Concerns are mounting that South Korea is being sidelined in the global AI semiconductor battle. Customized semiconductors, rather than general-purpose chips, are essential for the proper implementation of AI services designed by various companies, but Korean firms are struggling to make their mark. U.S.-based companies such as Nvidia, Apple, and AMD dominate the critical semiconductor chip design sector, operating as fabless (semiconductor design-focused) companies. Taiwan’s TSMC holds tight control over the semiconductor manufacturing (foundry) and packaging required by these designs.
Experts believe that the solid partnership between the U.S. Big Tech and Taiwan’s TSMC leaves little room for Korean companies to enter. This is exacerbated by Nvidia and AMD’s recent moves to elevate their alliance by establishing AI R&D centers in Taiwan. The U.S. and Taiwan dominate the non-memory market, which was valued at 620 trillion won (US$452 billion) in 2023, over 3.5 times larger than Korea’s memory market at 179 trillion won. Predictions from the American Semiconductor Association suggest that Korea’s market share for advanced semiconductors smaller than 10 nanometers, which was 31% in 2022, will shrink to 9% by 2032.
The atmosphere is thickening with South Korea increasingly being left out among the U.S.’s Chip4 alliance countries, including Japan and Taiwan. As the AI era advances, there is a growing trend to bypass Korea in the crucial areas of customized semiconductor design, manufacturing, and post-processing.
The emergence of Nvidia’s “AI accelerators” (semiconductors optimized for AI services) gives an insight into the current industry dynamics. Nvidia-designed GPUs are manufactured by Taiwan’s TSMC, which also handles the advanced packaging that integrates these chips with high-performance DRAM such as High Bandwidth Memory (HBM) to function as a single chip. The materials and equipment needed for this process are mostly supplied by Japanese companies.
The only role left for Korea is the HBM provided by SK hynix. Even here, variables have emerged due to Micron’s new entry, having secured a supply deal with Nvidia by promoting their products’ 30% lower power consumption. Nvidia’s upcoming Blackwell AI accelerators, slated for release in the latter half of this year, will feature Micron’s 6th generation HBM, HBM3E, alongside SK hynix’s products.
The triangular alliance between the U.S., Taiwan, and Japan, comprising “number one companies” in each sector, poses a formidable threat. It’s increasingly difficult to penetrate this alliance, which seems to be strengthening over time. This was evident at Computex 2024 in Taipei, where CEOs from U.S. semiconductor firms like Nvidia, AMD, Qualcomm, and Intel gathered on June 4. During the keynote on June 3, CEO Lisa Su emphasized the “very solid alliance with TSMC,” underscoring a strong partnership spirit.
Samsung Electronics has initiated an AI semiconductor “turnkey service,” managing to encompass design, production, and advanced packaging. However, there are no reports of significant orders yet. A semiconductor expert professor commented, “The turnkey strategy maximizes effectiveness when a company has overwhelming competitive power in all sectors. Samsung, not being number one in design, foundry, or packaging, faces challenges in attracting customers.”
For Korean companies to survive in the AI semiconductor competition, there’s a growing call to enhance “customer-tailored” sales strategies. This is driven by major semiconductor clients like Microsoft, Google, and Amazon, who demand chips that are optimized for their specialized AI services. An industry insider noted, “The era of developing good performance general products first and then lining up customers is over. To succeed in the AI semiconductor market, we must move beyond the illusion of being the top in memory semiconductors.”
