Crazy Rich Korea | AI Data Centres, Sovereign AI and the Compute Economy begins with a building that looks surprisingly ordinary from the outside.
Did you know? Korea’s Ministry of Science and ICT finalised a KRW 23.74 trillion budget for 2026, including about KRW 5.1 trillion for national AI transformation. The government’s AI infrastructure plan targets a cumulative 37,000 advanced GPUs in 2026, while the National AI Computing Center is being built through a public-private structure to expand access to compute for industry, research and public AI.
That makes “Korea AI”, “Korean AI data center”, “sovereign AI”, “National AI Computing Center”, “GPU Korea”, “Naver Cloud”, “Samsung SDS AI”, “Korean AI models”, “AI infrastructure Korea” and “Korea data center” powerful search concepts for understanding the physical industrial layer beneath artificial intelligence.
Crazy Rich Korea is not only building smarter software. It is building the electricity, chips, fibre, cooling and computing estates that make intelligence available on demand.
Did You Know? AI Has a Physical Address
Artificial intelligence feels weightless because the interface is usually a screen.
The computation happens somewhere else.
Rows of servers sit inside data centres.
Those servers contain accelerators, memory, storage and networking hardware.
They consume electricity and produce heat.
The “cloud” is a building.
Why Compute Became a Strategic Resource
Modern AI models require enormous computational resources for training and inference.
If a country depends entirely on foreign compute capacity, researchers and companies may face cost, availability and sovereignty constraints.
That makes GPU access increasingly similar to industrial infrastructure.
The bottleneck is not only talent.
It is whether talent can get enough machine time.
The 37,000-GPU Target Shows the Scale of Korea’s Response
MSIT’s 2026 planning described a cumulative national target of 37,000 advanced GPUs from government procurement and the sixth national supercomputer.
The purpose is to support national AI projects, public services, startups, research and regional AI transformation.
A GPU fleet becomes a shared industrial tool.
Compute is being treated like laboratory infrastructure.
Why GPUs Matter So Much to AI
GPUs were originally designed for highly parallel graphics workloads.
That parallel architecture turned out to be extremely useful for the matrix operations used in machine learning.
Modern AI accelerators now sit at the centre of the model-training economy.
One hardware category changed the frontier of software capability.
High-Bandwidth Memory Makes the GPU Useful
AI processors need data fast.
If memory cannot supply data quickly enough, expensive compute units sit waiting.
That is why high-bandwidth memory became strategically important.
Read Crazy Rich Korea | Semiconductors, HBM and the AI Chip Economy.
The AI data centre is a system in which compute and memory have to accelerate together.
Korea’s Semiconductor Strength Gives the Compute Economy a Local Foundation
Samsung Electronics and SK hynix already occupy important positions in memory and semiconductor manufacturing.
That gives Korea a rare advantage.
The country is trying to expand AI infrastructure while already participating in the hardware supply chain underneath it.
Software ambition and manufacturing depth can reinforce one another.
The National AI Computing Center Is a Public-Private Experiment
Korea’s National AI Computing Center is being structured through a special-purpose company backed by government and private-sector participation.
The logic is to combine public mission with private operating expertise.
Large AI infrastructure is expensive and demand can be uncertain.
Public capital can absorb some early coordination risk while private firms contribute technology and commercial discipline.
Why the Haenam Location Matters
The National AI Computing Center project selected the Solaseado area in Haenam through a consortium led by Samsung SDS.
Locating major compute infrastructure outside the Seoul metropolitan area supports regional development and can reduce pressure on the capital region’s power and land systems.
Data centres are becoming tools of regional policy.
Compute Changes Regional Geography
A research lab traditionally benefits from being near universities and talent.
A data centre cares deeply about power, fibre, land, cooling and risk.
Those location factors can pull high-value infrastructure into areas that were not historically technology hubs.
Digital industry creates a new geography around electricity and connectivity.
Why AI Data Centres Need Enormous Electricity
Every computation consumes energy.
At AI scale, millions of operations run across thousands of accelerators simultaneously.
Cooling systems consume additional power.
The data centre therefore becomes an energy-intensive industrial facility.
The digital economy returns us to the power grid.
Power Availability Can Become the Real AI Bottleneck
A company can order servers faster than a region can always build new transmission capacity.
Grid impact assessments, substations and generation therefore matter to AI expansion.
Korea has introduced incentives for AI data centres outside the Seoul region partly because infrastructure congestion is increasingly physical.
Compute policy becomes electricity policy.
Why Nuclear and Renewables Both Enter the AI Conversation
AI operators want reliable power and increasingly care about carbon intensity.
Korea’s energy mix includes nuclear power, fossil generation and growing renewable investment.
Read Crazy Rich Korea | Nuclear Power, SMRs and the Energy Technology Economy and Crazy Rich Korea | Offshore Wind, Renewable Energy and the Green Power Economy.
The AI future needs an energy portfolio underneath it.
Cooling Is the Second Half of Compute
Servers convert electricity into computation and heat.
That heat has to leave.
Air cooling, liquid cooling and advanced thermal systems become increasingly important as rack power density rises.
A data centre is partly a giant heat-management machine.
Why Liquid Cooling Is Growing
AI accelerators can generate much more heat per rack than conventional enterprise servers.
Liquid transfers heat more efficiently than air in many high-density configurations.
That can allow more compute to fit into the same floor area.
Thermal engineering becomes a productivity technology.
Water Use Can Become a Social Constraint
Some cooling systems consume water directly or indirectly.
In water-stressed regions, community concerns may rise even if the data centre creates investment.
Infrastructure should therefore be evaluated across electricity, water and land simultaneously.
Digital growth still consumes physical resources.
Fibre Is the Circulatory System of the AI Economy
A GPU cluster is valuable only if data can move into, through and out of it efficiently.
Data-centre interconnects, backbone networks and low-latency links therefore matter enormously.
MSIT’s 2026 investments include AI-network and data-centre interconnection technologies.
Compute becomes more useful when it is well connected.
Read This Beside Korea’s Telecom Economy
Telecom operators already possess fibre, network-management skills and data-centre experience.
Read Crazy Rich Korea | 5G, SK Telecom, KT and the Connectivity Economy.
The AI infrastructure market naturally attracts companies that already know how to move data reliably.
Sovereign AI Is About Capability, Not Isolation
Sovereign AI does not necessarily mean a country rejects global models or hardware.
It means maintaining enough domestic capability to build, adapt, govern and operate important AI systems under national priorities.
The emphasis is resilience and autonomy rather than technological autarky.
Why Language Makes Sovereign AI Especially Relevant to Korea
Korean language, law, public administration and cultural context differ from English-dominant training environments.
Domestic models can be tuned for Korean vocabulary, institutions and user expectations.
Read Crazy Rich Korea | Korean Language, Hangul and the Global Learning Economy.
Language-specific capability can become national digital infrastructure.
The Model Is Only One Layer of the AI Stack
An AI system depends on:
- chips;
- memory;
- data centres;
- networking;
- training data;
- foundation models;
- application software;
- security;
- governance; and
- skilled users.
Sovereignty becomes meaningful only when enough layers are understood and controlled.
Why Domestic NPUs Matter
NPU stands for neural processing unit.
Korea is supporting domestic AI-semiconductor development to reduce dependence on a single class of foreign accelerator and create local hardware capability.
The goal is not necessarily to replace GPUs everywhere.
Different workloads may favour different architectures.
Inference Is Becoming as Important as Training
Training a large model may happen occasionally.
Inference happens every time a user asks the model to do something.
As AI adoption expands, inference demand can eventually exceed training demand in aggregate.
That changes hardware economics toward efficiency and deployment scale.
Edge AI Changes Where Computation Happens
Not every AI task needs to travel to a giant central data centre.
Phones, cars, robots and factories can run some models locally.
That reduces latency and may improve privacy.
Read Crazy Rich Korea | Samsung, Galaxy and the Consumer Electronics Empire.
The compute economy is becoming distributed from hyperscale centres down to individual devices.
Physical AI Creates Another Huge Compute Customer
Robots need perception, planning and control.
Some training can occur in large data centres while inference happens on the machine itself.
Read Crazy Rich Korea | Humanoid Robots, Physical AI and the Smart Factory Economy.
The robot economy is also a compute economy.
Factories Generate Training Data the Open Internet Does Not Have
A factory produces sensor readings, inspection images, maintenance logs and process histories.
Those datasets may be extremely valuable for industrial AI because they describe real physical systems.
Korea’s manufacturing depth therefore creates data assets that purely software economies cannot easily reproduce.
Industrial Data Is Valuable Because It Is Hard to Collect
One million web pages can be scraped quickly.
One million examples of a rare machine failure may take decades to observe.
Scarcity makes operational data strategically important.
The mature factory becomes a data mine.
Why AI Data Governance Matters
Training data may contain personal information, copyrighted material or confidential industrial knowledge.
Access cannot simply be maximised without controls.
Data governance decides which information can be used, by whom and for what purpose.
The dataset becomes part of institutional trust.
Cybersecurity Is a Compute-Centre Safety System
A compromised AI cluster could expose valuable models, datasets or customer workloads.
Physical security, network segmentation, identity controls and monitoring therefore matter.
Read Crazy Rich Singapore | Cybersecurity, Digital Trust and the Security Economy.
Why Startups Need Shared Compute
A large incumbent can buy thousands of accelerators.
A young startup cannot.
National or cloud-based compute programmes can lower entry barriers for smaller teams.
Read Crazy Rich Korea | Startups, Unicorns and the Venture Capital Economy.
Shared infrastructure can democratise experimentation without making compute free.
Compute Subsidies Need Allocation Rules
When demand exceeds available GPU capacity, somebody has to decide which project gets access.
Priority may go to national missions, research, startups or public services.
The allocation system becomes an industrial-policy decision.
Scarcity creates governance.
Why Universities Need AI Infrastructure
Students and researchers cannot learn modern AI deeply if they only read papers about systems they cannot run.
Hands-on compute access is part of education.
Read Crazy Rich Korea | Education, Hagwons and the Knowledge Economy.
The AI talent pipeline depends on machines as well as teachers.
Supercomputer No. 6 Expands Scientific AI
Korea’s sixth national supercomputer is being built with 8,496 high-performance GPUs under a contract valued at KRW 382.5 billion including maintenance.
Its role extends beyond commercial AI into high-precision science and AI-enabled research.
The same computational infrastructure can support climate, materials, physics and model development.
AI Co-Scientists Change the Research Workflow
Korea’s 2026 science strategy includes AI foundation models and AI co-scientist systems for scientific discovery.
The machine may help generate hypotheses, search literature, design experiments or analyse results.
The goal is not to remove researchers.
It is to increase the number of useful experiments they can evaluate.
Why Nobel-Level Ambition Is Really an Infrastructure Statement
Breakthrough research cannot be scheduled.
But countries can increase the probability of breakthrough by improving talent, funding, compute and collaboration.
Ambition becomes meaningful when the experimentation capacity beneath it rises.
AI Models Are Becoming National Utilities
Government agencies, hospitals, schools and companies may increasingly rely on AI systems for routine tasks.
That makes availability and reliability more important.
A model outage can eventually resemble an IT infrastructure outage.
AI moves from novelty toward utility status.
Why Open Models Matter
Open-weight and open-source approaches can help researchers inspect, modify and deploy models without relying entirely on one vendor.
They can accelerate local experimentation.
The trade-off includes security, misuse and support challenges.
Openness is a capability choice, not a universal virtue.
Why Proprietary Models Matter Too
Closed commercial systems may offer stronger support, safety controls or rapid investment at scale.
A resilient ecosystem can use both open and proprietary models rather than forcing one ideology onto every application.
Diversity reduces dependence.
The Compute Economy Is Capital Intensive
AI can feel like a software boom.
The infrastructure resembles heavy industry.
Data centres, accelerators, cooling, power and networking require billions in capital.
The next generation of software is being built on industrial-scale fixed assets.
Capital Markets Will Decide How Fast Korea Can Build
Public funding matters, but private infrastructure investment will determine long-run scale.
Read Crazy Rich Korea | Chaebol, KOSPI and the Capital Markets Economy.
AI becomes an investment cycle connecting technology optimism to steel, concrete, electricity and finance.
Why Singapore Is a Natural Comparison
Singapore is already a major Asian data-centre and cloud hub but faces hard constraints around land, power and water.
Korea has a larger domestic industrial base and electricity system, but also faces regional grid and concentration challenges.
Compare Crazy Rich Singapore | Data Centres, Cloud and AI Infrastructure.
Both countries show that the AI race eventually becomes infrastructure planning.
What Students Can Learn from Korea’s Compute Economy
Computing
Study GPUs, NPUs, distributed systems, cloud architecture, networking and AI models.
Physics
Study electricity, heat transfer and optical networking.
Mathematics
Study matrices, optimisation, probability and scaling laws.
Economics
Study capital intensity, scarce resources, industrial policy and platform effects.
Geography
Study why data centres locate near power, fibre, land and cooling resources.
Civilisation mechanics
Study how intelligence becomes abundant only after a society builds enough physical infrastructure to manufacture computation reliably.
Ten Vocabulary Words for Understanding the Compute Economy
1. GPU
A graphics processing unit, widely used for massively parallel AI computation.
2. NPU
A neural processing unit optimised for machine-learning workloads.
3. Sovereign AI
Domestic capability to develop, operate and govern important AI systems according to national needs.
4. Data centre
A facility housing computing, storage and network infrastructure.
5. Inference
Using a trained AI model to generate outputs from new inputs.
6. Training
The process of adjusting a model’s parameters using data and computation.
7. Interconnect
High-speed links that move data among servers, accelerators or data centres.
8. Liquid cooling
Using liquid to remove heat from computing equipment.
9. Edge computing
Running computation closer to the user or physical device instead of only in central data centres.
10. Compute capacity
The amount of processing resource available for executing computational workloads.
Frequently Asked Questions
How much is Korea investing in AI in 2026?
MSIT’s final 2026 budget includes about KRW 5.1 trillion for national AI transformation, inside a total ministry budget of KRW 23.74 trillion.
How many GPUs is Korea targeting?
Government plans described a cumulative 37,000 advanced GPUs in 2026 from procurement and national supercomputing initiatives.
What is the National AI Computing Center?
It is a public-private national infrastructure project intended to provide large-scale AI computing capacity to research, industry and public programmes.
Why does Korea care about sovereign AI?
Domestic capability can improve resilience, support Korean-language and industry-specific systems, and reduce excessive dependence on external compute and model providers.
Why do AI data centres use so much electricity?
Large clusters run thousands of high-performance processors continuously and require substantial cooling and network infrastructure.
What is the larger economic lesson?
AI feels like software, but national AI capability is increasingly determined by who can build and power the factories of computation underneath it.
Helpful Reading Across the eduKate Graph
- Crazy Rich Korea | Semiconductors, HBM and the AI Chip Economy
- Crazy Rich Korea | 5G, SK Telecom, KT and the Connectivity Economy
- Crazy Rich Korea | Humanoid Robots, Physical AI and the Smart Factory Economy
- Crazy Rich Singapore | Data Centres, Cloud and AI Infrastructure
References and Current Sources
- Ministry of Science and ICT, MSIT’s 2026 Budget Finalized at KRW 23.7 Trillion, 2025.
- Ministry of Science and ICT, 2026 Work Plan and AI Highway GPU Expansion, 2026.
- Ministry of Science and ICT, Korea to Build Its 6th National Supercomputer in Partnership with HPE, 12 May 2025.
- Ministry of Science and ICT, National AI Computing Center Project, 2025.
Crazy Rich Korea Is Building Factories for Thought
A steel mill turns ore into metal.
A chip fab turns wafers into computation.
An AI data centre turns electricity into useful machine intelligence.
Did you know? That may be the deepest shift in the AI economy—the new factory does not ship boxes. It ships answers.