Trillion-dollar South Korean Sovereign AI Investment: Nvidia Wins, SK Hynix Under Pressure

South Korea is elevating artificial intelligence to a national strategy, making a bet of such scale that the market is taking notice. But in this nationwide competition, determining who will truly benefit is much more complex than it appears on the surface.

In July this year, the South Korean government announced plans to invest $919 billion to build 8.4GW of data center computing power by 2029, with a goal to expand total capacity to 18.4GW by 2035.

This plan will open up a vast sovereign AI market for Nvidia, with SK Group committing to a 2GW Rubin system purchase and Naver also signing a 200MW order.

However, for Samsung and SK Hynix, this is not a simple boon. South Korea lacks domestic suppliers capable of independently producing cutting-edge AI accelerators. According to SemiAnalysis, Nvidia may have already signed a long-term agreement with SK Hynix for SOCAMM and locked in preferential HBM pricing for 2027.

Meanwhile, the government-led "independent AI foundational model" competition is underway, but in the most critical elimination round, a puzzling decision was made—the startup Motif Technologies, which ranked first in the benchmark test, was excluded.

This result reveals the inherent contradictions of the sovereign AI race: the government’s intention to pursue technological independence often sharply contrasts with its actual resource allocation logic.

For investors, South Korea’s trillion-dollar commitment represents a significant expansion of Nvidia’s client base, but for Samsung and SK Hynix shareholders, this massive construction plan may not bring equivalent returns.

Why Nvidia Strongly Supports Open-Source and Sovereign AI

Nvidia CEO Jensen Huang recently published an article on the importance of open-source AI, which was subsequently co-signed by nearly all major AI companies (except Anthropic). There’s a clear business logic behind this stance.

Selling frontier model tokens via API pricing is the highest return-on-compute scenario. This means that Anthropic and OpenAI will increasingly account for the net new compute demand, and Nvidia faces the risk of its client base being overly concentrated among a few super-sized buyers.

Even if hyperscale cloud providers are included, truly core customers number only seven, and all except SpaceX are actively developing proprietary chips to reduce dependence on Nvidia.

Sovereign AI is a crucial entry point for Nvidia to diversify its customer base. The classic strategy framework "commoditize your complements" applies here: AI models and applications are complements to Nvidia GPUs, and the prosperity of open-source models translates to broader compute demand.

Nvidia’s recent $500 billion memorandum of understanding with the world’s largest capital allocators is a financial extension of this strategy—by branding "Nvidia AI factory compute" as an investable asset class, pension funds, insurance companies, and private credit funds can funnel capital into large-scale compute purchases.

Data Center Construction Boom: Nvidia Benefits, HBM Landscape Uncertain

South Korea has confirmed the selection of three data center sites under construction, totaling 4.4GW. On the construction side, SK Group, GS Group, and Naver are responsible for the initial construction of 5GW, 2.4GW, and 1GW, respectively, with SK Group also taking on the remaining 10GW for the second phase. President Lee Jae-myung has explicitly positioned AI infrastructure as a historic endeavor on par with the broadband push of the 1990s.

However, this large-scale construction plan is not a simple benefit for South Korean semiconductor companies.

Currently, South Korea still lacks domestic suppliers capable of independently producing cutting-edge AI accelerators. Rebellions and FuriosaAI have made progress, but in terms of deployment scale, software maturity, and commercial availability, they still lag significantly behind Nvidia. Thus, no matter how much money is invested, South Korea cannot build a competitive accelerator ecosystem relying solely on domestic manufacturing power.

This is precisely where the strategic value of Nvidia’s deepening relationship with Samsung and SK Hynix lies. Samsung is reportedly planning to build an AI factory equipped with over 50,000 Nvidia GPUs; SK Telecom’s 2GW DSX AI factory is expected to deploy Vera Rubin systems equipped with SK Hynix’s HBM4.

According to SemiAnalysis, Nvidia may already have a long-term agreement with SK Hynix for SOCAMM, including locked-in preferential HBM prices and large-scale procurement commitments for 2027, although the final outcome may still differ from current estimates.

If this assessment is accurate, this wave of infrastructure building in South Korea, while strengthening Nvidia's market position, may result in actual returns for Samsung and SK Hynix shareholders that are far more complex than the headline numbers suggest.

South Korea's "Squid Game": The Story of the Sovereign AI Tournament

In June 2025, the South Korean government officially launched the "independent AI foundational model" project, aiming to develop a large language model that Korean institutions can independently train, modify, and operate without reliance on foreign AI labs.

The most distinctive aspect of this project is its competition format. The government did not pre-designate a "national team," but instead used an elimination tournament system: all participants received subsidies for compute, data, and researchers, were reviewed every six months, and the resources of the eliminated were reallocated to the winners.

A total of fifteen consortia registered, ten were selected after initial document review, and by August 2025, five teams were finalized: Naver Cloud, LG AI Research, SK Telecom, NC AI, and Upstage.

In the first round, the government provided each team with the equivalent of about 3,000 H100s, invested about $45 million in data procurement, and gave each company about $1.4 million for overseas talent recruitment.

However, unexpected events occurred at the end of the first round. NC AI, which ranked lowest, was eliminated, but Naver Cloud was also disqualified.

The South Korean government then added Motif Technologies as the fourth team. The company was spun out from Korean AI infrastructure software firm Moreh in February 2025.

Best Open-Source Model Faces Elimination, Raising Doubts

In July 2026, the four teams released their latest models, with evaluation results revealed on August 18. The data showed that the two startups, Motif and Upstage, significantly outperformed the teams backed by major conglomerates, and their models had less than half the parameters of the latter, highlighting the performance gap even more.

According to SemiAnalysis, on the Artificial Analysis intelligence index, Motif 3 led Upstage by 10 points, making it the best Korean model, even surpassing the top two American open-source models, Inkling and Nemotron 3 Ultra.

Motif is a startup with fewer than 30 employees, has raised only $17 million, and could access less than 768 B200s for computing. The total compute cost to train Motif 3 (including experimental phases) was about $15 million.

Because of this, Motif’s elimination came as a major surprise. According to the competition’s scoring rules, benchmarks account for 40%, expert review for 35%, and user testing for 25%. Although Motif ranked first in the benchmarks, it fell to the bottom in the other two aspects and was finally eliminated.

The evaluation methodology is markedly opaque. In user testing, neither tasks nor criteria were disclosed, and the tests were not blind, possibly favoring more well-known large companies.

Expert reviewers reportedly expressed concerns that Motif’s technology could leak to foreign investors—but this directly contradicts the competition’s rule that all technical components must be open source.

Additionally, some scoring dimensions, such as "ecosystem impact" and "future plans," are structurally more favorable to major companies like LG and SKT.

Now, Motif may be forced to relocate outside of South Korea to secure the capital and computing resources needed for further research—an outcome that is supremely ironic for the brightest dark horse in this national AI competition.