The Iran-related conflict has severely disrupted Qatar's helium exports, driving global helium prices up approximately 50% and creating direct supply risk for semiconductor manufacturing. Air Liquide opened a new chip materials factory near Taichung, Taiwan on March 25 in response to supply diversification concerns. Approximately 200 specialized helium containers remain stranded near the Strait of Hormuz.
Helium is critical to chipmaking. ASML's EUV lithography machines, which produce every leading-edge chip below 7nm, rely on helium as a coolant due to its unmatched thermal conductivity among inert gases. Beyond lithography, helium creates inert atmospheres in deposition and etching chambers and serves as the industry standard for vacuum leak detection in fab tools. Helium-filled hard drives used in data centers also depend on steady supply.
Qatar accounts for roughly 30% of global helium production. Industry analysts estimate it could take up to five years to rebuild Qatar's helium export infrastructure, meaning even a ceasefire would not quickly restore pre-conflict supply levels. SEMI officials stated Taiwan's overall helium supply remains secure through alternative sources including the US and Canada, but cautioned that prolonged disruption would tighten an already constrained market. Air Liquide's decision to build dedicated helium infrastructure in Taiwan signals major industrial gas suppliers are preparing for a structural shift in supply routes.
Current supply is adequate through alternative sourcing, and fabs maintain buffer inventories. However, if the conflict persists and alternative supply cannot scale fast enough, helium joins the short list of single-point-of-failure inputs, alongside EUV photoresist and neon gas, that could constrain leading-edge wafer output. Unlike most fab materials, there is no synthetic substitute for helium.
Linked stocks: 2330 TT, ASML NA, 005930 KS, INTC US
TrendForce issued its latest memory pricing survey projecting Q2 2026 contract prices for general DRAM will rise 58-63% sequentially, while NAND Flash will surge 70-75%, driven by persistent supply shortages.
The Taiwanese research firm attributes the aggressive pricing to DRAM manufacturers aggressively shifting capacity toward high-bandwidth memory (HBM) and server applications, while simultaneously implementing a "catch-up pricing" strategy to narrow the price gap across different product categories.
PC DRAM is seeing upward price pressure despite downward revisions to PC shipments. Original equipment manufacturers facing lower fulfillment rates from suppliers are forced to pay premiums to secure inventory. For server DRAM, North American cloud service providers are clarifying their AI inference application scenarios, driving demand for both AI servers and general-purpose servers with large-capacity RDIMMs as the primary procurement target. Server DRAM generates the highest profit margins across all DRAM product categories, prompting manufacturers to prioritize bit output allocation to this segment.
In mobile DRAM, smartphone brands face mounting memory cost pressures and may adjust production plans starting in Q2 2026. Graphics DRAM faces reduced demand for notebooks and gaming consoles due to higher memory prices, with capacity allocation to GDDR remaining insufficient. Consumer DRAM customers focused on low-margin, low-price products now face memory costs exceeding selling prices in some cases.
While NAND manufacturers are increasing bit output through process upgrades and raising QLC ratios, the increase remains limited. Demand from AI servers remains robust, while PC and smartphone manufacturers are forced to reduce product capacity to suppress NAND Flash demand volumes. Enterprise SSD demand growth shows no signs of slowing as generative AI enters large-scale application, with significant shortages estimated this year and new capacity not expected to ramp until late 2027 or 2028.
Linked stocks: MU US, 005930 KS, SSNLF US
Nvidia Corp announced a $2 billion equity investment in Marvell Technology Inc and is opening its platform to allow Marvell to integrate custom AI chips and networking equipment, marking a significant expansion of Nvidia's AI infrastructure partnerships.
The investment and partnership announcement sent Marvell shares surging, as the deal positions Marvell as a key supplier of custom silicon and networking solutions for Nvidia's data center customers. The partnership centers on Marvell's custom chip design capabilities and networking technology, which will now be more tightly integrated with Nvidia's GPU-based AI systems.
This move allows Nvidia to offer customers more flexibility in designing AI infrastructure while maintaining its position at the center of the AI hardware ecosystem. Industry analysts view the deal as Nvidia's response to growing customer demand for custom silicon solutions alongside standardized GPU offerings, particularly as hyperscalers and cloud service providers seek to optimize specific workloads.
For Marvell, the partnership provides both capital and a strategic endorsement that could accelerate customer adoption of its custom AI accelerators and high-speed networking products.
Linked stocks: NVDA US, MRVL US
Taiwan Semiconductor Manufacturing Company (2330 TT) held a community meeting near its Dresden, Germany fabrication site to update local residents on construction progress. ESMC president Christian Koitzsch stated that the project remains on schedule since groundbreaking in August 2024, with initial production equipment expected to enter the cleanroom in the near term.
Koitzsch addressed water resource management concerns by stating that the facility will install a rainwater detention pond with approximately 3,000 cubic meters capacity. The equity structure shows TSMC holding 70% of ESMC, with European partners Bosch, Infineon, and NXP each holding 10%. Koitzsch emphasized that the company faces global competitive pressures and high costs requiring careful resource allocation. The facility will create approximately 2,000 high-paying jobs.
As construction enters peak phase, the site currently hosts approximately 1,000 workers daily, potentially increasing to 5,000. The company plans to build parking for 800 vehicles. Koitzsch noted that tram lines near the fab could cause vibration and electromagnetic interference affecting precision manufacturing processes, requiring further evaluation.
Linked stocks: TSM US, 2330 TT
South Korean AI chip startup Rebellions raised $400 million in pre-IPO funding backed by Samsung Electronics, the Korean government, and institutional investors, bringing total funding to 640 billion won as the company positions its custom inference accelerators against Nvidia's dominance in data center AI.
Rebellions develops ATOM, a purpose-built AI inference processor designed to run large language models and generative AI workloads at lower cost and power consumption than Nvidia's general-purpose GPUs. The inference market is the fast-growing segment of AI compute: while training requires massive GPU clusters, inference accounts for the majority of ongoing AI infrastructure spending and is where custom silicon can offer the largest efficiency gains over GPUs.
The Korean government's participation underscores the geopolitical dimension of AI chip competition. South Korea's semiconductor industry is dominated by memory (Samsung, SK hynix) but has limited presence in logic and AI accelerators. Rebellions represents Seoul's bet on building that capability domestically.
Linked stocks: 005930 KS, NVDA US