作者 | 哥吉拉 数据支持 | 勾股大数 据（www.gogudata.com） 甲骨文绩后股价暴跌引发美股对AI板块大跌的市场情绪还没平复，周五晚的博通带来更大的惊魂夜。 因为部分数据没能让狂热的市场预期满意，博通绩后单日重挫11.43%，一举击溃市场信心。 雪上加霜的是，除了甲骨文和博通的业绩利空冲击，彭博也同时报道了甲骨文将部分为Open AI开发的 数据中心的竣工时间从2027年推迟至2028年的消息。 还有一家网红AI基建公司Fermi被爆重要客户取消租约导致股价盘中崩盘暴跌46%，引发更大恐慌。 联合冲击下，一股极为强烈的抛售潮迅速蔓延，光模块、交换机、电力、存储等一众AI产业链股跟随 跳水，多只此前热门概念股如Astera Labs、Core Weave、Coherent 等跌幅都超过了10%，最终美股纳指 大跌1.91%。 一时间，"AI泡沫崩盘"的担忧在市场蔓延。 对于这一次越发强烈的AI抛售潮，我们应该更关注什么？ 01 暴跌真相 如何认真去研判甲骨文和博通的财报数据，其实可以发现AI行业的基本面还很稳固。 此次AI板块的全线大跌，核心逻辑绝非AI需求的突然收缩。 博通财报显示，未来18个 ...

Group 1 - The technology sector showed strong performance on September 11, with leading gains in communication, semiconductors, and electronic components [1] - The Tianhong Sci-Tech Innovation Index ETF (589860) rose by 4.15% with a turnover rate exceeding 11%, indicating active trading [1] - Key stocks in the Tianhong Sci-Tech Innovation Index ETF included Juguang Technology and Haiguang Information, both hitting the 20% daily limit up [1] Group 2 - The Tianhong Chip ETF (159310) surged over 6% in early trading, with major weighted stocks like Haiguang Information hitting the 20% limit up and Cambrian-U rising over 10% [1] - The Tianhong Sci-Tech Innovation Index ETF closely tracks the Sci-Tech Innovation Index, covering approximately 97% of the market capitalization of the Sci-Tech Innovation Board, focusing on small-cap hard technology companies [1] - Broadcom reported impressive third-quarter earnings, with revenue reaching a record high, up 22% year-over-year, exceeding analyst expectations [2] Group 3 - Broadcom's AI chip revenue grew by 63% year-over-year to $5.2 billion in the third quarter, with expectations for further acceleration to $6.2 billion in the fourth quarter [2] - OpenAI is set to produce its own AI chips in collaboration with Broadcom, aiming to reduce reliance on NVIDIA chips [2] - Analysts suggest that the demand for ASIC chips is strong, with a positive outlook for domestic ASIC industries, supported by Broadcom's quarterly performance and customer expansion [2]

Investment Rating - The report maintains an "Overweight" rating for the optical communication sector [4]. Core Insights - The optical communication sector has experienced significant volatility recently, but strong demand and large orders in the overseas AI computing field indicate that the fundamentals of the optical module industry remain solid. The AI-driven computing expansion cycle is far from over, and the recent market adjustments provide a better investment opportunity for long-term investors [1][26]. - The core logic driving long-term growth in the optical module industry remains unchanged, with exponential growth in AI computing demand necessitating faster and more efficient data transmission capabilities. Major overseas cloud service providers have significantly increased their capital expenditures, reflecting high industry prosperity [3][28]. Summary by Sections Investment Strategy - The report emphasizes the importance of focusing on the optical communication sector, particularly recommending leading companies in the optical module industry such as Zhongji Xuchuang and NewEase, as well as other related firms [10][11][18]. Market Review - The communication sector has seen a decline, with the optical communication segment performing relatively well compared to other sub-sectors. The report notes that the optical communication index increased by 0.1%, while other indices experienced declines [22][25][23]. AI Computing Infrastructure - Major global AI companies are accelerating their computing infrastructure development through large-scale collaborations and self-developed chip deployments. Companies like Google and Meta have significantly raised their capital expenditure forecasts for AI infrastructure [2][8][30]. Demand for Optical Modules - The demand logic for optical modules remains intact, driven by the ongoing need for enhanced data transmission capabilities due to the exponential growth in AI computing requirements. This is evidenced by substantial increases in capital expenditures from major cloud service providers [3][30]. Short-term Market Adjustments - Recent adjustments in the A-share optical communication sector are attributed more to market sentiment and fund flow changes rather than fundamental shifts in the industry. The report suggests that these adjustments do not hinder the long-term demand logic driven by AI [9][31]. Key Recommendations - The report recommends focusing on the optical communication sector and related companies, highlighting specific firms such as Zhongji Xuchuang, NewEase, and Tianfu Communication, among others. It also suggests monitoring domestic computing supply chains, particularly in liquid cooling segments [10][11][18].

Group 1 - Broadcom's latest financial report shows Q3 FY2025 revenue reached $15.952 billion, a 22% year-over-year increase, exceeding the previous guidance of $15.8 billion [1] - Adjusted net profit for the same quarter was $10.702 billion, reflecting a year-over-year growth of 30.15% [1] - AI-related business revenue for the quarter was $5.2 billion, up 63% year-over-year, surpassing the expected $5.1 billion [1] Group 2 - Broadcom's core competitiveness lies in custom ASIC chips for cloud service providers, which are designed for specific AI inference scenarios, gaining favor from tech giants like Google and Meta [3] - The CEO of Broadcom indicated that the record revenue in Q3 was driven by strong growth in custom AI accelerators, networking equipment, and VMware business [3] - Broadcom has received a $10 billion order for XPU chips, with market speculation suggesting the client may be OpenAI [3] Group 3 - The ASIC chip market is thriving, with Marvell reporting Q2 FY2026 revenue of $2.006 billion, a 58% year-over-year increase, driven by data center business revenue of $1.491 billion, which grew 69% [3] - Industry analysis indicates that while NVIDIA dominates the GPU market during the AI training phase, ASIC chips are more competitive in the inference phase due to their energy efficiency [4] - Major North American cloud providers are developing their own ASICs, with predictions that the inference market will surpass the training market as AI applications expand [4] Group 4 - NVIDIA has initiated strategies to respond to the rise of the ASIC market, including the launch of the NVLink Fusion platform to integrate custom ASICs with NVIDIA GPU racks [4] - In the domestic market, companies like Haiguang Information and Moore Threads are competing in the GPU space, while Huawei HiSilicon, Cambricon, Baidu Kunlun, and Alibaba Pingtouge are active in the ASIC sector [4] - The compound annual growth rate for high-end cloud AI accelerators from 2023 to 2028 is projected at 50% for GPUs and 52% for ASICs, with ASIC shipments expected to surpass GPUs by 2028 [4]

Group 1 - Former CSRC Chairman Yi Huiman is under investigation for serious violations of discipline and law, which may relate to financial corruption issues at ICBC and family connections [2] - The public fund fee reduction reform is nearing completion, expected to benefit investors by over 50 billion yuan annually, with significant changes to sales fee management [2] - The U.S. non-farm payroll data for August showed only 22,000 new jobs added, leading to increased expectations for a Federal Reserve rate cut [2] Group 2 - Notable investor Zhao Jianping has seen a floating profit of over 470 million yuan in Q3, with his holdings in A-share companies increasing significantly due to market upturns [3] - Broadcom reported a revenue of 15.952 billion USD for Q3 FY2025, with AI business revenue reaching 5.2 billion USD, and secured a new order worth 10 billion USD for XPU chips [3] - CITIC Securities suggests focusing on resources, innovative drugs, consumer electronics, chemicals, gaming, and military industries in September, anticipating a favorable environment due to potential Fed rate cuts [4] Group 3 - The cryptocurrency market experienced a widespread decline, with over 70,000 individuals facing liquidation, totaling 270 million USD in losses [5] - Controversy arose over mobile recharge policies, with major telecom operators allowing small or custom amounts through official apps, while third-party platforms faced criticism for high minimum recharge limits [5] - Xin Xuan Group responded to rumors regarding its founder being investigated, stating that the information is untrue [5]

Core Viewpoint - Broadcom has reported impressive earnings driven by strong demand for AI-related products, positioning itself as a significant competitor to Nvidia in the AI chip market [1][5]. Financial Performance - For Q3 of fiscal year 2025, Broadcom achieved revenue of $15.952 billion, a year-on-year increase of 22%, slightly exceeding the previous guidance of $15.8 billion [1]. - Adjusted net profit for the same period was $10.702 billion, reflecting a year-on-year growth of 30.15% [1]. - AI business revenue reached $5.2 billion, up 63% year-on-year, surpassing the previous quarter's guidance of $5.1 billion [1]. Market Position and Strategy - Unlike Nvidia, which focuses on general-purpose GPU chips, Broadcom specializes in designing custom ASIC chips tailored for cloud service providers, enhancing compatibility with their specific AI inference needs [2][3]. - Broadcom's CEO highlighted that the company expects AI semiconductor revenue to accelerate, projecting $6.2 billion for Q4, marking 11 consecutive quarters of growth [5]. - The XPU business accounted for 65% of Broadcom's overall AI revenue in Q3, with significant demand from three major clients [5]. Industry Trends - The competition between GPU and ASIC chips is expected to continue, with both types of chips serving different market needs in AI infrastructure [4][8]. - ASIC chips are gaining traction due to their efficiency in AI inference tasks, which are increasingly favored by cloud service providers [6][7]. - Analysts predict that by 2028, the shipment volume of AI ASICs will surpass that of GPUs, with a compound annual growth rate (CAGR) of 52% for ASICs compared to 50% for GPUs [9].

Investment Rating - The report maintains a "Recommendation" rating for the communication industry, expecting the industry index to outperform the benchmark index by more than 5% in the next 3-6 months [30]. Core Insights - Marvell's investor day highlighted a positive outlook, with significant growth expected in the data center market, projecting a potential market size increase from $75 billion to $94 billion by 2028, with a compound annual growth rate (CAGR) of 35% [15][14]. - The demand for customized AI chips is expanding, driven by a diverse customer base including traditional cloud computing companies and new AI infrastructure builders [15][14]. - Xiaomi launched its first AI smart glasses, positioning them as a personal smart device for the next era, showcasing advancements in edge AI technology [18][19]. Summary by Sections Industry Overview - The communication sector saw a weekly increase of 5.53%, outperforming the CSI 300 index by 3.58 percentage points [7][8]. - Year-to-date, the communication sector has risen by 7.36%, significantly outperforming the CSI 300 index, which has decreased by 0.33% [7][8]. Marvell Insights - Marvell's AI chips are transitioning from general GPU assembly to highly customized system-level collaboration, with XPU currently holding a 25% market share in AI computing [14][15]. - The fastest-growing segments within the data center market are expected to be the customized computing (XPU) market, projected to reach $40 billion with a CAGR of 47%, and the XPU component market, expected to reach $15 billion with a CAGR of 90% [15][14]. Xiaomi Product Launch - Xiaomi's AI glasses feature a lightweight design, advanced camera capabilities, and multi-modal smart interaction, aiming to enhance user experience in various applications [18][19][20]. - The glasses are equipped with a dual-chip architecture for efficient AI processing and energy management, supporting various functionalities including real-time translation and video calls [21][22]. Investment Recommendations - Key recommendations include focusing on major operators like China Mobile, China Telecom, and China Unicom, as well as leading companies in optical modules and AI chips such as NewEase, Tianfu Communication, and Shengke Communication [24].

Core Insights - Nvidia maintains a strong position in the AI chip market with a gross margin exceeding 60%, but clients are seeking alternatives to reduce dependency on Nvidia's GPUs [1][5][6] - Major ASIC chip manufacturers like Marvell and Broadcom are ramping up efforts to develop custom AI chips, targeting significant market growth by 2028 [3][4] Group 1: Market Dynamics - AMD has launched the MI350 AI chip series, claiming superior performance compared to Nvidia, indicating increasing competition in the GPU market [1] - Marvell has revised its 2028 data center market size forecast from $75 billion to $94 billion, with a compound annual growth rate (CAGR) of 35% [3] - Marvell's target market for custom AI chips has been increased to $55 billion from $43 billion, reflecting growing demand [3] Group 2: Competitive Landscape - Marvell has secured 18 custom chip projects and is negotiating over 50 more, aiming to increase its market share in custom computing and components from below 5% in 2023 to 20% by 2028 [3] - Broadcom reported AI revenue exceeding $4.4 billion in Q2 of FY2025, with expectations of growth to $5.1 billion in Q3, marking ten consecutive quarters of growth [4] - Cloud providers are increasingly collaborating with ASIC manufacturers to develop their own AI chips, which is causing concern for Nvidia [4][5] Group 3: Technological Developments - Google has significantly advanced its TPU capabilities, with peak performance reaching 4,614 TOPS and plans to scale up to 9,200 chips for parallel processing [5] - Nvidia is responding to competitive pressures by opening its NVLink ecosystem to allow ASICs to run on its platform, aiming to maintain its dominance in AI cloud computing [6] - Nvidia is also promoting "sovereign AI" initiatives globally, including plans to build an industrial AI cloud in Germany with 10,000 Blackwell GPUs [6] Group 4: Industry Growth - Nvidia's growth has been a major driver for the semiconductor IC industry, with a reported revenue growth rate of 125% compared to a maximum of 21% for other top fabless companies [6] - The semiconductor IC industry is expected to grow by approximately 19% in 2025, benefiting from strong demand and preemptive inventory stocking [7]

Summary of Key Points from the Conference Call Industry Overview - The conference call focuses on the overseas computing power chain, specifically in the PCB (Printed Circuit Board) sector, driven by the increasing demand for AI capabilities, particularly from companies like OpenAI and Xia [1][2]. Core Insights and Arguments - **AI Demand Growth**: The evolution of AI narratives is driving higher computing power demand, with significant needs from OpenAI and Xia. The increase in TOKEN numbers validates the strong demand for overseas inference computing power, indicating an acceleration in the iteration path of AI large models [1][2]. - **Investment Opportunities in PCB Sector**: From the second half of 2025 to 2026, investment opportunities in the PCB sector are concentrated around NVIDIA and the ASIC (Application-Specific Integrated Circuit) supply chain. The PCB market demand is expected to exceed 15 billion by 2026, with major suppliers including Shengyi Technology, Jinxiong, and Huitian [3][8]. - **Hardware Upgrades**: Major companies like Amazon, Google, and Meta are expected to undergo significant hardware upgrades. For instance, Amazon's T3 servers will enhance GPU acceleration cards and PCB value by over 20% due to material upgrades [5][11]. - **Google's Chip Production**: Google is projected to ship 2 million TPU chips in 2025, with an increase to 3 million in 2026. However, its PCB demand is relatively lower due to the use of lower-grade materials [9][11]. - **Meta's XPU Chips**: Meta anticipates producing millions of XPU chips, each requiring high-spec PCBs, with potential suppliers including Huitian and Tianpin [12]. Additional Important Insights - **Valuation of Core PCB Companies**: Current valuations of core PCB companies are reasonable, with potential for further price increases. Companies like Shengyi Technology are recommended for investment due to their strong position in the ASIC supply chain [3][13]. - **Market Dynamics**: The PCB market is expected to see significant growth, with a projected increase in demand due to the expansion of major clients like Amazon and Google. Shengyi Technology is expected to add substantial production capacity to meet this demand [14][16]. - **Profitability Forecasts**: Shengyi Technology's earnings are projected to reach 1 billion in 2025 and 1.7 to 1.8 billion in 2026, with a potential market value of 40 billion [14]. Huitian is also expected to achieve significant profit growth, with forecasts of 3.7 billion in 2025 and 4.6 billion in 2026 [15]. - **Emerging Trends in AI**: The resurgence of pre-trained models is anticipated to increase computing power demand, with a focus on building large clusters for AI training [22][26]. Conclusion - The overseas computing power chain, particularly in the PCB sector, is poised for significant growth driven by AI demands. Key players like Amazon, Google, and Meta are expected to lead hardware upgrades, creating substantial investment opportunities in the PCB supply chain. The market dynamics suggest a favorable outlook for companies involved in this sector, with reasonable valuations and strong growth potential.