[Core Information] AI-Related Applications Drive Revenue Growth For Two Major Distributors In September

1. Distributor performance: driven by AI, Wenye and Dalian set records

1. Dalianda: Revenue in September was NT$89.09 billion, a monthly increase of 13%

• September: Revenue reached NT$89.09 billion (approximately 20.89 billion yuan), a monthly increase of 13%, the second highest in the same period in history.
• Q3: Revenue was NT$244.47 billion (approximately 58.59 billion yuan), a record high in a single quarter excluding exchange rate effects.
• Cumulative (January-September): NT$743.75 billion (approximately 172.52 billion yuan), +14.6% year-on-year.
• growth core: Driven by generative AIAI server, power supply, PC/NB, storage iterationDemand surged.

2. Wenye: Revenue in September was NT$134.9 billion, +59.7% year-on-year, a record high in a single month

• September: Self-consolidated revenue was NT$134.9 billion, +59.7% year-on-year and +34.8% month-on-month, setting a single-month record.
• Q3: Revenue was NT$328.9 billion, +25.9% year-on-year and +26.8% month-on-month, exceeding the financial forecast upper limit (NT$283.5-299.5 billion).
• Cumulative (January-September): NT$835.9 billion, +19.8% year-on-year.
• growth core：Strong shipments of AI-related products+ Smartphone peak season + European and American markets recover.

3. TSMC: Revenue in September was NT$330.98 billion, +31.4% year-on-year

• September: Sales volume was NT$330.98 billion, +31.4% year-on-year.
• Q3: Revenue was NT$989.92 billion, +30% year-on-year, exceeding market expectations.
• Cumulative (January-September): NT$2.76 trillion, +36.4% year-on-year.
• growth core：Demand for AI chips surges, advanced processes (3nm/5nm/7nm) account for over 74%.

2. Storage industry: NAND shortage warning for ten years, AI reconstructs supply and demand

1. Phison CEO: NAND flash memory will be in short supply for 10 years

• core judgment: There will be a serious shortage of NAND flash memory starting in 2026,Lasts up to 10 years, a once-in-a-lifetime super cycle.
• demand side: The demand for AI inference/training has exploded, and the proportion of SSD in the data center has jumped from 20% to 80%–100%; the flash memory configuration of a single AI server has increased from 64TB to 96–120TB.
• supply side: Samsung, SK Hynix, and Micron diverted large amounts of capital toHBM, NAND production expansion is cautious; DDR4 production lines are dismantled and switched to DDR5, exacerbating structural shortages.
• price impact: NAND prices have doubled in half a year, with 1TB TLC rising from US$4.8 to US$10.7, and original manufacturers have suspended quotations many times.

2. Industry chain dynamics: Domestic production is accelerating, and storage price increases are transmitted to the entire chain

• Dike shares: Acquired 62.5% equity of Jiangsu Jingkai for RMB 300 million to develop advanced DRAM packaging and testing.
• Shengbang Shares: Acquired 77.54% equity of Yicunxin and entered into non-volatile storage and analog chips.
• Rockchip: The shortage of DDR4 drives customers to transform to DDR5, which affects Q3 growth in the short term, but continues to grow at a high rate in the medium and long term.

3. AI computing power infrastructure: giant acceleration slots, OpenAI dual-path layout

1. AMD×OpenAI: 6 GW GPU long-term supply agreement

• Signed multi-year, multi-generation product agreement based onAMD Instinct GPUPowering OpenAI’s next-generation AI infrastructure.
• First 1 GWMI450 GPUDeployed in the second half of 2026; collaboration extends to future generations such as MI350X.
• Strategic significance: OpenAI reduces its dependence on NVIDIA, and AMD enters the top AI computing power supply chain.

2. Broadcom ×OpenAI: 10 GW custom AI accelerator + 800G NIC

• joint development10 GW of custom AI accelerator, using BroadcomEthernet + CPOThe full-stack solution will be deployed in the second half of 2026.
• launchThor Ultra (800G AI Ethernet Network Card), the industry’s first 800G NIC, can interconnect hundreds of thousands of XPUs and support trillion-parameter models.
• launchTomahawk 6-Davisson, 102.4Tbps CPO switching chip, building an open Ethernet AI network ecosystem.

3. Microchip: The world’s first 3nm PCIe Gen6 switch

• support160 channels, 64GT/s/channel, providing a CPU/GPU/storage high-speed interconnection base for AI clusters.

4. Terminals and chips: AI penetrates PCs/robots, and process iterations accelerate

1. Intel: 18A process Panther Lake PC chip

• For high-end AI notebooks, the GPU/CPU performance is 50% higher than Lunar Lake**.
• Mass production will ramp up at the end of 2025 and will be fully launched in January 2026 to compete for PC market share.

2. Meixinsheng: Robot/AI end-side chips are shipped in batches

• DToF chips are mass-produced, entering well-known robot brands; entering AR/AI glasses customers, it is expected to increase sales.

3. Smartphones: Q3 +3% year-on-year, storage costs squeeze profits

• Omdia: Q3 global shipments in 2025 will increase by 3% year-on-year; Samsung (19%), Apple (18%), and Xiaomi (14%) lead the list.
• Hidden worries: Storage price increases have led to rising BOM costs, putting pressure on profitability; manufacturers compete for AI/data center production capacity, exacerbating the shortage of terminal storage.

5. Core conclusions and trends

1. AI becomes the number one engine for semiconductor growth: The performance of distributors (Wen Ye/Dalianda) and foundries (TSMC) has been fully verified, and the demand for AI servers, storage, and network chips continues to explode.
2. Storage super cycle established: Ten years of shortage of NAND, structural shortage of DDR4/DDR5, prices continue to rise, and domestic substitution is accelerating.
3. Computing infrastructure reconstruction: Adopted by OpenAIAMD GPU + Broadcom NetworkDual paths challenge NVIDIA’s dominance; PCIe 6.0, CPO and other technologies accelerate the implementation of AI clusters.
4. Supply chain security first: Original factory production capacity is concentrated in high-value areas, and distributors need to strengthenInventory management, long-term agreements, alternative material plans, balancing short-term profits and long-term risks.