A Warning Sound from The Hague

I am an IT analyst and storyteller who looks at the core of technology through the lens of the humanities. Today, I want to share a somewhat serious story that might hold the future of us and our children in its entirety. It is the story of ‘AI Sovereignty (Sovereign AI)’.

Have you ever imagined this? One day, the email and cloud services you use daily for work suddenly become inaccessible overnight due to the policies of a country on the opposite side of the globe. Sounds like a far-fetched sci-fi movie, right?

Surprisingly, this actually happened in 2023. Prosecutors at the International Criminal Court (ICC) in The Hague experienced the sudden suspension of access to Microsoft’s (MS) cloud services they used daily. They were investigating Russian war crimes, and their work nearly came to a halt in an instant. Why? When the U.S. government placed the ICC on a sanctions list, Microsoft, a U.S. company, stopped the service to comply with domestic law.

This incident sent chills down the spines of policymakers worldwide. It starkly revealed how dangerous it is to entrust all our data to massive foreign corporate servers in exchange for convenience, and how technological dependence can instantly undermine our sovereignty.

Now, this chilling fear is approaching us as an unprecedented tsunami in the AI era. If a cloud service shutdown was like cutting off a single military unit’s supply line, AI dependence is like handing over the nation’s central nervous system entirely.

Chapter 1. Discovery of a New Continent: The Operating System of the AI Era Called ‘Foundation Model’

You’ve probably heard the terms GPT or Gemini countless times recently. These massive AI models are technically called ‘Foundation Models.’ As the name suggests, they are the ‘foundation’ of the AI ecosystem—the starting point of everything.

Why is this so important? Let’s take a brief time machine ride back to the IT era.

• 1980s, PC Era: Many companies wanted to develop programs running on PCs, but the platform that made it all possible was Microsoft’s ‘Windows.’ By controlling the Windows operating system (OS), Microsoft set all the rules of the PC era and amassed enormous wealth from every software company selling products on it.
• Late 2000s, Smartphone Era: Then Google’s ‘Android’ and Apple’s ‘iOS’ dominated as new OSes. No matter how brilliant your app was, you had to follow the app store rules and pay the fees (so-called ’toll’) set by Google and Apple.

Exactly. The Foundation Model is the ‘Operating System of the AI Era.’

If past OSes ruled the ‘digital territories’ of PCs and smartphones, the Foundation Model now sucks up the world’s text, image, and video data like a giant vacuum cleaner, learning all human knowledge and building a new ‘AI territory’ on top. All rules and order in this territory are decided by the ’landlord’ who owns it.

The painful reality is that most of this vast new continent is monopolized by a few ‘digital empires’—giant tech companies from the U.S. and China like Google, Microsoft, OpenAI, Meta, Baidu, and Tencent. To borrow Park Tae-woong, Chairman of Hanbit Media’s expression, we are merely spectators of the discovery of this AI new continent, facing the miserable fate of becoming ’tenant farmers’ forced to rent land within the boundaries they set.

Chapter 2. “Why Not Just Rent?”: Three Hidden Poisons Behind the Sweet Temptation

At this point, many ask a very rational and reasonable question:

“There are already such well-made models. Why should we pour tens or hundreds of billions of won and manpower into building a new one? Isn’t it much more efficient and economical to just rent them via API?”

That is a perfectly valid question. You don’t need to build a tire factory to make a car. But the Foundation Model, the ‘AI engine,’ is not a simple component. It hides three fatal traps we cannot ignore.

First, Security: A Digital Dagger Piercing the Heart

Recall the ICC case. What if it had been our Ministry of Defense, National Intelligence Service, or nuclear power plant?

Imagine processing sensitive military operation plans, diplomatic secrets with key allies, citizens’ medical and financial records, and national critical infrastructure blueprints on a foreign company’s Foundation Model. Would we truly have final control over that data? In times of sudden international turmoil, can we guarantee 100% that they won’t suspend services for their own interests or even analyze our data to use as a ‘digital dagger’ against us?

Just as the U.S. controls semiconductor equipment exports to pressure rivals, in the future, ‘AI model access rights’ will become the most powerful diplomatic weapon. Without our own ‘AI engine’ to protect us, our nation’s fate could be helplessly subject to their decisions.

Second, Culture: Ourselves in a Distorted Mirror

A more subtle and frightening problem is cultural and linguistic dependence.

Most global Foundation Models are trained primarily on English data from the U.S. and Western countries. When asked, “Whose territory is Dokdo?” they respond, “Dokdo is a disputed territory between South Korea and Japan,” which is unacceptable to us. They simply provide the ‘most statistically probable’ answer based on their training data.

Is this limited to the Dokdo issue? AI that fully understands and reflects our history, the process of achieving democracy, social values, and the subtle nuances unique to the Korean language (e.g., ‘jeong,’ ‘han,’ ’nunchi’) can only be created with our own data and by our own hands.

Relying on foreign AI for education, media, and content creation is tantamount to handing our children a ‘history textbook biased by foreign perspectives.’ Our identity will be trapped in a ‘distorted mirror’ they created, slowly losing its light. Just as search engines changed how we acquire knowledge, AI will shake our ways of thinking and values. Losing sovereignty in this invisible war could reduce us beyond economic colonies to ‘cultural and spiritual colonies.’

Third, Economy: The Eternal Shackles of ‘Digital Tenant Farming’

The most practical problem is economic. As long as we rent foreign AI engines without our own Foundation Model, we will forever pay expensive license fees and remain ’technological tenant farmers.’

We have seen smartphone app developers give 30% of their revenue to Google and Apple as ‘app store fees.’ This dependence will deepen in the AI era. All future industries—finance, healthcare, law, manufacturing, content—will innovate based on AI. But if the heart of that innovation is someone else’s engine, what happens?

No matter how excellent our AI services are, a significant portion of the value they create will be continuously paid to U.S. and Chinese ’technology landlords’ as engine usage fees. This is like tenant farmers who toil on fertile land but have most of the harvest taken by landlords. The ‘digital rent economy’ structure, where all industry profits are absorbed by a few global big tech companies owning AI engines, will become entrenched. This is the reality of ‘digital colonialism.’

Chapter 3. The Thorny Journey: Four Massive Mountains Toward ‘Our Own Engine’

Amid this urgency, the concept of ‘Sovereign AI’ has emerged. It means processing our data within our territory, under our laws and regulations, with AI made by our own hands. The alpha and omega, the key to realizing Sovereign AI, is having ‘our own Foundation Model.’

Recently, the Korean government appointed two experts who led the development of Naver’s ‘HyperCLOVA X’ and LG’s ‘ExaONE’ as top AI policy officials. This is a clear response to the pressing times: “We will no longer rent. We will build our own AI engine and establish our own AI territory.” It is a strong declaration of intent.

This is not a clumsy ‘follower strategy’ chasing the U.S. or China. It is closer to a ‘declaration of technological independence’ to decide our fate ourselves in the face of the great wave of AI civilization.

Of course, the path will not be easy. We face four massive mountains standing in our way.

• First, the Mountain of Capital: Astronomical “Ammunition” Needed Training a massive AI model requires enormous costs. Tens of thousands of the latest GPUs must run nonstop for weeks or months. Nvidia’s newest AI chip, ‘Blackwell,’ costs $40,000 (about 55 million KRW) each. Securing thousands of these chips, building data centers, and consuming massive electricity requires trillions of won in funding. Companies like Google and MS enter this war backed by huge cash flow, but it is a scale difficult for our companies and government to handle.
• Second, the Mountain of Data: Lack of “High-Quality Korean Data” AI model performance depends on the quantity and quality of training data. English dominates global web data. Korean data is not only quantitatively insufficient but also difficult to secure ‘high-quality’ data free from copyright issues and bias. Building and refining specialized data in fields like law, healthcare, and finance requires national-level effort.
• Third, the Mountain of Talent: A Global War for the “AI Brain” The world is in a silent war to secure the best AI researchers and developers. Global big tech offers billions in salary and stock options to build ‘AI dream teams’ in Silicon Valley. Can we realistically compete? We must critically assess whether domestic universities’ AI education and corporate research environments are attractive enough to retain talent.
• Fourth, the Mountain of Awareness: Beyond the Trap of “Short-Term Efficiency” Perhaps the hardest mountain is internal perception. From a short-term efficiency and economic viewpoint, “just rent it” always sounds more reasonable. There is even cynical skepticism: “We’re not making semiconductors; why do AI ourselves?” Convincing the public and policymakers of the long-term strategic value of AI sovereignty and building social consensus is as challenging as technology development.

Chapter 4. South Korea’s Compass: What and How Should We Do?

Should we despair and give up facing these four massive mountains? No. We have our own way and the path we can best take. In the 1970s, we built semiconductor factories and car engines on barren land despite everyone saying it was impossible. That resilience remains.

Where should South Korea’s compass point in the AI era?

First, a “Strong Small Nation” AI Strategy Toward “Optimal,” Not “Best”

Trying to build a general-purpose model bigger than GPT-5 right now might be reckless. Catching up with Google and MS’s capital and data scale is realistically difficult. But we have a wise card: ‘selection and concentration.’

Instead of an all-purpose ‘omni AI,’ we build ’expert AI’ specialized in fields we excel at. Pour all our capabilities into world-class manufacturing, semiconductors, healthcare, and K-culture content to build a ‘vertical sovereign AI’ unmatched in those fields. For example, AI optimized for semiconductor design and yield management, precision medical AI based on Korean genetic and medical records, and content generation AI targeting global K-pop fans. Small but powerful AI—this is the path.

Second, Forming “Team Korea”: Uniting Public, Private, and Academia

Developing AI engines is now a ’national competition’ impossible for individual companies alone. The government must create a ‘playing field’ where companies can freely compete with steady long-term R&D investment and bold regulatory innovation, looking 10 to 20 years ahead. Universities must become ’talent academies’ producing core experts ready for industry. Large companies like Naver, Kakao, LG, SKT, and KT should compete in developing their models but unite as ‘one team’ on national goals like data infrastructure and talent cultivation.

Third, Building a “Data Dam”: Stockpiling the Oil of the AI Era

If data determines AI performance, we must systematically collect, refine, and safely utilize high-quality public and industrial data scattered across the country to build a ’national data dam.’ Privacy and data security are paramount, but locking down all data under the pretext of ‘security’ is like fearing maggots and not making soy sauce. Laws and systems must be revised to actively use safely de-identified data for AI training, and social consensus must be urgently built.

If We Fail to Build Our “AI Factory” in the 21st Century

Why did we endure hunger to build the Gyeongbu Expressway, light the blast furnace at Pohang Steelworks, and construct semiconductor factories in the 20th century? It was a fierce declaration of independence: we would not pass down poverty and dependence to our descendants but decide our own fate.

In the 21st century, that fierce industrial battlefield has simply shifted to the ‘AI territory.’ Building our own ‘AI factory’ and creating our own ‘AI engine’ carries the same weight and meaning as building highways and steelworks in the past.

AI sovereignty is no longer a discourse or choice for technology experts alone. It is a survival issue determining whether we and our children will live as proud citizens of a ‘digital sovereign nation,’ not as subjects of a ‘digital colony.’

A future where the heart of an AI engine made by our own hands beats strongly—that is the South Korea we must build together. The journey will surely be tough and arduous, but once again, we can achieve it.