According to a report released by McKinsey on July 20, the global semiconductor industry is increasingly driven by a "winner-takes-all" dynamic. In 2023, the top 5% of companies by annual sales—including NVIDIA, TSMC, SK Hynix, and Broadcom—captured a staggering $159 billion in economic profit. In contrast, the middle 90% of companies shared just $5 billion, while the bottom 5% collectively lost $37 billion. Remarkably, the profit earned by the top tier even exceeded the industry's total economic profit of $147 billion.

This sharp divergence emerged within just two to three years. During the COVID-19 pandemic (2021–2022), the middle 90% of firms earned over $30 billion annually, averaging $130 million per company. But by 2023, amid a boom in AI semiconductors, average profit dropped to $38 million and plunged further to $17 million in 2024—an 88% decline in two years.

McKinsey projects that AI-focused semiconductor companies will grow at a compound annual rate of 18–29% through 2030, while growth for traditional chipmakers is expected to remain at just 2–3%. The core reason for this growing divide is that leading companies are setting new standards for emerging chip products, effectively controlling the rules of next-generation technologies.

While traditional components follow industry-wide standards set by JEDEC, newly introduced products allow early entrants to define proprietary standards. For instance, when SK Hynix developed its first-generation HBM in 2013, it simultaneously established the standard. Today, NVIDIA is doing the same with its proprietary SoC-attached memory module (SOCAMM), designed to power personal AI supercomputers. As chip design shifts toward customization, such practices are likely to become the norm, consolidating market power among a few players.

This shift has left South Korea's semiconductor sector at risk of marginalization. Although South Korea controls over 50% of the global memory market, it lags significantly behind the U.S. and Taiwan in AI-critical chips like GPUs and ASICs. Aside from SK Hynix's HBM, few Korean companies are part of NVIDIA's AI supply chain. Domestic AI chip startups exist, but lack the capital and talent to compete beyond niche segments where NVIDIA has yet to enter.

To counter this trend, Korean industry experts are calling for a strategic pivot to next-generation memory products such as second- and third-generation HBM. Emerging technologies like CXL, PIM, and LPCAMM—offering low-power and high-speed performance—are seen as opportunities. Samsung is reportedly advancing faster than SK Hynix in CXL product development, potentially shifting the competitive landscape.

Experts also stress that Korea must adopt a broader approach to ecosystem and funding development. This includes moving beyond tax incentives toward direct subsidies or equity investment, and leveraging Taiwan's public-private model as a blueprint. Attracting foreign R&D centers to Korea could also enrich the domestic AI semiconductor ecosystem.

Lee Eun-young, Managing Director at Samil PwC's Management Research Institute, stated that Korea lacks sufficient domestic capability for critical AI chip components. She emphasized the need for stronger investment in R&D, talent, and capital to break out of a self-reinforcing cycle of underperformance.

As the AI era redefines semiconductor market dynamics, success will increasingly depend on setting the rules—not just playing by them.