Use-Able ㅡModel
U - User-centered 사용자 중심
S - Sustainable 확산, 확장 가능한 구조
E - Exportable 지역 → 전국 → 해외(ODA)로 이식 가능
A - Affordable 이해할 수 있는 비용 구조
B - Based on Demand 수요에서 시작하는 콘텐츠
L - Living Infra 살아있는 실증 공간 (전시 아님)
E - Ecosystem-led 협의체로 생태계가 스스로 유지
The position and policy of the Korean government as an ODA donor country, 2026
재정경제부 유상 ODA 중심 K-AI패키지 추진전략 (26.7.20)
Strategy for Promoting the K-AI Package Centered on Concessional ODA (Ministry of Finance and Economy, July 20, 2026)
AI-embedded 인프라를 포함한 입체적 관점의 ｢K-AI 패키지｣ 마련
•
한국형 AI 시그니처 사업모델 개발을 위해, 한국이 사업메뉴를 수원국에 제공하고 수원국이 사업을 선택하는 체계 구축
•
사업메뉴에 경쟁력 있는 우리 기술‧부품을 최대한 포함
•
사업수요 발굴과 금융지원 방안 검토시 EDCF, KSP, MDB 신탁기금‧협조융자, 개발금융 등을 유기적으로 연계
인프라에 맞춤형으로 접목되는 AI 솔루션과 이를 구동할 인프라를 유기적으로 blending(혼합)하여 제공하는 전략
Developing a K-AI Package from an integrated perspective that includes AI-embedded infrastructure
•
Establish a framework in which Korea provides a menu of Korean AI signature project models and the partner country selects the solutions that best fit its needs.
•
Incorporate competitive Korean technologies and components into the project menu to the greatest extent possible.
•
Systematically connect EDCF, KSP, MDB trust funds and co-financing, and development finance when identifying project demand and designing financing arrangements.
A strategy that blends infrastructure with tailored AI solutions and the enabling systems required to operate them.
국무총리 주재, 제58차 국제개발협력위원회 (26.7.23)
58th Committee on International Development Cooperation, Chaired by the Prime Minister (26.7.23)
AI 활용 ODA 정책 추진 방향
- ‘제안형 ODA’를 본격적으로 추진하고 문화ㆍAI 등 우리나라의 새로운 강점 분야를 활용하여 협력국의 수요에 부합하는 사업을 확대한다
- AI 기술 발전의 혜택을 모두가 누리는 ‘글로벌 AI 기본사회’를 실현하기 위해 협력국의 AI 생태계 조성을 지원하고 협력국의 AI 도입·확산 준비 수준과 수요에 따라 AI 기반 구축부터 현지형 AI 모델 개발 지원까지 단계적으로 추진: 협력국 특화형 AI 사업 추진
Policy Direction for AI-Enabled ODA
- Korea will actively pursue proposal-driven ODA and expand projects that address partner countries’ needs by leveraging emerging areas of national strength, including culture and AI.
- To realize a “Global AI Basic Society” in which the benefits of advances in AI are shared by all, Korea will support the development of AI ecosystems in partner countries. Assistance will be delivered in stages—from establishing AI foundations to supporting locally adapted AI models—according to each country’s readiness and demand for AI adoption and diffusion: country-tailored AI programs.
What CDR thinks
Education must be approached with seriousness—especially short-cycle vocational training, where every hour must translate into practical capability.
Equal access to opportunity matters. Yet facilities built with substantial public investment too often become underused assets, surviving mainly as references in a system integrator’s delivery portfolio.
Robotics and AI are evolving too quickly for a hardware-first model to remain effective.
This often leads planners to assume that a full-scale data center is required. In practice, however, the GPU workload for vocational robotics training can be much lighter than expected. We are not building a sovereign foundation model. For robotics training and inference, the priority is to optimize utilization and deploy the right level of computing capacity for actual operational demand.
Some centers invest heavily in GPU servers only to see them become underutilized experimental platforms for a small number of researchers. Infrastructure should instead be sized, governed, and measured according to training outcomes and real industrial use.
What we need is not a supercar, but a skilled driver capable of completing the course. The path ahead is not a straight line; it is complex, changing, and full of real-world constraints.
Practical technologies can be organized, validated, shared, and scaled. The fastest way to understand what works is to test it in real operating conditions.
ASEAN countries are fully capable of implementing more practical robotics/AI education solutions. But action must begin now; further delay risks leaving their workforces permanently on the wrong side of the AI divide.
To properly develop this topic, we need a top-down approach rather than a bottom-up one.
Proposal for ADB TA
Global Physical AI solution for Vocational Education
Core Proposal — Transform ASEAN vocational training from hardware-delivery programs into a sustainable Physical AI capability ecosystem built on Simulation-First, Software-Defined, and Sovereign AI principles.
This strategic proposal uses a US$1.5 million technical assistance grant from the ADB e-Asia and Knowledge Partnership Fund as the entry point for building a US$50–70 million EDCF/MDB national-scale replication pipeline.
Full Slide Deck — 15 Slides
Additional explanation
1. Problem Statement — The Physical AI Divide Is Becoming Irreversible
•
Advanced manufacturing economies are rapidly transitioning toward Physical AI, digital twins, autonomous manufacturing, and software-defined robotics.
•
Developing Member Countries (DMCs) risk structural exclusion from global value chains if they remain dependent on manual and low-technology production.
•
Conventional TVET ODA models suffer from heavy upfront CAPEX, equipment damage and downtime, single-vendor lock-in, and obsolescence within three to five years.
•
The solution must go beyond equipment donation and integrate learning, certification, operation, and scale-up into one ecosystem.
2. Paradigm Shift — Simulation-First and Sim2Real
A skilled driver capable of completing the course matters more than an expensive supercar. Learners train and prove their competence in a virtual environment before touching physical equipment.
1.
Virtual Dojo — Train robotics and Vision-Language-Action models in high-fidelity physics simulations.
2.
Digital Certification — Digitally manage standardized competency assessments and learning records.
3.
Physical Deployment — Deploy only validated people and models to real, multi-brand robotic cells.
This approach minimizes equipment damage and production downtime during training while reducing dependence on the limited availability of expensive hardware.
3. AI Driver’s License — A Physical AI Qualification Framework
•
Just as drivers must pass simulation and practical tests before using public roads, operators of high-value AI and robotic systems should be formally validated.
•
An open ROS 2-based platform assesses basic operation, perception and decision-making, safety, and multi-brand system capabilities.
•
STEP-based learning management and competency records support a certification framework designed for recognition across countries.
•
Employers gain access to verified talent, while training institutions can continuously update curricula in response to industry demand.
4. Sovereign AI and Universal Interoperability
•
Use the Asset Administration Shell (AAS) as a common wrapper connecting robots, sensors, and industrial IoT systems from different manufacturers.
•
An open, vendor-neutral architecture lowers the cost of equipment replacement and technology upgrades.
•
Factory data, robot trajectories, and operational knowledge remain within national and institutional boundaries.
•
The shared ASEAN Knowledge Hub exchanges protected insights and benchmarks—not raw industrial secrets.
5. Transnational ASEAN Alliance Architecture
Country | Potential Lead Partners | Proposed Role |
Thailand | MHESI · DIP | Advanced manufacturing and industrial adoption |
Indonesia | BPSDMI · Ministry of Industry | Large-scale manufacturing workforce development |
Vietnam | MOLISA · DVET | Integration with the vocational education system |
Philippines | TESDA · DTI | Competency certification and MSME adoption |
Rather than repeatedly building isolated national centers, the program establishes a federated knowledge-sharing architecture. Curricula, models, and operating practices validated in one country can be rapidly transferred to others through common standards.
6. Infrastructure — Modular K-AI Labs and Edge Data Centers
•
Containerized, modular labs deployable within existing TVET facilities
•
Integrated simulation stations, physical robotic cells, and GPU-powered Edge AI data centers
•
Local processing of all computation and critical data
•
Optional solar and BESS microgrids for 24/7 operation in environments with unreliable power
•
Phased expansion and utilization optimization based on actual training demand rather than oversized centralized data centers
7. Sustainability — Robot-as-a-Service (RaaS)
Overcoming the Grant Death Valley — Training assets are offered as services to local industry so that centers continue operating after grant funding ends.
1.
K-AI Labs train and certify students and incumbent workers.
2.
Certified talent and robotic assets are made available to local MSMEs.
3.
MSMEs access robotics and AI through monthly subscriptions instead of major upfront capital expenditure.
4.
Revenue is reinvested in equipment maintenance, curriculum updates, and cloud and AI service operations.
The goal is not an aid-dependent facility, but a regional Physical AI hub sustained by its own revenue.
8. Four-Stage Implementation Roadmap
Stage | Core Activities | Financing and Outcomes |
1. Policy & Planning | Baseline assessment, feasibility study, master plan, and institutional design | Confirm country-specific conditions and demand |
2. Pilot & PoC | Pilot initial K-AI Labs, the AAS Cloud Hub, and the AI Driver’s License | US$1.5 million ADB e-Asia/KPF TA |
3. National Scale | National rollout of modular labs, Edge DCs, and the certification framework | EDCF and MDB co-financing |
4. Ecosystem Activation | RaaS, local system integrators, MSME adoption, and joint knowledge creation | Self-sustaining operation and regional replication |
9. The Role of the US$1.5 Million TA
The TA is not a one-off facility investment. It is high-value preparation and demonstration funding designed to reduce the risk of subsequent large-scale investment.
•
Demonstrate Modular Labs and Edge DCs
•
Establish an AAS-based Cloud and Knowledge Hub
•
Develop semi-grant TVET programs and training content
•
Validate the RaaS business model
•
Design monitoring, evaluation, and national scale-up mechanisms
Validated results will catalyze a US$50–70 million EDCF/MDB national replication pipeline.
10. Strategic Value
•
Capital Efficiency — Separating learning from hardware improves cost efficiency compared with equipment-heavy training models.
•
De-risked Assets — Continuous software and model updates extend the useful technological life of hardware.
•
Infinite Scalability — Cloud delivery, common standards, and reusable content reduce the marginal cost of adding countries.
•
Human Development — People’s skills and certifiable competencies—not equipment—become the core assets.
•
Digital Inclusion — Developing-country workforces become co-developers and operators, rather than passive consumers, in the Physical AI era.
Proposed Key Performance Indicators
Domain | Illustrative Indicators |
Workforce | People trained and certified; employment and career-transition rates; participation of women and youth |
Technology | Sim2Real success rate; number of integrated brands; utilization and downtime |
Industry | Participating MSMEs; productivity gains; recurring RaaS revenue |
Sustainability | Share of self-generated revenue; maintenance cost coverage; hours of energy autonomy |
Scale-up | Number of adopting countries and institutions; jointly developed content; follow-on investment mobilized |
No Nation Left Behind in the Physical AI Era.
Together, we can build the final bridge that transforms the latent potential of developing-country workforces into active participation in a regional industrial AI alliance.
