schools need a stronger pipeline to make that restructuring pay off².

Eligible institutions are public and private high schools, including the latter stage of secondary schools and the upper secondary departments of special needs schools². Here is how the program has grown, using published figures.

Fiscal year	Adoption status	Subsidy scheme
FY2024	1,010 schools adopted (746 public, 264 private)3	First year of the program. Applications ran January 31 to February 29, 20243
FY2025	Around 200 new schools plus around 1,000 continuing schools4	10 million yen for new schools (12 million yen for priority categories), 5 million yen for continuing schools (7 million yen for priority categories). Prefecture-wide cross-school initiatives also funded at 10 million yen per prefecture4
FY2026	1,249 schools (920 public, 329 private), including 80 in priority categories1	10 million yen for new schools (around 100), 5 million yen for year-two schools (7 million yen for priority categories), 3 million yen for year-three schools (5 million yen for priority categories)2

From 1,010 schools in the first year to 1,249 two years later. MEXT's budget material assumes a program scale of around 1,300 public and private schools⁵, so this is no longer an experiment for a handful of pioneers. The priority categories, which carry higher subsidy amounts, come in three types — global, distinctiveness and appeal, and professional — with 20, 10, and 50 schools respectively adopted for FY2026 (including 10 in a semiconductor-focused slot)¹. The funding source is 5.2 billion yen secured in the FY2025 supplementary budget. MEXT's program briefing materials list example activities: promoting enrollment in Informatics II and advanced mathematics courses, cross-disciplinary and inquiry-based learning, and high school-university articulation⁵.

One point that is easy to miss: the range of eligible expenses is fairly wide. Beyond equipment, it covers outsourcing fees, consumables, personnel costs excluding teaching staff, honoraria, and travel². In other words, the subsidy can pay for external instructors and teacher training, not just hardware. That matters for the application tips later.

For the bigger picture, Japan's high school reform is now anchored by the N-E.X.T. High School Initiative published in February 2026 and the 295.5 billion yen High School Education Reform Promotion Fund behind it; the DX High School program is positioned as one of the related measures⁵. I covered the overall reform architecture in my guide to the N-E.X.T. High School Initiative.

How adopted schools use generative AI: named examples from MEXT's case studies

Now to the main question: what actually happens in classrooms. A common misreading of DX High School is that it is a hardware subsidy. It is true that high-performance PCs and 3D printers dominate the spending. But read MEXT's published case studies closely and the interesting part is not the equipment — it is where each school places generative AI within the learning process. Most of what you find in a web search is written by vendors selling tools; articles that verify school practice against primary sources are rare. Here are examples confirmed in MEXT's case study collections and official school communications.

Tokyo Metropolitan Koiwa High School: building local LLMs in Informatics II

Koiwa High School, a Tokyo metropolitan school adopted in FY2024, runs its Informatics II course as year-long project-based learning⁶. Students form groups of up to four, manage their plans with Gantt charts, and sprint toward a first milestone: presenting at the September school festival. Among the project examples MEXT's material lists is building a local LLM — a large language model running on the students' own machines — using high-performance PCs, Python, and VS Code. Other projects include sports data analysis with heart rate monitors, AI-based pose analysis using the VitPose model, and 3D modeling⁶.

What stands out is the positioning of generative AI. The material says students use generative AI proactively as a consultation partner, in an environment where teachers step in with advice when students get stuck⁶. Students also use generative AI and course materials to work through statistical methods beyond the high school syllabus — multiple regression, principal component analysis — and then apply them to data they collected themselves. The AI is not there to produce answers; it is scaffolding for reaching more advanced content. And the fact that a public metropolitan school is running practice at this level should encourage public school teachers who assume this is private-school territory.

Fujimigaoka High School: generative AI plus 3D printers for inquiry-driven making

Fujimigaoka, a private girls' school in Tokyo, was adopted under the banner of building a DX learning environment to raise students who can live proactively in the AI era⁷. One pillar is hands-on making that combines generative AI with 3D printers. Students first study the characteristics of generative AI and copyright, then use prompt engineering to generate images, convert them into 3D models with CAD, and print them. The sequence runs from design idea to physical object, teaching problem-solving through making⁷.

Equally instructive are the school's course design and numeric targets. Its elective, Integrated Computer Science, runs advance learning toward launching Informatics II, and the plan explicitly commits to targets of a 30 percent Informatics II enrollment rate and a 40 percent science-track university progression rate, both for FY2028⁷. The budget design also goes beyond hardware: alongside 3D printers and high-spec PCs, it includes student-facing courses and the provision of teaching materials and lesson plans for teachers. Equipment, student courses, and teacher training planned as a set. Among the case studies, this is the one I would hand to a school drafting its first application.

Beyond MEXT's collections, some schools publish their practice directly. Kogakuin University Junior and Senior High School, in its second year of DX High School adoption, has brought generative AI squarely into Informatics II⁸. The school introduced a paid ChatGPT team plan and changed how programming is taught: instead of building up syntax from scratch, students prompt the AI to generate a program's basic structure first, then modify and extend it toward their own goals. The teacher in charge describes generative AI as an extension of the brain, according to the school's official site⁸. With a working scaffold in front of them, students start reading code through the lens of what they want to change — a neat inversion of the classic programming class, where only the students who can already write move forward.

Four patterns: where to place generative AI in the learning process

Put the examples side by side and clear patterns emerge. I have not audited all 1,249 schools, so this is an interpretation of MEXT's case studies and public information, but I sort the practices into four patterns.

Pattern	Role of generative AI	Example
1. Treat it as subject matter	Make the workings of LLMs themselves the object of study in Informatics II	Local LLM building (Koiwa)6
2. Use it as a sparring partner	A consultation partner and scaffold in inquiry and projects	Learning statistics with AI as a "consultation partner" (Koiwa)6
3. Use it as a creative tool	Build it into making, from image generation to CAD and 3D printing	Generative AI with 3D printers (Fujimigaoka)7
4. Redesign instruction around it	Rebuild how programming is taught on the assumption AI writes code	Generate the structure, then modify and extend (Kogakuin affiliated school)8

Pattern 1 means opening the box rather than just using the tool. Run a local LLM on your own machine and questions like the relationship between model size and capability, the influence of training data, and compute constraints stop being abstract knowledge and become lived experience. Since Informatics II covers data science, programming, and information systems, making the LLM itself the subject fits the course's intent — and it is the single best use of the high-performance PCs the subsidy buys.

Pattern 2 has the lowest barrier to entry and the broadest reach. Sharpening questions in inquiry learning, getting help understanding statistical methods, having the AI critique a presentation draft. MEXT's Guidelines for the Use of Generative AI in Primary and Secondary Education (Ver. 2.0), published December 26, 2024, put forward human-centered use in schools and strengthening information literacy in a world where generative AI exists⁹. The sparring-partner pattern fits that philosophy: students keep the roles of posing questions and making final judgments, while the AI serves as a partner for deepening thought. I wrote a step-by-step guide in my article on inquiry-based learning with generative AI, and if theme-setting is the bottleneck, the 50 AI inquiry themes collection should help.

Pattern 3's strength is that hands move and objects remain. Generating an image is not the end; carry it through CAD into a 3D print and generative AI starts working as an idea-expanding device. Note that Fujimigaoka designed this alongside instruction in copyright and the nature of AI⁷. In the process of making, students inevitably hit the question of whose work a generated design is — which turns information ethics from a lecture into a felt problem.

Pattern 4 is the hardest and the most powerful. In an era when generative AI writes code, how much sense does it make to start from syntax memorization? The Kogakuin affiliated school's practice answers that question through course design. But this pattern only works if teachers themselves use generative AI deeply; deciding what to let the AI generate and where students must think demands command of both the subject and the tool.

Where to start? My view is that schools should enter through patterns 2 and 3, and move to 1 and 4 once the foundation is in place. The reason is teacher workload. Patterns 1 and 4 require informatics-teacher expertise and carefully built instructional design, while 2 and 3 sit easily on top of existing inquiry classes and recover gracefully from failure. Build small wins in year one, then push into the core in year two — a sequence that happens to match the subsidy's declining schedule.

Application tips: what gets judged is a learning blueprint, not an equipment list

Now the practical part for schools considering an application. First, the calendar. For FY2026, applications were accepted from January 21 to February 27, 2026, and adoption was announced on April 1¹. The cycle opens just after New Year and lands at the start of the school year. Whether future rounds follow the same timing has not been announced, but the safe assumption is that a plan not solidified by autumn will not make it.

The first thing to check is alignment with the program's purpose. As the press release puts it, what the program supports is curricula that emphasize informatics and mathematics, and cross-disciplinary, inquiry-based learning using ICT¹. The briefing materials list example activities: promoting enrollment in Informatics II and advanced mathematics, cross-disciplinary and inquiry-based learning, and high school-university articulation⁵. So even a generative AI plan must be wired through to those outcomes — how it expands Informatics II enrollment, how it deepens inquiry. "We will install a state-of-the-art generative AI environment" is an equipment list, not a plan.

Second, design for the three-year taper. The subsidy steps down from 10 million yen to 5 million to 3 million². The system itself is telling you to build the environment in year one and shift weight to use and adoption from year two. Read in reverse: a plan that buys hardware in year one and stops there runs out of breath. Do what Fujimigaoka did — plan equipment, student courses, and teacher training as a set from the start, so the "use it hard" phase actually runs⁷. And as noted, eligible expenses include outsourcing, honoraria, and non-teaching-staff personnel costs², so budgeting for external instructors and training is squarely within the rules. Personally, I think simply committing to roughly 70 percent equipment and 30 percent investment in people changes how convincing a plan reads — that split is my own rule of thumb, not an official criterion.

Third, set numeric targets. Fujimigaoka wrote a 30 percent Informatics II enrollment rate and a 40 percent science-track progression rate into its plan⁷. Numbers create pressure, but they are also the clearest signal of a plan's resolution. For generative AI, whether you can specify target grades and subjects, sessions per year, and teacher-training counts is what separates a real plan from a poster.

Finally, do not close the plan inside a single school. The priority categories are a narrow gate at 80 schools for FY2026, but slots like the semiconductor focus tie into regional industry¹, and FY2025 added support for prefecture-wide cross-school initiatives at 10 million yen per prefecture⁴. Weaving in boards of education, neighboring schools, and universities also speaks directly to the program's articulation aims.

Keeping generative AI from becoming a one-season showpiece

Let me end with a step back. The three years of DX High School will get the equipment and environment in place. The question is what happens after. Whether generative AI learning keeps running in year four and beyond depends, I believe, not on hardware specs but on whether teachers and students carry the experience of having thought and built something together with AI. The student who ran a local LLM, the student who sharpened a question through AI sparring, the student who turned an AI-generated design into a physical object — those experiences outlast the machines. A rollout that ended with a how-to seminar disappears with the next staff transfer.

At TIMEWELL, through WARP for Schools, we help schools design generative AI-based inquiry learning and walk alongside students as they build AI products with their own hands. We have trained more than 500 people and run a partnership program with the Tokyo Metropolitan Government (WARP ENTRE). If your school is weighing how to handle generative AI in a DX High School plan or post-adoption curriculum, feel free to reach out — starting with a simple exchange of notes is fine. And if you want students to go as far as product development, the hands-on guide to high schoolers building AI should be useful.

Summary

• DX High School is a MEXT program supporting informatics- and mathematics-focused curricula and ICT-enabled, cross-disciplinary inquiry learning; 1,249 schools (920 public, 329 private) were adopted for FY2026
• Subsidies are fixed at 10 million yen for new schools, 5 million in year two, and 3 million in year three, and can fund courses, outsourcing, and honoraria — not just equipment
• MEXT's case studies suggest four generative AI patterns: subject matter, sparring partner, creative tool, and instruction redesign
• The realistic entry points are the sparring-partner and creative-tool patterns, moving to the other two as capacity grows
• Write the application as a three-year learning blueprint with numeric targets, wired to Informatics II enrollment and university articulation — not as an equipment list

A number like 1,249 schools means this is no longer a pioneers' experiment; the standard equipment of Japanese high school education is shifting. The difference will be made not by the machines but by the design of what students learn on top of them. Whether you are drafting next year's application or entering year two as an adopted school, try re-reading your plan against one question: where exactly does generative AI sit in the learning?

---

References

Related articles

• What Is the N-E.X.T. High School Initiative? Japan's High School Reform Explained
• What Are "Skills AI Cannot Replace"? How High School AI Education Is Changing
• Inquiry-Based Learning x Generative AI: Practical Steps to Empower Students
• 50 AI Themes for Inquiry-Based Learning
• A Hands-On Guide to High School Students Building AI

Footnotes

1. MEXT press release, "Announcement of Schools Adopted for the FY2026 High School DX Acceleration Project (DX High School)," April 1, 2026 (in Japanese) ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶

2. MEXT, "FY2026 High School DX Acceleration Project (DX High School)" (in Japanese) ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶

3. MEXT press release, "Announcement of Schools Adopted for the FY2024 High School DX Acceleration Project (DX High School)," April 16, 2024 (in Japanese) ↩ ↩²

4. MEXT, "FY2025 High School DX Acceleration Project (DX High School)" (in Japanese) ↩ ↩² ↩³

5. MEXT briefing material, "6. Related Measures," April 2026 (in Japanese) ↩ ↩² ↩³ ↩⁴

6. MEXT, "FY2025 DX High School Case Studies: Tokyo Metropolitan Koiwa High School (Informatics II, project-based learning)" (in Japanese) ↩ ↩² ↩³ ↩⁴ ↩⁵

7. MEXT, "FY2024 DX High School Case Studies: Fujimigaoka High School" (in Japanese) ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷

8. Kogakuin University Junior and Senior High School, "Informatics II: ICT x Generative AI" (in Japanese) ↩ ↩² ↩³

9. MEXT, "Guidelines for the Use of Generative AI in Primary and Secondary Education (Ver. 2.0)," December 26, 2024 (in Japanese) ↩