01. Standard Passive Component Demand Warms, Broadening Industry Recovery
On September 15, Omdia data showed that global semiconductor revenue reached $425 billion in Q2 2026, surging 31.4% quarter-over-quarter and surpassing Q1's 29.2% to hit a new all-time high, bringing the first-half total to $752 billion. Memory served as the biggest growth driver, with DRAM, NAND, and NOR all strengthening in tandem, while demand for high-performance products such as HBM continued climbing alongside AI server deployments. Memory ICs now account for over half of total semiconductor revenue.
The non-memory segment also posted quarter-over-quarter growth exceeding 10% this quarter, far outpacing the historical seasonal average of roughly 3%. Omdia noted that of 97 quarters tracked since 2002, only 10 have seen double-digit sequential growth, yet the industry has now achieved four consecutive quarters of double-digit expansion since Q3 2025. The trend is expected to continue in Q3 on the back of robust AI computing and memory demand.
02. High-Capacity NOR Flash Prices Expected to Double in Second Half
On September 17, Commercial Times reported that TrendForce pointed to AI demand crowding out capacity, leaving supply of 256Mb and above high-capacity NOR Flash tight, with average prices projected to surge 90%–110% in the second half of 2026. First-half NOR contract prices had already accumulated gains of 100%–120%, with second-half performance diverging by capacity: high-capacity segments are set to rise 90%–110% on strong AI server and automotive demand, while 128Mb and below will see Q4 gains moderate to 10%–20% as mainland Chinese fabs release capacity.
AI servers use three to five times more NOR per unit than traditional servers, and retimers also require standalone NOR, driving demand for high-capacity and Octal SPI NOR. Winbond Electronics holds the top market share, ramping 45nm NOR production while migrating processes toward 25/20nm, though new capacity will do little to close the gap in the near term.
03. Samsung Looks to Expand Memory Module Outsourcing; Transcend, Apacer Favored for Orders
On September 17, Economic Daily reported that Samsung, seeking to concentrate resources on HBM, is planning to outsource production of additional DDR5 module and SSD capacity, prioritizing HBM advanced packaging at its Cheonan and Onyang facilities while pressing partner firms to expand ahead of schedule.
Taiwanese manufacturers Transcend Information and Apacer Technology are the leading contenders for the business. Transcend has collaborated with Samsung for over a decade and possesses SMT lines and module manufacturing capabilities; Apacer counts Samsung as a primary raw material supplier. Analysts believe that if Taiwanese firms secure these orders, they will not only boost revenue but also deepen partnerships with original manufacturers, secure material sources, and strengthen customer stickiness amid supply shortages.
04. Infineon Sells NOR Flash Business; Winbond Takes Over for $1.12 Billion
On September 17, SemiMedia reported that Infineon has signed a definitive agreement with Winbond Electronics to sell its NOR Flash and F-RAM business for $1.12 billion in cash, with the transaction expected to close in the second half of 2027 pending regulatory approval. Winbond will acquire 100% of the business, headquartered in Silicon Valley and to be renamed Spansion, to expand into automotive, industrial, and infrastructure memory markets. Infineon will retain specialty memory products such as SRAM and HYPERRAM, focusing on core growth areas going forward.
05. AI Agents Drive Surge in CPU Demand
On September 17, Commercial Times reported that Intel CEO Lip-Bu Tan said the rapid proliferation of AI agents is fueling a surge in CPU demand, with the company currently able to satisfy only about 50% of demand from leading-edge customers. CEOs of multiple tech giants have personally called requesting increased supply, but Intel has had to apologize due to capacity constraints.
As AI shifts from training to inference, CPUs are playing an increasingly critical role in orchestration, control, and multi-thread scheduling. On the process front, 18A has entered full-scale production, with 14A slated to begin production in Q1 next year and yield improving by roughly 7% annually. Separately, Reuters reported on the 16th that SK Hynix is in talks with Intel to produce memory in the United States, potentially leasing Intel's Ohio capacity or forming a joint venture to lock in AI memory supply.
06. Apple Reportedly Returns to Server Market; Foxconn Poised for Major Orders
On September 18, Economic Daily reported that foreign media indicate Apple plans to launch AI inference servers, marking a return to the server market roughly 15 years after discontinuing Xserve. The products may adopt Nvidia's NVLink Fusion interconnect technology and feature Apple's self-developed M8 Ultra chip, with an expected launch in 2029. Analysts are bullish on Foxconn as Apple's most important assembly partner and a primary ODM for NVLink switches, positioning it to benefit from major orders. Foxconn's Houston facility is already assembling Apple AI servers at a rate of roughly 10 units per hour.