01. Passive Component Standard Products See Demand Recovery, Broadening Industry Rebound
On September 9, Economic Daily reported that Yageo's latest signals indicate simultaneous warming demand for both standard and specialty products, suggesting AI-driven passive component demand is spreading from high-end niche items to general-purpose parts. Yageo stands to benefit first, with SAE Magnetics, Holy Stone, and Nichidenbo also poised to gain.
02. HBM Crowding-Out Effect Persists, DRAM/NAND Supply Tightens; CXMT HBM Yield at Just 25%
On September 9–10, Commercial Times and China Times reported that HBM's share of DRAM wafer starts is projected to rise from 19% in 2026 to 28% in 2028, squeezing conventional DRAM supply. Excluding HBM, wafer start growth will be only mid-single digits annually, while NAND supply expansion is also constrained, making oversupply unlikely by 2028. Average HBM3e pricing is estimated to climb from $1.50 to $3.00 per Gb, with HBM4 rising from $2.20 to $4.16. Mainland China's CXMT has achieved only around 25% yield in HBM3 pilot production, primarily due to insufficient TSV expertise—front-end yield sits at roughly 30% and back-end packaging at 70%, meaning nearly 80 out of every 100 chips are scrapped. Samples have been delivered to Alibaba and Cambricon for testing.
03. TI Issues Third Price Hike of the Year, Accelerating Industrial Chip Domestic Substitution
On September 7, TI issued its third price adjustment notice of the year, raising prices across multiple products. AI computing demand is driving up analog chips such as power management ICs, with wafer capacity being prioritized for AI servers, while rising raw material costs are pushing the industry into a pricing upcycle. China's industrial chip demand is estimated at approximately 200 billion yuan in 2025, yet American firms hold over 80% share in the high-end segment, with TI long dominating the DSP market.
04. Memory Prices Surge; Kioxia Moves to Curb the Rally
On September 10, Commercial Times reported that Kioxia CEO Nobuo Otaya stated bluntly that "memory prices have risen enough," instructing sales teams not to raise prices sharply for data center operators to avoid undermining long-term AI industry investment. While Q1 NAND average prices surged 70% quarter-over-quarter, a repeat of such magnitude is unlikely.
Otaya clarified that no joint production talks are underway with SK Hynix, citing potential antitrust scrutiny and difficulties coordinating capacity at the SanDisk joint venture fab. The company will focus on high-margin AI data centers rather than competing with Chinese vendors in the low-cost consumer market, and has begun shipping its 10th-generation 332-layer 3D NAND while partnering with Nanya Technology to develop metal-oxide alternatives to silicon.
05. Qualcomm and Amazon Reach AI Chip Agreement Worth Up to $60 Billion
On September 10, Central News Agency reported that Qualcomm said Amazon could purchase up to $60 billion in AI data center chips and related products under a long-term collaboration. Qualcomm granted approximately $4 billion in warrants, allowing Amazon to purchase shares at $161.26 each as procurement milestones are met.
The two sides will jointly develop AI inference chips and high-speed optical interconnect technology, while Qualcomm expands its use of AWS for chip design. Amazon joins Microsoft and Meta in Qualcomm's data center roster, with the chipmaker expecting data center revenue to reach $15 billion by 2029.
06. Apple Raises Prices Across New and Legacy Models, Soaring Memory Costs Fuel "AI Inflation"
On September 10, Central News Agency reported that Apple's iPhone 18 Pro and Pro Max will start at $1,199 and $1,299 respectively, up $100 from their predecessors, while older models are also seeing price hikes—breaking the convention of new models upgrading while legacy devices get discounted.
Samsung's Galaxy A-series has already risen NT$500–3,000, with ASUS and Lenovo also adjusting prices. TrendForce data shows that memory's share of the iPhone Pro 256GB bill of materials has climbed from roughly 10% a year ago to 34% in Q3 2026, and is projected to exceed 40% by the first half of 2027. As AI data centers and consumer electronics compete for memory resources, costs are spilling over from infrastructure to end devices, creating an "AI inflation" effect.