Group 1 - The core viewpoint of the report is that 2026 will witness strong growth in the AI supply chain, driven by the expansion of capital expenditures from hyperscale cloud service providers, rapid advancements in AI hardware technology, and a tight supply chain for components [1] Group 2 - Capital expenditures from top global cloud service providers are expected to increase by 34% year-on-year in 2026, reaching approximately $424 billion, primarily fueled by the ongoing expansion of cloud services and internet businesses, with cloud service revenue growth rebounding to over 28% [2] Group 3 - The report predicts a prosperous coexistence of AI GPUs and custom chips (ASICs) in 2026, with Nvidia's Blackwell platform expected to enter large-scale delivery, anticipating the delivery of around 60,000 racks throughout the year [3] - Google and Amazon are leading the deployment of custom chips, with Google's TPU v7 and Amazon's Trainium T3 expected to be widely deployed in 2026, while Meta plans to launch its ASIC solution in the second half of the same year [3] Group 4 - Traditional general-purpose servers are also showing a recovery growth in the high single to mid-single digits, alongside the AI boom [4] Group 5 - The report highlights an uneven distribution of growth benefits across the supply chain, with markets for rack power supplies, cooling, and PCB substrates expected to continue expanding, while tight supply of various raw materials, especially memory, creates a "seller's market" benefiting component suppliers [5] - Due to high prices of storage components, the report has downgraded the 2026 PC shipment forecast, expecting a 4% decline, while smartphone growth is also expected to slow to 2% due to rising commodity prices [5] Group 6 - The report suggests that the driving factors for stock prices in 2026 will be sales and profit growth rather than further valuation expansion, identifying the best opportunities among suppliers that can benefit from the increase in AI server deployments and enhance unit value or added value [6] - Recommended stocks include server ODM manufacturers (Quanta, Hon Hai, Wistron, Wistron NeWeb) and component suppliers (Delta, Jentech, AVC), as well as Largan Precision, which benefits from upgrades in foldable screens and variable aperture specifications [6] - The report takes a more cautious stance on brand companies (ASUS, Lenovo, Gigabyte, MSI) and ODM and component companies (Compal, Pegatron, Inventec, JMicron) that are less likely to benefit from AI server growth due to high input costs, especially in storage [6]

2026年初，全球内存行业再次迎来高爆发，"超级存储周期"的提法再次被摆放到前台。 美银发布一份研究报告指出，DRAM将在2026-27年开启史上最强超级周期，全球DRAM销售额预计从 2025年的1303亿美元飙升至2026年的2098亿美元，同比暴涨61%。 美银预计，2027年这个数字将进一步增至2314亿美元。趋势背后，AI需求爆发、供应结构性短缺与技 术迭代形成三重共振。 在报告中，美银大幅调高了SK海力士、三星电子、南亚科技等核心标的目标价，认为内存行业正从传 统周期性产业向AI驱动的成长性赛道蜕变。 01 超级周期的五大核心支撑 美银报告认为，内存行业此前深陷"扩产-过剩-降价-减产"的周期循环，但2026年启动的超级周期呈现本 质不同——AI带来的刚性需求与供应端的约束形成长期错配，推动行业进入高质量增长阶段。 首要驱动是AI内存需求的爆发式增长 以英伟达Rubin Ultra GPU、谷歌TPU v7为代表的AI加速器，对HBM(高带宽内存)、GDDR7等高端内存 的搭载量呈数倍提升，单台AI服务器的HBM搭载量已从早年的340GB增至1.17TB，英伟达下一代Rubin Ultra更是有望 ...

海外 投资评级：看好（维持） 行业走势图 -10% 0% 10% 20% 30% 40% 2024-12-23 2025-01-23 2025-02-23 2025-03-23 2025-04-23 2025-05-23 2025-06-23 2025-07-23 2025-08-23 2025-09-23 2025-10-23 2025-11-23 Wind 恒生指数 2025 年 12 月 21 日 数据来源：聚源 相关研究报告 《存储依然高景气，H200 有望获批对 华出口 —行业周报》-2025.12.14 《谷歌 Gemini3 增益生态，存储有望 持续高景气—行业周报》-2025.12.7 《AI 需求持续验证，提升算力需求可 预见性—行业周报》-2025.11.23 国内 L3 级准入试点许可，百融云硅基员工+ Anthropic Skill 范式发布，关注 Agent 产业机会 ——行业周报 | 初敏（分析师） | 张可（分析师） | 杨哲（分析师） | | --- | --- | --- | | chumin@kysec.cn | zhangke1@kysec.cn | yangzhe@ky ...

Core Insights - The article discusses Google's TPU v7 interconnect architecture, focusing on the ratio of TPU to copper cables and optical modules, highlighting the technical aspects of the TPU design and its cooling solutions [1][6][7]. TPU Rack Interconnect Architecture - One of the notable features of TPU is its ability to achieve large-scale world size expansion through the ICI protocol, with a TPU Pod capable of accommodating up to 9216 Ironwood TPUs [2]. - Each TPU rack consists of 16 TPU trays and a varying number of host CPU trays, along with a top-of-rack switch and power units [2]. - The TPU tray contains a TPU board with four TPU chips, each equipped with multiple interfaces for interconnectivity [2]. Cooling Solutions - Google has adopted liquid cooling for TPU racks since the TPU v3 era, with a 1:1 ratio of TPU trays to host CPU trays in liquid-cooled racks, compared to a 2:1 ratio in air-cooled racks [6]. - The market anticipates that 2024 will be the "year of liquid cooling," as more ASIC servers begin to adopt this technology, indicating significant market growth potential [6]. Market Projections - In 2026, Google is expected to ship 2.5 million TPU v7 units, leading to a liquid cooling market space of approximately $2.8 to $3.2 billion [7]. - By 2027, shipments are projected to exceed 5 million units, with the value of liquid cooling per rack potentially increasing to $90,000 to $100,000, resulting in a market space of $7 to $8 billion [7]. Interconnect Design - The TPU v7 utilizes a 3D torus topology for interconnectivity, where each TPU connects to six neighboring nodes across three dimensions [8]. - Internal connections within the TPU tray use copper cables, while external connections utilize optical modules and OCS for inter-unit communication [9][12]. Optical Connectivity and Market Demand - A TPU Pod with 9216 TPUs will require approximately 11,520 copper cables and 13,824 optical modules, indicating a significant demand for optical components in the market [16]. - Google is projected to need around 15,000 OCS switches by 2026, with a market space for OCS estimated at $2.2 billion based on a price of $150,000 per switch [17][18].

Investment Rating - The industry investment rating is "Positive" [2] Core Insights - The electronic sub-industry has significantly recovered, with the Shenwan Electronics secondary index showing year-to-date changes of: Semiconductors (+39.75%), Other Electronics II (+43.95%), Components (+89.82%), Optical Electronics (+5.55%), Consumer Electronics (+42.54%), and Electronic Chemicals II (+38.20%) [2][9] - North American key stocks mostly rose, with notable increases for companies like Micron Technology (+180.99%) and Intel (+102.29%) year-to-date [10] - Google's TPU v7 demonstrates cost advantages over GPU-based systems, challenging the GPU-dominated computing market. The total cost of ownership (TCO) for TPU is approximately 30%-40% lower than NVIDIA's GB200 system [2][24] - The demand for AI infrastructure is growing significantly, with NVIDIA reporting that cloud GPUs are sold out, indicating a supply-demand imbalance. TrendForce predicts over 20% year-on-year growth in global AI server shipments by 2026 [2][3] Summary by Sections Market Performance - The Shenwan Electronics secondary index has shown substantial recovery, with weekly changes for various segments: Semiconductors (+5.72%), Other Electronics II (+7.59%), Components (+8.10%), Optical Electronics (+5.23%), Consumer Electronics (+6.08%), and Electronic Chemicals II (+3.93%) [9] - Key North American stocks have shown positive performance, with significant increases for companies like Tesla (+9.99%) and Qualcomm (+2.93%) [10] Technology Competition - Google's TPU v7 has emerged as a strong competitor in the computing market, leveraging superior system-level engineering to achieve higher model performance utilization rates compared to NVIDIA GPUs [2][24] - The competition between GPU and TPU is seen as a redistribution of market share in a growing market, with both technologies expected to experience rapid growth [2] Investment Opportunities - Recommended companies to watch include: For overseas AI - Industrial Fulian, Huadian Co., Pengding Holdings, Shenghong Technology, and Shengyi Technology; For domestic AI - Cambricon, Chipone, Haiguang Information, SMIC, and Shenzhen South Circuit [3]

Market Overview - The market experienced a pullback after an initial rise, with the total trading volume in Shanghai and Shenzhen reaching 1.78 trillion, a decrease of 28.8 billion from the previous trading day [1] - The Shanghai Composite Index fell by 0.15%, while the Shenzhen Component Index rose by 1.02%, and the ChiNext Index increased by 2.14% [1] - The Communication ETF showed strong performance, peaking at nearly 8% during the day and closing up by 5.61%, while the AI ETF on ChiNext rose by 4.89% [1] Factors Driving Market Upward - Increased expectations for interest rate cuts, with the likelihood of a 25 basis point cut by the Federal Reserve in December rising from approximately 40% to 80% following comments from several Fed officials [2] - Google's recent launch of the Gemini 3 model, which has outperformed competitors in various benchmarks, contributing to positive sentiment in the AI sector [2][4] AI Model Performance - The Gemini 3 Pro model has shown superior performance across multiple benchmarks, achieving 95.0% in mathematics without tools and 100% with code execution, indicating its advanced capabilities compared to other models [3] - The model also excelled in scientific knowledge, achieving 91.9%, and demonstrated strong performance in visual reasoning and multimodal understanding [3] Hardware Developments - Google’s TPU v7 has exceeded expectations in terms of memory capacity and energy efficiency, potentially impacting the competitive landscape for AI hardware [5] - There are reports that Meta Platforms is considering a significant investment in Google’s TPU for its data center needs, indicating a shift in the competitive dynamics of the AI hardware market [5] Future Market Outlook - There is a gradual improvement in risk appetite as pessimistic expectations are being digested, with the potential for continued recovery in the U.S. stock market [7] - Despite potential changes in the chip landscape due to Google's growth, A-share related component manufacturers are expected to benefit regardless of which company leads in chip competition [7] - The AI industry is anticipated to continue its rapid growth, with increasing demand for computational power, which may positively impact hardware and software sectors [9]

Market Overview - The computing sector experienced a significant rise, with the Communication ETF (515880) increasing by 7% and the ChiNext AI ETF (159388) rising over 6% [1][2]. Factors Driving the Market - Increased expectations for interest rate cuts have emerged, as several Federal Reserve officials, including allies of Powell, expressed support for a potential 25 basis point cut in December, raising the likelihood from approximately 40% to 80% [2]. - Google's recent release of the Gemini 3 model, which has outperformed its predecessors in various benchmarks, has contributed to positive market sentiment [2][4]. AI Model Performance - The Gemini 3 Pro model has shown superior performance across multiple benchmarks, including achieving 95% in mathematics and 91.9% in scientific knowledge assessments, indicating its advanced capabilities compared to competitors [3]. Hardware Developments - Google's TPU v7 (Ironwood) has exceeded expectations in terms of memory capacity and energy efficiency, posing a competitive challenge to NVIDIA's offerings. Google anticipates a doubling of computing power every six months to meet rising AI service demands [5]. Market Outlook - There is a gradual improvement in risk appetite as pessimistic expectations are being absorbed by the market, with strong support levels in the A-share market despite rapid style shifts [6]. - While the competitive landscape for computing chips may evolve, NVIDIA is expected to maintain revenue growth due to its existing market share and new demands from sovereign AI projects [6]. - The AI industry is progressing rapidly, with significant investments expected to sustain growth, although short-term volatility should be monitored [7].