Core Insights - The report from GF Securities highlights the significance of the Scale-Up architecture in addressing the challenges posed by training and inference of trillion-parameter models, particularly focusing on the advancements in AI computing power [1][2]. Group 1: Current Trends in AI Computing - The core characteristics of model inference, such as long context, high concurrency, and real-time interaction, are driving the upgrade of Scale-Up systems [2]. - The need for high-performance networks during the inference phase is emphasized, as the architecture allows all accelerators' memory to be presented as a single shared pool, addressing bandwidth and latency requirements [2]. Group 2: Domestic Developments in Scale-Up - Alibaba Cloud introduced the new generation of Panjiu AI Infra2.0 AL128 supernode server, which aims to optimize computing power and communication synergy, achieving a 50% improvement in inference performance compared to traditional architectures [3]. - Tencent's ETH-X supernode project is being developed in two phases, focusing on optimizing GPU and memory communication and exploring full optical interconnection for Scale-Up [3]. - Huawei announced a three-year computing power plan, with the Ascend 950 supernode expected to launch in Q4 2026, indicating a strong commitment to advancing Scale-Up technology [3]. Group 3: Investment Opportunities in Optical Interconnection - Current Scale-Up networks primarily use copper cabling, which has limitations in system design and transmission distance, making it less suitable for future expansions [4]. - The advantages of optical interconnection are highlighted, as it can cover much greater distances compared to copper, which is limited to around 7 meters for certain applications [4].

Core Viewpoint - Huawei is set to launch its latest AI chip, Ascend, in South Korea next year, aiming to penetrate the domestic AI infrastructure market and provide an alternative to NVIDIA [2]. Group 1: AI Chip Launch - Huawei plans to officially release AI computing cards and related solutions in South Korea next year [2]. - The upcoming chip, Ascend 950, is expected to begin mass production next year [2]. - Unlike NVIDIA, Huawei intends to sell its products in cluster units rather than individually [2]. - The strategy includes not just providing AI cards and servers but also accelerating industrial applications through end-to-end solutions encompassing hardware and software [2]. Group 2: Operating System Development - Huawei will offer its self-developed open-source operating system, HarmonyOS, to domestic enterprises in South Korea to foster ecosystem development [3]. - The ownership of HarmonyOS has transitioned from Huawei to an open-source organization responsible for its operation and upgrades [3]. - HarmonyOS is designed for use not only in smartphones but also in various smart home devices [3]. Group 3: Market Strategy - There are no plans to launch smartphones in South Korea next year [4].

Summary of Key Points from the Conference Call Industry Overview - **DDR5 Spot Price Increase**: The spot price for 16Gb DDR5 increased by 10% this week, reaching a new high of nearly US$7, compared to a low of US$4 in 2023-2024. The price for legacy DDR4 8Gb also rose by 6% [1][10][12]. - **Gradual Price Recovery**: Year-to-date, DDR5 prices have shown a gradual recovery, with only 1-2% week-over-week increases prior to this week [1]. - **Catalysts for Price Increase**: The sudden price increase is attributed to: 1. Stronger demand due to increased memory content and replacement cycles for old PCs and servers. 2. Capacity shifts from DDR5 to HBM4 and other memory types. 3. Inventory restocking by OEMs ahead of a projected supply shortage in Q4 [1]. - **Annual Contracts**: Some OEMs and memory module manufacturers are looking to secure annual contracts for DDR5 at higher prices, a shift from the conventional quarterly contracts due to concerns over long-term supply shortages [1]. Memory Technology Developments - **NVIDIA's Vera Rubin NVL 144 CPX**: This new GPU is expected to utilize three types of memory, totaling approximately 100TB, which is 2.5 times larger than the current Blackwell Ultra NVL 72. The memory types include HBM4, GDDR7, and SOCAMM [2]. - **SOCAMM vs HBM**: SOCAMM offers lower bandwidth compared to HBM4, suggesting a limited price premium compared to LPDDR5 but is cheaper than GDDR7 [2]. Samsung's HBM3e Qualification - **Samsung's HBM3e**: Reports indicate that Samsung's 12-hi HBM3e has passed NVIDIA's quality testing, aligning with management expectations. However, a large order is not anticipated soon due to Hynix's strong market position [3]. China's Semiconductor Industry - **Positive Data**: China's semiconductor equipment imports rose by 12% year-over-year in August, exceeding market expectations. This increase is partly due to Korean chipmakers' investments in fab upgrades ahead of US restrictions [4]. - **Huawei's GPU Line-Up**: Huawei announced new GPU line-ups, including the Ascend 950 for 2026, which still rely on legacy HBM2-equivalent memory rather than advanced HBM4+ [4]. Price Trends and Market Dynamics - **DRAM Price Trends**: The current spot prices for various DRAM types are as follows: - 16Gb DDR5: US$6.8 (up 10% WoW, 39% YoY) - 16Gb DDR4: US$10.2 (up 6% WoW, 168% YoY) - 8Gb DDR4: US$6.1 (up 6% WoW, 211% YoY) [7][10]. - **Contract Price Trends**: DDR4 prices have seen a significant increase, with a crossover occurring in Q2 2025 where DDR4 prices surpassed DDR5 prices [27][29]. Conclusion - The memory industry is experiencing a significant price rally, particularly in DDR5 and DDR4 segments, driven by demand and supply dynamics. The developments in memory technology, particularly from major players like NVIDIA and Samsung, indicate a competitive landscape with potential for further price adjustments and contract negotiations in the near future.

Group 1 - The People's Bank of China maintains the 7-day reverse repurchase rate and the Medium-term Lending Facility (MLF) rate unchanged, indicating a steady approach while observing external monetary easing and domestic inflation recovery [2] - Global economies are showing resilience against tariff impacts, with manufacturers adapting through production shifts, price adjustments, and inventory management, suggesting a potential return to demand-driven fundamentals if policy expectations stabilize [3] - China's rare earth magnet exports reached a seven-month high in August, increasing by 10.2% month-on-month to 6,146 tons and 15.4% year-on-year, supported by demand from the new energy vehicle and wind power sectors [4] Group 2 - Oil prices have risen temporarily due to geopolitical risks, but institutions predict potential oversupply in Q4 and 2026, creating downward pressure on long-term prices, with short-term dynamics influenced by geopolitical factors and inventory levels [5] - Huawei unveiled its AI chip and computing roadmap, introducing the Ascend 950, 960, and 970 series, along with the Atlas computing platform, highlighting breakthroughs in self-developed high-bandwidth storage [6] - Volkswagen Group has set aside approximately €5.1 billion to address delays in Porsche's electric vehicle launches, reflecting adjustments in the automotive industry's transition to new energy vehicles amid global demand slowdown and increased competition in China [7] Group 3 - BYD has implemented new payment norms for suppliers, committing to complete acceptance within three working days and payment within 60 days, aimed at alleviating cash flow pressures in the supply chain and stabilizing supplier engagement [8]