[Paek Jongmin's Deep Tech] Beyond Nvidia's Preferred Customer: Toward a Computationally Sovereign Nation
Platform Dependence and Computational Sovereignty Behind the Enthusiasm for Jensen Huang's Visit to Korea
Cooperation with Nvidia Is Essential, but AI Infrastructure Strategy Should Avoid Single Dependence
The U.S. Government Also Diversifies It
Watching the fervor surrounding Nvidia CEO Jensen Huang’s visit to Korea in the first half of 2026, following his trip in the second half of 2025, I can’t help but feel a strange irony. Despite being customers who pay astronomical prices—including the use of taxpayers’ money—to purchase products, we cheered for the seller as if he were a savior. The government, along with major conglomerates such as Samsung, SK, Hyundai Motor, LG, and Naver, all lined up for Nvidia. It is a bizarre scene in which the country paying huge sums for products is far more enthusiastic than the company selling them.
There are valid reasons for the enthusiasm toward Jensen Huang. Korea has proven its global competitiveness in high bandwidth memory (HBM), a core component of artificial intelligence (AI) used in Nvidia’s graphics processing units (GPUs), and the stock price surge of Samsung Electronics and SK hynix has reaffirmed the world-class competitiveness of Korean companies.
However, we must look at the other side of this excitement with a cool head. Cooperation with Nvidia is necessary. The real question is whether we will become Nvidia’s strategic partner or remain just a top customer.
Importing GPUs in large quantities is not merely about buying semiconductors. It also means adopting server architectures, communication networks, software, and developer ecosystems. If we lack an appropriate strategy for this, dependence could become a serious concern.
We should not become complacent with our achievements in HBM. HBM is a valuable asset that breaks through the bottlenecks of the AI era. However, supplying memory does not mean we control the entire AI industry. Even if GPUs with our HBM are running, if the switches for the AI models operating on top of them are held by another country, Korea’s AI sovereignty remains fragile.
A recent incident illustrates this point: The U.S. government restricted access by foreign nationals to Anthropic’s latest models, ‘Fable 5’ and ‘Mythos 5’. There are even rumors that the restrictions began due to issues with a Korean company. Even a nation supplying HBM—key to solving AI bottlenecks—was not exempt.
The actions of our neighbor China add pressure from another direction. China recently overtook the United States in supercomputer rankings with its domestically developed ‘LineShine.’ The fact that China is establishing its own independent computational paths despite U.S. sanctions should not be taken lightly. China now exudes confidence that it can defeat the U.S. in this competition.
Chinese open-source AI models are also advancing rapidly. China is releasing AI models to the public and racing ahead in a “speed battle.” Analysts say this process involves utilizing American models. As a result, China’s open-source AI model ‘GLM’ is said to have nearly caught up to ‘Claude,’ one of the most prominent coding AI models from the U.S., according to users.
In contrast, we remain locked into a policy of developing our own AI foundation models using Nvidia GPUs, pursuing only the most orthodox approach. Concerns are rising in the AI industry about this adherence to orthodoxy. Leading companies are mixing various training methods and combining large-scale and lightweight models. Korea should not remain stuck in a mindset of building everything from scratch.
Just two or three years ago, even after the emergence of the ChatGPT momentum, we emphasized fostering domestic AI semiconductors—neural processing units (NPUs)—over securing foreign GPUs. The direction was not wrong. The issue was our inability to simultaneously address the urgent need for GPUs and the long-term cultivation of a domestic semiconductor ecosystem.
At this point, the U.S. national supercomputer strategy offers valuable lessons. The U.S. does not fill its core AI infrastructure solely with Nvidia. Depending on the purpose, it uses systems based on AMD, Intel, or Nvidia—a so-called computational portfolio strategy. Companies are also seeking alternatives to Nvidia. Korea, too, must not depend solely on Nvidia. We need to foster AMD GPUs, domestic NPUs, and open-source software ecosystems together.
The allocation of public GPU usage rights also needs to be reconsidered. There are aspects where academia requires computational resources. However, professionals in the field lament, “Companies that need to commercialize models and prove them through sales cannot obtain the GPUs they need.” The success or failure of national AI infrastructure depends not on how many chips are secured, but on how quickly they reach those who can actually create services with them.
Hot Picks Today
"Two Spend 500,000 Won in a Flash, Yet Wait Two Hours?" ... 'Dreamland' Shunned by Kids and Foreign Tourists
- The Era of Getting 200 Million Won After 30 Years Is Over... A Japanese Bank's "Perpetual Department Head" Who Just Clocked In Cries Out
- Spain Hit by Its Worst Tragedy... Lee Kangin, Moreno and Real Madrid Join Soccer World in Mourning
- They Said, “Make It Your Home for Life” - Then Suddenly Kicked 200 Passengers Off... What Happened to the Around-the-World Cruise?
- "130 Rice Bowls Sold Out in a Day"... 'Unfiltered Face' Adds 8 Million Won to Night Market Sales: China's Bungeunwoo
Enthusiasm alone does not make a strategy. We do not have much time left to answer whether we can move beyond mere ‘sovereign AI’ to become a ‘computational sovereign nation.’
© The Asia Business Daily. All rights reserved. Unauthorized AI training and use prohibited.