A global surge in demand for DDR4 memory chips is driving prices sharply higher as major DRAM manufacturers reduce legacy production and supply chain constraints push third-quarter allocations to their limits. Taiwanese firm Nanya Technology has reportedly ceased issuing price quotations after selling its entire third-quarter production, leaving buyers to await allocations in the fourth quarter amid escalating price pressures.

South Korea's top chipmakers, SK Hynix and Samsung Electronics, are heading into 2025 with dramatically diverging fortunes, as the global race for AI-driven memory dominance accelerates. Fueled by surging demand for high-bandwidth memory (HBM), SK Hynix is expected to outpace Samsung's semiconductor division by nearly KRW13 trillion in first-half 2025 operating profit, according to industry estimates.

Samsung and SK Hynix are advancing vertical channel transistor (VCT) technology through 4F² DRAM prototypes as a transitional step toward 3D DRAM, while Micron is bypassing VCT entirely to focus directly on true 3D DRAM development, signaling a strategic divergence in next-generation memory architectures.

As competition intensifies in the high-end memory market, Micron has taken the lead over South Korean competitors by becoming the first memory supplier for Nvidia's next-generation memory module, SOCAMM. Industry sources reveal that SOCAMM uses LPDDR5X chip stacking technology. With Micron expanding its outsourced packaging and testing (OSAT) orders for LPDDR5X, leading memory OSAT provider PTI is a primary beneficiary. This partnership not only helped Micron win significant contracts but also boosted PTI's second quarter 2024 shipments of LPDDR5X packaging, greatly enhancing its operational performance and future growth prospects.

SK Hynix is doubling down on diversifying its supply chain for thermal compression bonding (TCB) equipment, crucial for producing high-bandwidth memory (HBM), while Samsung Electronics appears to be moving in the opposite direction, prioritizing internalization. The contrasting strategies are fueling debate across the semiconductor industry.

Benefiting from a rapid recovery in memory market demand during the first half of 2025, memory packaging and testing companies are sharing in the gains. Although the US 90-day tariff exemption period is about to expire, increasing uncertainties around order visibility starting from the third quarter are expected to make customers adopt more conservative stocking strategies.

As Nvidia prepares to begin supplying samples of its next-generation AI accelerator "Rubin," industry sources report that SK Hynix is accelerating shipments of sixth-generation high-bandwidth memory (HBM4). Some HBM4 units intended for Rubin samples have already been supplied in small quantities.

The US expanded semiconductor restrictions to include Nvidia's H20 chip, previously the only compliant AI processor available to Chinese buyers, as the Trump administration deepens technology curbs on Beijing.

China continues to lag behind memory giants SK Hynix, Micron, and Samsung Electronics in developing high bandwidth memory (HBM) technology. Although US export controls on semiconductor tools, such as lithography systems, etching equipment, and TSV processes, present long-term barriers, they have had limited short-term impact on China's HBM research and domestic production.

Canon Machinery, a subsidiary of Japan's Canon Inc., is intensifying its development of hybrid bonding technology as it seeks to expand beyond its conventional die-bonding equipment. The company plans to introduce its advanced packaging solutions to the market after 2026, aiming to meet the increasing demand from leading semiconductor manufacturers.

The recent increase in mature process DRAM prices is expected to boost Samsung Electronics' DRAM business revenue, offsetting the sluggishness in its high bandwidth memory (HBM) business and fueling momentum for its performance in the second quarter of 2025.

AMD on June 12 officially confirmed that its next-generation AI accelerators—the Instinct MI350X and MI355X—will feature 12-high HBM3E memory supplied by Samsung and Micron. The confirmation marks a strategic win for Samsung, which has faced challenges qualifying its HBM3E for use in Nvidia's AI chips.

As data centers face increasing demands for AI training and inference workloads, high-bandwidth memory (HBM) has become a critical competitive edge for memory manufacturers. Micron announced that its 12-layer stacked 36GB HBM4 samples have been delivered to multiple major customers. Utilizing the 1β DRAM process technology, mass production is expected to begin in 2026 to support customer growth on next-generation AI platforms.

SK Hynix has unveiled a long-term roadmap for next-generation DRAM technologies, laying out a vision that aims to drive sustainable innovation over the next 30 years as the global semiconductor industry undergoes rapid transformation.

Nvidia has reportedly tapped Micron Technology as the first supplier of its next-generation small outline compression attached memory module (SOCAMM) chips, putting the US-based memory maker ahead of Samsung Electronics and SK Hynix in the race to power next-gen AI servers.

As mass production of sixth-generation HBM4 nears, South Korean chip giants Samsung Electronics and SK Hynix are aggressively reconstructing their semiconductor equipment supply chains to strengthen competitiveness, reported Korean medium IT Chosun.

Chipmaker Micron Technology Inc. said it will spend about US$200 billion on US manufacturing, research and development, the latest company to pledge large-scale investments in the country since President Donald Trump won election.

SK Hynix President and CEO Kwak Noh-jung warned that growing tariff-related uncertainty could heighten market volatility in the second half of 2025. He shared these views during the company's quarterly "THE Communication Event Together" at its Icheon headquarters in South Korea, which was also streamed live to business sites across the country, according to ETNews, Hankook Daily, and Yonhap News.

Samsung Electronics has reportedly failed its third attempt at obtaining Nvidia's certification for 12-layer HBM3E chips in June 2025, according to a recent report by a Hong Kong brokerage cited by South Korea's Businesspost. The tech giant is now targeting a fourth certification attempt in September.

Taiwan-based memory module vendors Adata Technology and Team Group Inc. saw notable revenue gains in May 2025, driven by continued price increases for DRAM and NAND flash. Adata reported NT$4.169 billion (approx. US$139.3 million) in consolidated revenue—its best May performance on record and highest monthly total since April 2010. Team Group posted NT$1.202 billion (approx. US$40.2 million), marking a 6.6% month-over-month rise and its second-best May result to date.

Memory manufacturers are reducing production capacity for older processes, driving DDR4 prices to skyrocket. Although Samsung Electronics' last buy order (LBO) for DDR4 was originally scheduled to end in June 2025, with production ceasing by the end of the year, memory module makers are already struggling to secure supply.

In a sign of China's accelerating ambitions in the semiconductor race, ChangXin Memory Technologies (CXMT), the country's leading DRAM manufacturer, is rapidly advancing development of next-generation memory technologies. Departing from its original plan to commercialize DDR5 using a 17-nanometer process, CXMT is now opting for a more advanced 16-nanometer node. A 15-nanometer process is also under development, with mass production targeted for late 2026.

SK Hynix is approaching the final delivery deadline for its May orders of thermal compression bonding (TCB) equipment from Hanmi Semiconductor and Hanwha Semitech, as the industry focus shifts to which vendor will secure contracts for the second half of 2025.

According to a press release on June 10, Micron Technology announced it has begun shipping samples of its 36GB HBM4 with 12-high stacking to multiple customers. The new memory is built on Micron's 1-beta DRAM process and features a 2048-bit interface capable of exceeding 2.0 TB/s per stack — more than a 60% performance improvement over its predecessor.