Even Inside The HBM Boom, Korea's AI Position Is

Even Inside The HBM Boom, Korea's AI Position Is More Fragile Than The Headline Profits Suggest

Samsung Electronics and SK Hynix together hold roughly 80% of the global high-bandwidth memory market, and both companies are reporting record AI-driven profits in 2026. But Seoul Economic Daily has now published warnings from senior Korean technology analysts that the country's AI position is more fragile than the headline numbers suggest. The argument is that Korea has won the memory layer of the AI infrastructure stack but is losing the foundation-model layer, and that imbalance has structural consequences for the country's technology future.

What The Warning Actually Says

The Seoul Economic Daily piece, published April 26, summarises commentary from senior analysts at Samsung Securities, Mirae Asset, and the Korea Institute for Industrial Economics and Trade. The shared concern is that Korea's competitive advantage in semiconductors and AI infrastructure is real but narrow, and that the absence of a globally competitive foundation-model platform leaves the country exposed to value capture by US, Chinese, and Indian model providers who buy Korean memory.

Samsung's Q1 2026 results showed strong year-on-year growth driven by HBM and DRAM demand, and SK Hynix's Q1 2026 fivefold profit jump reaffirmed the country's chip-side dominance. But the warning is that profit pools further up the AI value chain, including model serving, agentic AI applications, and consumer AI products, are growing faster than the chip-side pool. Korea has not yet built a globally competitive participant in that upper layer.

Korea has built the strongest chip position in the AI age. The question is whether that is enough, and the honest answer is probably not.

The Market Position Numbers

The numbers underneath the warning are striking. SK Hynix held 62% HBM market share in Q2 2025, with Micron at 21% and Samsung at 17%. Samsung is investing aggressively to recover ground, including HBM4 and the next-generation HBM4E roadmap, with the Pyeongtaek P5 facility expected to be operational by 2028. SK Hynix's M15X facility is targeted for mid-2027.

Bank of America estimates the 2026 HBM market at $54.6 billion, up 58% from the previous year. Samsung is targeting roughly 50% capacity expansion in 2026, while SK Hynix has announced a more than fourfold increase in infrastructure investment versus its previous announcement. Both companies are reportedly planning a roughly 20% HBM3E price hike for 2026, ahead of the HBM4 transition.

The chip-side picture therefore looks structurally strong, with three Korean and US suppliers controlling effectively the entire global HBM market. The foundation-model picture is the gap. Korea has Naver and LG AI Research, both of which produce capable Korean-language models. Neither has emerged as a global frontier competitor in the way that DeepSeek, Sarvam, or even Mistral has positioned in their respective regions.

Why This Matters Beyond Korea

The Korean position is a useful case study for the rest of Asia. Japan, Taiwan, and increasingly Singapore have built significant chip-side AI capability, but the foundation-model and agentic AI layers remain dominated by US and Chinese players. India's Sarvam AI, Indonesia's BDx-Nvidia sovereign stack, and DeepSeek in China are the only Asian players currently positioned at frontier scale in foundation models.

That asymmetry has long-term consequences. The chip-side pool is real but capped by the underlying demand for AI compute. The foundation-model and applications pool is uncapped because it scales with how much economic activity moves to AI-mediated channels. Korea's medium-term AI economic position depends on whether Naver, LG, or a new domestic entrant can credibly compete at the upper layer, or whether the country accepts a permanent role as the highest-value supplier to other people's AI businesses.

The risk for Korea is not that it loses the chip war. The risk is that it wins the chip war and still ends up lower in the AI value chain than it should be.

What Could Change The Trajectory

Three potential developments could reset Korea's position. The first is a serious foundation-model investment by Samsung or LG that produces a globally competitive open-weights model. Both companies have the capital, the data infrastructure, and arguably the engineering depth, but neither has so far prioritised foundation-model work at the scale needed.

The second is a Korean policy push that channels public capital into a national foundation-model effort, similar to what India is doing through the AI Mission. The Korean AI Basic Act, which entered force three months ago, has provisions that could be repurposed for this kind of programme, and there are signs the National Assembly is starting to consider it more seriously.

The third is acquisition or strategic partnership with a frontier non-Korean lab. Anthropic's partnership with NEC in Japan provides a template that a Korean-Japanese-US triangulation could follow, although Korean political sensitivities about technological autonomy make this a harder path than it might appear from outside.

For deeper context, see our coverage of SK Hynix's Q1 2026 fivefold profit jump, TSMC's 58% Q1 profit jump, and Korea's AI Basic Act enforcement update.