Below are the most-read DIGITIMES stories from the week of Mar 9-15, 2026:
NAND prices surge as AI storage demand tightens supply
NAND flash prices have surged sharply as supply constraints intensify, with some manufacturers raising quotations by as much as 50% overnight, according to Phison Electronics. The Taiwanese controller supplier said demand from AI infrastructure is rapidly expanding enterprise storage requirements, pushing enterprise SSD products to about 30% of its revenue in the first quarter of 2026.
Phison has responded by securing long-term agreements with six NAND manufacturers and two DRAM suppliers, and by building inventory exceeding NT$50 billion (approx. US$1.6 billion) to ensure supply stability. The company is increasingly shifting toward enterprise design services tied to hyperscale cloud providers and AI server makers, while also advancing next-generation controller technologies such as PCIe Gen 6 and Gen 7 to support future AI system requirements.
Nvidia narrows CoWoP PCB partners
Nvidia has reduced the number of potential PCB suppliers for its emerging Chip-on-Wafer-on-PCB (CoWoP) advanced packaging technology to three companies: Taiwan's Zhen Ding Technology, Unimicron, and China's Kinwong. The technology, which aims to replace traditional ABF substrates with PCBs, is still in early verification but has attracted significant industry attention as an alternative packaging approach. However, the technical barriers, including the integration of high-end materials and extremely precise routing requirements, have led several manufacturers to withdraw from the program.
With no confirmed mass-production timeline and the continued dominance of CoWoS packaging in the near term, many PCB vendors are prioritizing other markets, such as networking servers and satellite systems, where investment returns are more certain.
Shifting AI chip production away from China
Nvidia has halted production of China-bound H200 AI chips and redirected manufacturing capacity at TSMC toward its next-generation Vera Rubin architecture. The decision reflects ongoing regulatory uncertainty as US export restrictions and potential Chinese countermeasures complicate sales prospects. Although Nvidia had prepared to ship more than one million H200 chips to Chinese customers, approval delays and licensing limits have prevented meaningful shipments.
By reallocating production resources to Vera Rubin, which is expected to power next-generation AI systems from major US technology companies, the company is prioritizing markets with clearer demand visibility while reshaping supply chain planning for AI accelerators.
Winbond aligns DDR3 pricing with DDR4
Memory supplier Winbond has raised prices for its 4Gb DDR3 chips to match those of equivalent DDR4 products, reflecting severe supply shortages across the DRAM market. The company is simultaneously accelerating its shift toward DDR4, which now accounts for 60-70% of output, while maintaining limited DDR3 production for legacy applications such as IoT devices and networking equipment.
Capacity across Winbond's facilities remains fully booked throughout the next two years, with further price increases expected as structural supply constraints persist. The company is also repurposing manufacturing capacity to increase NOR flash production to meet growing demand in automotive, industrial, and communications sectors.
Samsung secures DRAM deal for Apple's foldable iPhone
Samsung has reportedly won the contract to supply 12GB LPDDR5X memory for Apple's first foldable iPhone, scheduled for release in the second half of 2026. The agreement reflects tightening DRAM supply conditions, with LPDDR5X prices reportedly doubling compared to 2025 levels. Although Apple continues to diversify suppliers, Samsung's production scale is expected to make it the primary memory provider.
At the same time, Samsung Display (SDC) is also positioned as the sole supplier of foldable panels for the device. Initial shipments of the foldable iPhone are expected to remain relatively modest, but the deal reinforces Samsung's leadership in both memory and foldable display technologies.
Applied Materials launches US$5 billion R&D hub
Applied Materials has partnered with Micron Technology and SK Hynix to establish the EPIC Center, a collaborative semiconductor research facility in Silicon Valley with planned investments of up to US$5 billion. The center will focus on accelerating the development of next-generation memory technologies, including DRAM, high-bandwidth memory, and NAND, which are increasingly critical for AI and high-performance computing systems.
Engineers from participating companies will work directly with equipment developers to shorten commercialization timelines for new technologies. The initiative reflects growing industry pressure to improve memory performance and energy efficiency as AI infrastructure spending surges worldwide.
India's semiconductor expansion as part of long-term strategy
India's plan to bring four semiconductor facilities into commercial production in 2026 marks a milestone for the country's industrial ambitions. Still, it is unlikely to alter global wafer supply in the near term significantly. Analysts note that most of the new facilities are assembly, test, marking, and packaging plants rather than wafer fabs, indicating their primary role will be to strengthen downstream semiconductor capabilities.
Even with India's flagship Tata Electronics fab expected to produce around 50,000 wafers per month, the country's total capacity will represent only a small fraction of global output. Industry experts, therefore, view the initiative as a long-term ecosystem-building effort rather than a development capable of shifting global semiconductor supply dynamics in the immediate future.
Article edited by Jack Wu
