On January 21, the Economic Daily reported that Samsung and SK Hynix are significantly reducing their NAND Flash production to focus on high-end DRAM and HBM production. This has led to a continued supply shortage in the global NAND chip market, causing prices to soar once again. Market estimates suggest that the global NAND chip supply gap will widen by 3% to 4%. Taiwan’s Phison and ASolid are expected to be the two major beneficiaries, while module manufacturers like ADATA, Transcend, and Team Group will also benefit.
According to South Korean media Chosun Biz, citing data from Omdia, Samsung’s NAND chip production this year will decrease from 4.9 million units last year to 4.68 million units, a reduction of about 4.5%. SK Hynix’s production will drop from 1.9 million units last year to 1.7 million units, a decrease of 10.5%. Combined, the two companies’ production will fall from 6.8 million units last year to about 6.38 million units, a year-on-year reduction of approximately 6.2%.
Industry analysts note that NAND chip and DRAM production capacities can be converted. Under the premise of maximizing profits, Samsung and SK Hynix are shifting their “relatively less profitable” NAND chip production capacity to high-end DRAM production. Meanwhile, with the expansion of AI inference applications, data centers’ demand for high-capacity SSDs and storage resources is increasing, making NAND chips an integral part of AI infrastructure and driving prices upward.
On January 22, the Economic Daily reported that the increasing power consumption of AI servers and the evolution of data center computing platforms toward high current, high voltage, and high power density have driven the demand for power management and system stability. This has made chip resistors a “strategic material” in the AI era, bringing new growth opportunities for Taiwan-based companies like Yageo and Walsin.
Chip resistors primarily function to control and limit current and voltage in circuits and are widely used in mobile phones, computers, and communication devices. In the era of AI and high-performance computing, their importance has been further amplified, with usage increasing by 1–2 times compared to traditional servers. These are mostly high-power, high-precision, and high-reliability products with higher average selling prices (ASPs). Analysts believe that Taiwan-based companies, leveraging their geopolitical advantages, will be prioritized in the AI and data center supply chains.
On January 19, the Commercial Times reported that Yageo, the global leader in chip resistors, has initiated its first price hike in five years. Related company Kaimei first notified Chinese agents of a 15% increase in thick-film resistor prices, followed by Uniohm, which announced that due to rising material costs, it will increase prices for 0402–1206 size chip resistors.
Yageo accounts for over 30% of the global chip resistor production capacity and has absolute pricing power, with a monthly production capacity of over 120 billion units. The usage of chip resistors in AI servers is 1–2 times that of general servers. U.S.-based CSPs, citing national security reasons, tend to avoid using mainland Chinese chip resistors. The geopolitical impact on supply and demand makes Yageo’s price hike particularly noteworthy.
On January 22, SemiMedia reported that SK Hynix has completed the process upgrade of its Wuxi DRAM factory in China, transitioning from the 1z process to the more advanced 1a process. The Wuxi factory is a key production hub, accounting for 30% to 40% of the company’s overall DRAM output. The factory’s 12-inch wafer production capacity is approximately 180,000 to 190,000 wafers per month, with about 90% of production now using the 1a process. The upgrade was first announced in January 2024 and took about two years to complete.
On January 20, the United Daily News reported that TrendForce released its latest research report. North American cloud service providers are continuing to strengthen their investments in AI infrastructure, with global AI server shipments expected to increase by over 28% annually in 2026. The development of AI inference services will drive a replacement and expansion cycle for general-purpose servers, with global server shipments (including AI servers) expected to grow by 12.8% annually in 2026. The combined capital expenditure of the top five North American cloud service providers is expected to increase by 40% annually in 2026, with Google and Microsoft actively increasing their purchases of general-purpose servers.
In terms of AI chips, GPUs are still the largest category, accounting for 69.7%, but ASIC AI server shipments are expected to rise to 27.8%, with growth rates surpassing those of GPU AI servers. Google is expected to become the market leader in ASICs.
On January 21, the Commercial Times reported that TSMC is accelerating the optimization of its mature process production lines to prepare for the AI era and free up manpower. Some chip makers have already discussed order transfers with foundries, with some DDIC and PMIC products needing to be transferred to other factories. In the short term, this will help improve the supply-demand structure and alleviate the low-cost impact from mainland Chinese capacities.
Since 2025, TSMC has strategically reduced its 8-inch capacity and plans to shut down some factories by 2027. Although TSMC remains competitive in high-voltage and special processes, some orders have already been transferred to other foundries like World Advanced.
Semiconductor industry analysts note that older factories have low automation levels and room for improved production efficiency. While AI and advanced chip orders are still concentrated at TSMC, some mature process orders are shifting to other foundries, such as UMC. In 2025, UMC established an 8-inch cooperation with U.S. foundry Polar, and its 12-nanometer collaboration with Intel is expected to provide the first PDK to customers in June 2026, potentially attracting some transferred orders.
