证券研究报告 | 2025年8月24日 国信通信·算力租赁专题报告 Neocloud引领算力租赁发展，国内市场发展可期 行业研究 · 行业专题 通信 投资评级：优于大市（维持评级） 证券分析师：袁文翀 021-60375411 yuanwenchong@guosen.com.cn S0980523110003 证券分析师：张宇凡 021-61761027 zhangyufan1@guosen.com.cn S0980525080005 Public 投资摘要 全球算力景气度延续，随着英伟达GB系列高密度算力机柜加速出货，全球高端算力景气度进一步提升。当前AIGC浪潮下，全球服务器出货量 持续增长，咨询机构IDC预计2028年全球人工智能服务器市场规模有望达到2,227亿美元，其中生成式人工智能服务器占比将从2025年的 29.6%提升至2028年的37.7%。从需求端来看，模型迭代加速背景下训练端需求仍维持高位，推理侧需求随着应用的渗透逐步提升；从供给 端来看，以英伟达B/Rubin、AMD MI系列为代表的高性能算力芯片持续迭代，2025年下半年GB300有望加速交付。根据CSP厂商的Capex 指引，预计 ...

如果您希望可以时常见面，欢迎标星收藏哦~ 来 源 ：内容 编译自 eenews。 X-Fab 的 Ulrich Bretthauer 博士表示，150mm CMOS 的淘汰不仅仅是产品的变化，而是一种结 构性行业风险。 电子产品和系统制造商依赖于可靠的集成电路供应，尤其是在汽车、工业、医疗和航空航天领域。 半导体器件的买家，尤其是在这些特定领域的买家，严重依赖稳定的集成电路供应。因此，过去几 个月一些代工厂停止了基于150毫米（6英寸）晶圆的CMOS工艺，这让一些公司陷入困境；这扰 乱了供应链，使设计工程师不得不重新开始设计，从而影响了战略规划。 150 毫米晶圆上 CMOS 芯片生产的停止标志着 0.6 微米及更大尺寸工艺的终结，这给汽车、工 业、医疗和其他行业的制造商带来了挑战。这些成熟的节点仍然广泛用于模拟和混合信号 IC，包 括传感器接口和电源管理芯片。对于许多设计团队来说，150 毫米 CMOS 芯片寿命终止 (EOL) 的 突然宣布几乎没有预兆。在某些情况下，这一消息会引发紧急会议，以评估库存、启动重新设计并 重新验证长期运行的系统。 随着工艺节点尺寸的不断缩小，晶圆尺寸也从 150 毫米（6 英 ...

Summary of Conference Call on Optical Networking Industry Industry Overview - The optical networking industry is expected to experience a super upcycle driven by the adoption of ASIC (Application-Specific Integrated Circuit) and SiPh (Silicon Photonics) technologies, with significant demand growth anticipated for 800G and 1.6T transceivers in 2026, projected at 45 million and 8 million units respectively [1][2][13]. Key Companies Discussed 1. **Innolight (300308.SZ)** - Top pick in the sector due to strong market share and margin upside from technology upgrades [1][4]. - Expected gross/net margin could reach approximately 45% due to BOM (Bill of Materials) optimization and lower costs associated with SiPh technology [39][41]. - Revised earnings estimates for FY25-27E show an increase of 25%-85% due to higher shipment numbers and margin improvements [41][43]. 2. **Eoptolink (300502.SZ)** - Anticipated to gain market share with a significant shift towards SiPh technology, expected to reach a mix of ~40% in 2026 [32][69]. - Target price raised to Rmb321 based on a 20x FY26E PE, reflecting strong growth in 800G and 1.6T segments [33][70]. - Earnings estimates for FY25-27E increased by 5-25% due to higher revenue and margin assumptions [33][43]. 3. **Suzhou TFC Optical Communication (300394.SZ)** - Focus on acquiring new customers for passive optical components and optical engines, with a target price raised to Rmb140 [46]. - Earnings estimates for FY25-27E increased by 6-42% based on stronger demand for 1.6T optical engines [46][48]. Core Insights and Arguments - The ASIC uptrend is expected to solidify the demand for optical transceivers, particularly with the increasing attach rate per AI accelerators [2][11]. - The market has underappreciated the gross/net margin upside from SiPh migration, which is projected to drive sector EPS revisions and re-ratings [2][3][11]. - Despite concerns about AI capex peaking, the inference AI capex is viewed as more sustainable, supporting ongoing growth in the optical networking sector [3][28]. - Customer preference remains strong for Tier-1 suppliers like Innolight and Eoptolink, which is expected to enhance their market positions [40][30]. Additional Important Points - The anticipated adoption of LPO (Linear-drive Pluggable Optics) for scaling up ASICs could represent an upside risk for long-term demand [12]. - The actual mass adoption of co-packaged optics (CPO) is not expected until 2029-2030, allowing for continued strength in the 800G market in 2026 [30]. - The gross margin for 800G/1.6T SiPh transceivers is expected to reach 45-50%, driven by ongoing technology upgrades and increased industry penetration [30][31]. Conclusion - The optical networking industry is poised for significant growth driven by technological advancements in ASIC and SiPh, with leading companies like Innolight and Eoptolink positioned to capitalize on these trends. The market's current valuation may not fully reflect the potential earnings growth, presenting a favorable investment opportunity.

Industry Impact - Trump's latest tariff overhaul is squeezing US bitcoin miners [1] - ASIC imports are shifting abroad due to the tariff changes [1] Regulatory Concerns - Industry expert warns about the impact of the tariff overhaul [1]

Summary of the Conference Call for Aojie Technology Industry Overview - Aojie Technology holds a leading position in the cellular IoT module market, with a market share of nearly 50% for Cat 1 products, benefiting from an expanded product matrix and increased downstream applications, alongside industry growth trends [2][6] - The Chinese baseband chip market is approximately $20 billion, accounting for 30% of the global market, indicating high elasticity [11] Company Insights - The core team of Aojie Technology has extensive experience in high-tech industries, with Chairman Dai Baojia being a founder of Ridi Technology, and team members from Marvel, providing a solid foundation for technological innovation and business expansion [2][8] - Aojie Technology has made significant advancements in baseband and customized chip sectors, particularly with the 3D stacked cloud chip project, showcasing the capabilities of Chinese engineers [3][4] Product Development and Market Position - Aojie Technology's product offerings in the IoT market include Cat 1, Cat 4, and Cat M modules, with Cat 1 accounting for 47% of global IoT module shipments [5] - The company has successfully introduced its 4G quad-core mobile chip to the Latin American market and plans to showcase its octa-core smart chip SR8,666X at MWC2025, with expectations of achieving a brand breakthrough in Q3 [11][12] Competitive Landscape - In the Cater four market, Qualcomm and Aojie are the main competitors, with Qualcomm dominating overseas markets while Aojie focuses on China and developing markets, gradually achieving domestic substitution [9] - Aojie Technology's competitive edge in the cellular IoT module market is supported by a broad product matrix, large-scale production capabilities, and cost advantages [6] Future Expectations - The company anticipates significant progress in its mobile baseband chip sector, with expectations of achieving a brand breakthrough in Q3 and further advancements in Q4 or early next year [12] - Aojie Technology's ASIC capabilities are crucial for the domestic computing power industry, with a strong emphasis on customized solutions [15] Technological Innovations - The 3D DRAM stacking technology is highlighted as a key path in the AI era, offering advantages such as high bandwidth, low power consumption, and flexibility, which are essential for meeting the demands of cloud computing [17] - Aojie Technology is leveraging its expertise in 3D DRAM technology to create compliant solutions that align with the procurement needs of domestic CSP manufacturers, indicating a broad market potential [17] Conclusion - Aojie Technology is well-positioned in the rapidly evolving IoT and mobile chip markets, with a strong focus on innovation, competitive advantages, and a clear growth trajectory, making it an attractive investment opportunity [12][18]

Core Viewpoint - The article discusses the recent developments in the AI chip market, particularly focusing on AMD and NVIDIA's efforts to re-enter the Chinese market and the competitive landscape they face from domestic suppliers and ASIC chip manufacturers [1][2]. Group 1: Market Developments - NVIDIA's announcement to resume sales of its AI chip H20 in China and the introduction of a new GPU for the Chinese market has positively impacted its stock price, which rose by 4.04% [1]. - AMD's MI308 product is under review for export to China, with potential losses of $800 million due to previous export restrictions [2]. - In 2024, sales from China accounted for over 24% of AMD's total revenue, highlighting the importance of the Chinese market for both companies [2]. Group 2: Competitive Landscape - The competition in the AI chip market is intensifying, with domestic Chinese suppliers increasing their market share. IDC data indicates that domestic AI chip market share rose to over 40% in the first half of the year, while NVIDIA's share was 65.2% last year [2][3]. - TrendForce predicts that the market share of foreign AI server chips will decrease from 63% last year to 41.5% by 2025, indicating a shift towards domestic suppliers [3]. - Domestic cloud providers are currently relying on NVIDIA chips but are also exploring self-developed ASIC chips as alternatives [3]. Group 3: Strategic Initiatives - NVIDIA is actively promoting the construction of "AI factories" globally, targeting industrial enterprises to expand its market presence [4]. - AMD is working to integrate its chips into data centers of major AI firms like OpenAI and Meta, while also collaborating with governments and research institutions to build high-performance computing and AI infrastructure [4].

Summary of Conference Call Records Industry Overview - The conference call discusses the optical module and related industries, particularly focusing on the demand for optical modules, PCBs, and other network-side products driven by significant growth in TOKEN users and Annual Recurring Revenue (ARR) [1][2][4]. Key Insights and Arguments - **Growth in AI Monetization**: From April to June, TOKEN usage increased four to five times compared to the beginning of the year, indicating a substantial rise in inference computing demand, which benefits the related supply chain [1][5]. - **Impact of Tariffs**: Current optical module products are not affected by U.S. tariffs, but potential 232 industry taxes could be a concern. A 25% tax increase would have limited impact on customers due to existing tariffs [4][12]. - **Market Performance**: North American cloud service providers reported better-than-expected revenue, and major model companies are seeing significant growth in TOKEN users and ARR, which is driving demand for optical modules and PCBs [2][5]. - **Future Demand Projections**: By 2025, demand for computing power will increasingly be driven by inference rather than training, with ASICs expected to account for a larger share of the market, significantly increasing the value of optical modules and PCBs compared to GPUs [1][6][8]. - **AI Revenue Growth**: AI-related revenues are projected to reach hundreds of billions to a trillion dollars by the end of the year, with companies like OpenAI and Anthropic showing rapid ARR growth [1][6][7]. Additional Important Points - **Market Valuation**: Current market valuations for leading companies are relatively low, not fully reflecting the growth potential of optical modules and PCBs. A PE ratio of at least 20 is suggested for these sectors [3][12][15]. - **Technological Developments**: The introduction of new technologies, such as the 1.6T optical module, is expected to accelerate growth in the industry, with significant releases anticipated in the coming quarters [5][14]. - **Cost Efficiency**: ASICs have a lower cost per Flops compared to GPUs, with a cost efficiency ratio of 2-3 times better, leading to a wide range of cost fluctuations [9][10]. - **Long-term Growth Predictions**: By 2026, the growth rate for optical modules and PCBs is expected to reach 20-30%, driven by the increasing share of ASICs in the market [10][11]. Conclusion - The optical module and PCB industries are poised for significant growth driven by advancements in AI, increased demand for inference computing, and favorable market conditions. The current market undervaluation presents potential investment opportunities, with a strong outlook for the coming years.

Group 1 - The article highlights the sectors that will benefit from the "anti-involution" policy [1] - The successful delivery of the first NVIDIA GB300 NVL72 is expected to lead to a peak season for the supply chain, with analysts optimistic about the growth in Q2 and Q3 due to the rapid ramp-up of NVIDIA Blackwell and the strong development of ASICs [1] - A total of 388 institutional research reports were published today, with SF Express receiving an upgraded rating and 15 companies getting initial coverage, including four stocks like Ruijie Networks receiving in-depth coverage from New Fortune analysts [1] Group 2 - In the individual stock institutional attention ranking, Xinhecheng made its debut, with the top five being Xiaogoods City, Xinhecheng, Taotao Automotive, Kidswant, and Zijin Mining [1]

Core Viewpoint - The PCB concept stocks are experiencing significant activity, with companies like Yihau New Materials, Huadian Co., and Jingwang Electronics reaching new highs, indicating a bullish trend in the market [1] Group 1: Market Dynamics - The M8 PCB market space for AI servers and switches is projected to be between 50-60 billion by 2026, according to Tianfeng Securities [1] - Major global manufacturers are intensifying their infrastructure development to meet the surging demand for inference computing power, which is driving the growth of the PCB industry [1] Group 2: Company Performance - Yihau New Materials (301176), Huadian Co. (002463), and Jingwang Electronics (603228) have all achieved new record highs in stock performance [1] - Jin'an Guoji (002636) has seen a consecutive two-day stock increase, while Aoshikang (002913) and Zhongyi Technology (301150) are also experiencing upward trends [1] Group 3: Investment Trends - There is a notable influx of dark pool funds into these stocks, suggesting increased investor interest and confidence in the PCB sector [1]

Core Viewpoint - The article discusses the increasing competition in the AI chip market, particularly how major tech companies like Google and Meta are accelerating their development of custom chips to reduce reliance on Nvidia's GPUs, with predictions that ASIC shipments will surpass Nvidia's AI GPU shipments as early as next year [1][2]. Group 1: Market Dynamics - Nvidia has historically dominated the AI chip market, holding over 80% market share in AI servers, while ASIC-based servers currently account for only 8% to 11% [2][4]. - Google is expected to ship between 1.5 million to 2 million of its self-developed AI chips (TPUs) this year, while Amazon's AWS is projected to ship 1.4 million to 1.5 million ASICs, bringing their combined shipments close to half of Nvidia's estimated annual GPU shipments of 5 million to 6 million [2][4]. Group 2: Cost Efficiency - The key advantage driving tech giants to develop their own chips is the reduction in Total Cost of Ownership (TCO), with ASICs potentially saving 30% to 50% in TCO compared to GPUs [3]. - Google claims its TPUs can deliver three times the performance of Nvidia GPUs per unit of energy consumed, highlighting the efficiency of custom chips [3]. Group 3: Competitive Landscape - Meta is focusing on launching its new high-performance ASIC chip "MTIA T-V1" in Q4 of this year, aiming to outperform Nvidia's next-generation AI GPU "Rubin" [5]. - Despite ambitious plans, Meta faces production challenges due to limited advanced packaging capacity from TSMC, which can only provide 300,000 to 400,000 units, creating a bottleneck [5]. Group 4: Nvidia's Response - In response to the competitive threat, Nvidia has opened its proprietary "NVIDIA NVLink" communication protocol to facilitate integration with other companies' CPUs or ASICs, aiming to retain its major clients [6]. - Nvidia's established software ecosystem, CUDA, remains a significant barrier for competitors, as it allows AI developers to efficiently build and deploy applications, maintaining Nvidia's competitive edge [6].