纯粹以"再做一个大模型"为核心卖点的创业项目正快速降温，取而代之的，是轻量化小模型、AI Agent以及设备端AI等方向。 两年前，AI创业的关键词还是"百模大战"。从通用大模型到行业模型，参数规模不断刷新纪录，融资金额屡创新高，市场一度形成"谁不做大模型，谁就会 被淘汰"的共识。 但近阶段，风向正在发生明显变化。无论是在硅谷，还是在国内创投圈，纯粹以"再做一个大模型"为核心卖点的创业项目，正快速降温。取而代之的，是轻 量化小模型、AI Agent以及设备端（On-Device）AI等方向。 这并非技术热情消退，而是一场由成本、商业化和资本逻辑共同驱动的理性回归。在2026年的国际消费类电子产品展览会(CES)上，这一趋势尤其明显。第 一财经独家采访了设备端（On-Device）AI创业公司Aizip的联合创始人陈羽北。他是典型的在美华人工程师代表，曾于清华大学和加州大学伯克利分校分别 获得学士和博士学位，如今在加州大学戴维斯分校任教并进行创业项目。 陈羽北表示，设备端AI正在成为创业公司的新赛道。通过"用大模型生成小模型"的方法，通过收集数据、购买数据和大模型蒸馏三种途径获取所需数据，使 AI能够直接在终端 ...

Core Insights - The article discusses the significant restructuring of investment banking teams at major firms like Goldman Sachs, Morgan Stanley, and JPMorgan in response to the booming AI infrastructure market, particularly in the TMT (Technology, Media, and Telecommunications) sector [1][2] - Goldman Sachs' recent report indicates a strong demand for AI server clusters, which is expected to continue growing through 2026, alongside robust demand for optical networking equipment and a moderate supply increase in the DRAM storage chip market [1][3] Investment Banking Restructuring - Goldman Sachs is creating a new global infrastructure technology business unit, integrating its telecom and CoreTech teams, led by Yasmine Coupal and Jason Tofsky [1] - Another new unit, focused on global internet and media, will be led by Brandon Watkins and Alekhya Uppalapati, as the firm seeks to capitalize on large-scale investment opportunities in the AI sector [2] AI Infrastructure Demand - The demand for AI servers is projected to remain strong, with significant growth expected in AI ASIC clusters led by Google, outpacing AI GPU shipments, which will also see robust growth [2][3] - Nvidia's Rubin architecture AI GPU clusters are set to begin production in mid-2026, with strong capacity ramp-up anticipated in the latter half of that year [3] Market Predictions - Goldman Sachs' stock strategists predict that the S&P 500 index will reach around 7600 points next year, indicating a potential 10% upside from current levels, driven by widespread AI technology adoption and resilient economic growth [3][4] - The overall earnings per share for S&P 500 companies are expected to jump by 12% next year, with a further 10% increase in 2027, reflecting the positive impact of AI on corporate profitability [3][4] AI Investment Wave - The article highlights that the current wave of investment in AI infrastructure is just beginning, with projections estimating that the total investment could reach between $3 trillion to $4 trillion by 2030, driven by unprecedented demand for AI computing power [4]

来源：市场资讯 （来源：浙商证券融资融券） 12月9日，市场缩量调整，指数涨跌不一，个股多数收跌。三市共成交1.92万亿，较上个交易日缩量1300多亿；共有4000多股收跌。当日， 北向资金成交总额2196亿元。 短期来看，市场小幅调整消化获利盘，不改震荡反弹之势，接下来还有12月份的第二场重磅会议，利好预期有望推动市场继续反弹走高。 中期来看，美联储2026年降息前景持续升温。近日，在12月议息会议召开前夕，下一任美联储主席热门候选人、美国白宫国家经济委员会 主任凯文·哈西特突然表示，美联储应该继续降低利率。哈西特在降息问题上更为激进，此前他在接受采访时表示，如果自己是美联储主席 会立即降息，而且支持在12月降息50个基点。后续若哈西特成功就任美联储主席，或将在2026年启动更加激进的降息政策。 配置上，关注AI算力概念。近期，谷歌TPU实现了从硬件到软件栈的深度整合，支撑了近期风靡全球的谷歌Gemini3和Nano Banana Pro的大 模型训练，成功证明了在大模型时代，ASIC可以成为继GPU之后又一个重要的AI算力核心支撑，开启AI算力新方向。国内方面，近期多家 国产AI芯片企业上市，有望推动国产 ...

本期内容提要: 本周电子细分行业大幅修复。申万电子二级指数年初以来涨跌幅分别为：半导体（+39.75%）/其他电子 Ⅱ（+43.95%）/元件（+89.82%）/光学光电子（+5.55%）/消费电子（+42.54%）/电子化学品Ⅱ （+38.20%）；本周涨跌幅分别为半导体（+5.72%）/其他电子Ⅱ（+7.59%）/元件（+8.10%）/光学光电 子（+5.23%）/消费电子（+6.08%）/电子化学品Ⅱ（+3.93%）。 信达证券近日发布电子行业周报：半导体（+39.75%）/其他电子Ⅱ（+43.95%）/元件（+89.82%）/光学 光电子（+5.55%）/消费电子（+42.54%）/电子化学品Ⅱ（+38.20%）；本周涨跌幅分别为半导体 （+5.72%）/其他电子Ⅱ（+7.59%）/元件（+8.10%）/光学光电子（+5.23%）/消费电子（+6.08%）/电 子化学品Ⅱ（+3.93%）。 以下为研究报告摘要： 当前AI基建的增长性需求远超单一技术路线承载能力。英伟达在FY26Q3财报会上表示云端GPU已经售 罄，反映了公司当前供不应求的局面。根据TrendForce预测，2026年受惠于北美大型CS ...

Core Viewpoint - The unexpected slowdown in U.S. retail sales growth highlights a decrease in consumer spending amid a weakening labor market and temporary inflation caused by tariffs, yet the resilience in retail sales supports the narrative of a "Goldilocks" economic environment in the U.S. [1] Retail Sales Data - In September, U.S. retail sales showed a modest increase of 0.2% month-over-month, falling short of the expected 0.4% growth, following a strong 0.6% increase in August. Excluding autos and gas, sales rose only 0.1% [5][6] - Among 13 categories, 8 recorded growth, primarily in gas stations and personal care stores, while auto sales declined for the first time in four months, and spending on electronics, clothing, and sports goods also decreased [6][8] Consumer Behavior - The data indicates that middle and low-income consumers are becoming more cautious due to rising inflation and employment challenges, leading to a pause in spending [5][7] - Retailers like Walmart and TJX have noted that shoppers are increasingly seeking discounts and essential goods, while Home Depot has warned of delayed large home purchases [7] Economic Outlook - Morgan Stanley economists predict that the OBBBA tax cuts from the Trump administration will significantly boost economic growth starting in 2026, alongside the temporary nature of inflation from tariffs and ongoing AI infrastructure investments by tech giants [1] - The NRF forecasts a record number of shoppers during the upcoming Thanksgiving and Black Friday shopping weekend, which could provide a significant boost to the U.S. economy in Q4 and 2026, as consumer spending accounts for 60%-70% of GDP [9][10] Federal Reserve and Interest Rates - Following the retail data release, expectations for a Federal Reserve rate cut in December have increased, with an 80% probability of a rate cut according to CME FedWatch Tool [3][4][7] - There is a notable division among Federal Reserve officials regarding the decision to cut rates, reflecting concerns over consumer affordability [7]

Core Viewpoint - The unexpected slowdown in U.S. retail sales growth highlights a cautious consumer spending environment amid a weakening labor market and temporary inflation due to tariffs, reinforcing the narrative of a "Goldilocks" soft landing for the economy [1][5][9] Retail Sales Data - In September, U.S. retail sales rose by only 0.2% month-over-month, falling short of the expected 0.4% increase, following a strong 0.6% growth in August [5][6] - Excluding autos and gas, sales increased by just 0.1%, indicating a slowdown in consumer spending momentum [5][6] - The "control group" sales, a key indicator for GDP calculations, declined by 0.1%, marking the first drop in five months [8] Consumer Behavior - The data suggests that middle and low-income consumers are feeling the pressure from inflation and employment challenges, leading to more cautious spending [5][7] - High-income consumers continue to support overall spending, but there are signs of strain among lower-income shoppers, with many seeking discounts and essential goods [7][9] Economic Outlook - Morgan Stanley economists predict that the OBBBA tax cuts and the temporary nature of inflation from tariffs will contribute to a strong economic growth effect starting in 2026, supporting the "Goldilocks" narrative [1] - The upcoming Thanksgiving and Black Friday shopping events are expected to significantly boost consumer spending, which accounts for 60%-70% of U.S. GDP [9][10] Federal Reserve and Interest Rates - Following the retail data release, expectations for a Federal Reserve rate cut in December have increased, with an 80% probability of a cut according to CME FedWatch Tool [3][4][7] - There is a notable division among Federal Reserve officials regarding the decision to cut rates, reflecting concerns over consumer affordability [7]

Core Viewpoint - The recent decline in major tech stocks, such as Nvidia, has led to a correction in the A-share technology sector, intensifying discussions about an "AI bubble" in global markets. Concerns stem from the fact that the demand for computing power from AI giants exceeds their free cash flow growth, leading to cash flow consumption and debt expansion [1][2]. Group 1: AI Industry Dynamics - The AI industry is currently viewed as being at the bottom of the Kondratiev wave cycle, suggesting that it is still in a developmental phase rather than a bubble [1]. - There is a prevailing opinion that the market focus may shift from "AI computing power" to "AI applications," indicating a potential new phase in AI investment opportunities [1][4]. Group 2: AI Applications Across Industries - AI applications are penetrating various sectors, including finance (e.g., smart risk control, intelligent investment advisory), healthcare (e.g., AI-assisted diagnosis, drug development), education (e.g., personalized learning), manufacturing (e.g., quality inspection, supply chain optimization), and more [1][2]. - The application of AI is not limited to a specific industry but is a cross-industry theme, highlighting its broad relevance [1]. Group 3: Investment Opportunities in AI - Companies providing AI applications are seen as essential "enablers" across industries, while semiconductor and chip companies are considered the foundational "builders" of the AI ecosystem [2]. - Several indices, such as the Artificial Intelligence Index and the ChiNext AI Index, have seen significant gains exceeding 60% this year, while other indices related to software and computing have underperformed [2][3]. Group 4: Recent Developments and Market Reactions - Notable recent developments include Google's launch of the Gemini3 model and Alibaba's introduction of the Qianwen app, both of which have generated positive market responses despite broader tech stock declines [3][4]. - In the recent market rebound, sector-specific ETFs related to media, gaming, software, and big data have outperformed hardware-focused indices, indicating a shift in investor sentiment towards AI applications [4].

Core Insights - The technology cycle of AI is unlikely to end in the short term, with a continued focus on computing power as a key driver [1] - NVIDIA's recent earnings report exceeded expectations, indicating that the demand for computing power is still rapidly growing [1] - Google's Gemini3 model has received positive market feedback for its strong capabilities [1] - The hardware sector for computing power is expected to see significant growth in light modules and other components in the coming year [1] Industry Overview - The ongoing investment in computing infrastructure both domestically and internationally is expected to sustain high demand in the light module market [1] - The Communication ETF (515880) is the largest in its category, with light modules accounting for 52% and servers for 22% of its composition as of October 28 [1] - When combined with fiber optics and copper connections, these components represent over 81% of the ETF, reflecting the fundamental strength of overseas computing power [1]

Core Viewpoint - Nvidia reported strong earnings, but investor sentiment shifted from excitement to caution, leading to a broad market pullback in the US stock market [1][3] Group 1: Nvidia's Financial Performance - Nvidia's Q3 revenue for fiscal year 2026 reached $57.01 billion, with an expected Q4 revenue of approximately $65 billion, both exceeding Wall Street expectations [3] - Despite the strong financial results, concerns about Nvidia's growth potential and market share nearing its peak persist among investors [3] Group 2: Market Reactions and Economic Indicators - The communication ETF (515880), which has over 50% exposure to optical modules and is closely linked to the overseas AI industry, experienced a significant pullback of over 4% [1] - The US non-farm payroll report for September showed an increase of 119,000 jobs, surpassing the expected 50,000, while the unemployment rate rose to 4.4%, the highest since October 2021 [3] Group 3: AI Industry Fundamentals - The AI industry continues to accelerate, with significant advancements in domestic GPU production and strong performance in AI benchmarks by Google's Gemini 3 Pro [4] - The optical module market is experiencing growth due to increased shipments of computing chips and clear upgrades in technology, with expectations for continued expansion in market size and penetration rates [4] Group 4: Future Outlook - The AI industry trend remains positive, with capital expenditures expected to grow, particularly in high-end products like optical modules [5] - The communication ETF (515880) has a scale of nearly 13 billion yuan, ranking first among similar products, with optical modules accounting for 52% of its holdings [5][6]