AI drivenEnterprise-level QLC demand explodeswithHDD out of stock transfer order, completely reversing NAND flash Q4 price expectations, with contract prices expected to rise across the board by 5%-10%.At the same time, industry mergers and acquisitions, SiC ecology, AI chips and passive components and other dynamic and intensive implementation of multiple lines, the semiconductor industry chain is accelerating towards theHigh computing power, high storage, and high integrationOrientation reconstruction.
1. NAND flash memory: Prices increased by 5%-10% in Q4, and QLC became the core driving force
1. Price Reversal: From Consolidation to General Increase
- Original expectation: consumer demand is overdrawn in the first half of the year, and prices will enter consolidation in Q4.
- Reality reversal:HDD is seriously out of stock + delivery time is extended, CSP (cloud service provider) quickly shifts storage needs toQLC Enterprise SSD, urgent orders poured in; SanDisk took the lead to rise by 10%, Micron suspended quotations, and the atmosphere on the supply side became stronger.
- Institutional forecast: TrendForce confirmed by TrendForce,Q4 NAND flash memory contract prices will increase by 5%-10% across the board in 2025。
2. Core driver: QLC demand explodes, AI reconstructs storage landscape
- demand side: AI server SSD configuration upgraded from 64TB to 96TB/120TB,QLC becomes the first choice due to high density and low cost; HDD proportion dropped from 90% to 80%, SSD proportion quickly increased to 20% and headed towards 80%-100%.
- supply side: Supply has tightened after original factories cut production and destocked, and production capacity has declined.Large capacity QLC, HBM, DDR5The supply of consumer-grade NAND has been squeezed due to the tilt towards high-value products.
- Cycle judgment: Beijing Junzheng said,The storage price increase cycle may continue until 2026, the consumer side is more obvious.
3. Impact on the industrial chain
- Original factory: Samsung, Micron, and SK Hynix increased the proportion of QLC output and suspended quotations to control prices.
- Modules / Distribution: Inventory appreciation and bargaining power increase, but cash flow and price decline risks need to be dealt with.
- downstream: The BOM costs of servers, PCs, and mobile phones are rising, and terminal prices are under pressure.
2. Industry mergers and acquisitions: accelerated integration of distribution and storage
1. Yachuang Electronics: Wholly owned by Ou Chuangxin and Yihai Nengda
- intended toIssue shares + cashAcquired the remaining 40% of Ou Chuangxin and the remaining 45% of Yihai Nengda for a transaction price of 317 million yuan.
- Strategic significance: strengthening theElectronic components distribution + analog chip designcontrol in the field to improve resource integration and operational efficiency.
2. Beijing Ingenics: Storage + NPU dual-line advancement
- storage: LPDDR4 is upgraded from 25nm to 20nm/18nm/16nm, and 4G-32G large-capacity products will be launched next year;Large-capacity NOR Flash is growing faster than DRAM。
- AI: The existing NPU has multi-T computing power and will be launched in the middle of next year.4T computing power IPC products, continue to develop NPU modules with higher computing power.
3. Power and control: SiC ecological expansion, high-end MCU iteration
1. Infineon × ROHM: SiC packaging strategic cooperation
- for each otherSecond supplier of SiC power device packaging, improve customer design and procurement flexibility.
- Technology interoperability: ROHM adopts InfineonTOLT、D-DPAKOther top cooling packages; Infineon adopts RohmDOT-247Half-bridge package, extending the SiC half-bridge solution.
- Application coverage: vehicle chargers, photovoltaics, energy storage, AI data centers.
2. Renesas: RA8T2 series MCU, the benchmark for high-end motor control
- Core configuration:1GHz Cortex-M85 + optional 250MHz Cortex-M33, supports real-time control + communication function integration.
- Performance highlights: Helium technology improves DSP/ML performance;1MB high-speed MRAM+TCM, ensuring real-time motor control; dual Gigabit Ethernet and EtherCAT slave stations to meet industrial/AIoT high-speed interconnection.
4. AI infrastructure: comprehensive upgrade of EUV, image sensing, and passive components
1. ASML: EUV orders surge, AI-driven equipment demand
- 2027 delivery plan:10 sets of High-NA EUV+56 sets of EUV。
- Customer order expansion: Intel increased High-NA EUV from 1 to 2 sets, EUV from 3 to 5 sets; Samsung EUV increased from 5 to 7 sets; SK Hynix plans to install within two years20 sets of EUV, all for HBM and advanced storage.
2. Howe: OV50R image sensor, standard for high-end mobile phones
- Specifications:50 million pixels, 1.2μm, supports four-in-one pixel binning.
- Core advantages:110dB ultra-high dynamic range (HDR), 12.5 million pixels 120fps, 4K 60fps three-channel HDR, 100% QPD focusing, and excellent dark performance.
- Progress: Samples have been produced.Mass production in Q1, 2026。
3. Fenghua Hi-tech: AI server and robot passive components are implemented
- Product: LaunchMedium-high voltage, high-temperature and high-capacity MLCC, alloy resistors, and high-current inductors dedicated to AI servers, cooperate with leading AI server manufacturers.
- Application: Some specifications of products have been used inIntelligent robotfield.
5. Core conclusions and trends
- NAND super cycle established: AI+QLC+HDD transfer order driver Q4 prices will increase by 5%-10%, and the cycle may be extended to 2026, QLC will become the mainstream of enterprise-level storage.
- Accelerating industrial chain reconstruction: Distribution mergers and acquisitions strengthen vertical integration, the SiC ecosystem is open and compatible, and MCUs evolve towards high integration and high computing power.
- AI computing power demand spillover: EUV, image sensors, and passive components are fully upgraded, and fully penetrated from the data center to the terminal.
- Domestic substitution deepens: Beijing Ingenics, Fenghua Hi-Tech, etc. continue to make breakthroughs in the fields of storage and passive components, adapting to AI and high-end applications.