Samsung, SK Hynix Capital Expenditures Surge 35% in the First Half of the Year! Nvidia Not Among Samsung's Top Five Customers

Driven by artificial intelligence demand, Samsung Electronics and SK Hynix are aggressively expanding their production capacity, with the two companies’ combined semiconductor facility investments for the first half of 2026 growing by more than 30% year-on-year.

On August 17, according to Korean media outlet Etoday citing the companies’ semi-annual reports for 2026, the combined semiconductor facility investments by Samsung Electronics and SK Hynix in the first half of the year reached 43.198 trillion KRW, up 35.1% from 31.98 trillion KRW in the same period last year. Both companies reported capacity utilization rates at 100%, highlighting the extremely strong demand for AI-related memory.

In terms of customer structure, Nvidia’s share of SK Hynix’s revenue fell from 24% for the full year of 2025 to 13.35%, while Nvidia did not even make the list of Samsung’s top five customers during the same period. This reflects the differences in AI business strategies between the two companies.

Soaring AI Demand: Two Memory Giants Accelerate Capacity Expansion

Looking at the investment pace, SK Hynix has been even more aggressive in its expansion.

The DS division of Samsung Electronics saw a year-on-year increase of 23.5% in facility investment in the first half of the year, primarily focusing on next-generation memory, advanced process capacity, and related infrastructure. In comparison, SK Hynix’s facility investments soared 56.4% year-on-year, with emphasis on meeting the demand for HBM, high-capacity DRAM, and enterprise SSDs.

Both companies have reached 100% capacity utilization, which means their existing production capabilities are virtually operating at full load. Against the backdrop of continued AI infrastructure buildout, demand for high-end memory products such as HBM is rising rapidly, prompting Samsung and SK Hynix to expand production capacity ahead of the next generation of products.

In other words, this round of capital expenditure is not simply to catch up with current demand, but is a race for capacity and technological edge in the next wave of AI memory growth.

R&D Investment Hits Record Highs, SK Hynix Nearly Doubles Growth Rate

Beyond capital expenditure, both companies are also ramping up their R&D investments.

In the first half of the year, Samsung Electronics’ total R&D spending hit 27.363 trillion KRW, a 51.5% year-on-year increase, marking a record high for a half year. However, Samsung did not separately disclose the R&D spending for its DS division. SK Hynix’s R&D investment growth was even more pronounced, reaching 6.043 trillion KRW in the first half, up 98.4% year-on-year, which is already about 90% of its full-year 2025 R&D expenditure.

The competition among memory manufacturers is extending beyond traditional process and capacity competition, further evolving into the AI computing ecosystem.

According to Korean media reports, Micron recently launched the $250 million Micron Ventures Paradigm Fund, the company’s largest-ever venture fund, focusing on AI model architectures, computing, enterprise applications, and physical AI. The aim is to proactively prepare for emerging AI technology trends and establish a partner network relevant to next-generation memory and computing products.

AI Memory Demand Shifting from “Relying on Nvidia” to “Increasingly Diversified”

The semi-annual reports show that changes in customer structure are one of the most significant highlights of this period.

According to Seoul Economic Daily, SK Hynix’s revenue from Nvidia in the first half of 2026 reached 17.6087 trillion KRW, accounting for 13.35% of total revenue, lower than the 24% share for the whole year of 2025.

This shift does not signal a decrease in Nvidia’s importance. On the contrary, SK Hynix is expanding supply to large technology companies developing their own AI chips, thus making its customer structure more diversified. As SK Hynix works with Nvidia to advance HBM4 supply for the Vera Rubin platform, Nvidia’s contribution to its revenue could rebound in the second half of the year.

Samsung’s customer structure is even more concentrated. According to its semi-annual report, the top five customers account for about 25% of total revenue, and Nvidia is not among them.

This means that, amid the explosive growth in AI chip demand, the relationships between memory manufacturers and their customers are shifting: Nvidia remains one of the most important clients in the HBM market, but as tech giants like Alphabet and Amazon accelerate their own ASIC development, the sources of AI memory demand are becoming more dispersed.

For Samsung and SK Hynix, the competition ahead is no longer just about “who can produce more HBM,” but also who can more rapidly enter the expanding AI chip customer base and secure more key clients for next-generation products such as HBM4.