行业研究丨深度报告丨软件与服务 [Table_Title] AI 沉思录(一)：从智驾看 AI Agent 落地范式 %% %% %% %% research.95579.com 1 丨证券研究报告丨 软件与服务 cjzqdt11111 报告要点 [Table_Summary] 本篇报告以当前行业的几个非共识观点引入，通过梳理复盘最先以 AI Agent 模式落地的智驾场 景在不同阶段下产业发生的变化，进一步来探讨，当前如何观察研判 AI Agent 在不同阶段、不 同行业的落地拐点，以及如何把握不同阶段下的投资机会。本轮技术周期下的应用爆发将逐步 拉开序幕，我们认为当前阶段应当关注数据、推理云/芯片和算力优化加速方向的"铲子股"， 以及具有明确产品思路+能够保障投入的玩家，下一阶段寻找率先兑现 ROI 或流程替代方向的 公司，长期来看更加关注自身护城河的构建。 分析师及联系人 [Table_Author] 宗建树 SAC：S0490520030004 SFC：BUX668 请阅读最后评级说明和重要声明 2 / 37 %% %% %% %% research.95579.com 2 [Table_Title2 ...

证券研究报告 速腾聚创 (2498 HK) 港股通 Q2 毛利率继续提升，EMX 有望快速 放量 2025 年 8 月 24 日│中国香港 其他电子 速腾聚创公布 25 半年财报：激光雷达出货量 26.68 万颗，其中 ADAS 交付 量 22.05 万颗，机器人交付 4.63 万颗。实现收入 7.83 亿元，同比增长 7.72%。 值得关注的是，1）公司 Q2 单季度收入 4.6 亿元，同比增速提升至+24.4%， 小鹏、问界等客户转向影响开始逐步减弱；2）Q2 单季度毛利率实现 27.7%， 创过去三年新高，ADAS 毛利率达到 19.4%；3）公司新一代产品 EMX 即 将于 Q3 量产，采用全新芯片化收发架构，下半年开始有望持续放量。我们 看好公司受益智驾和机器人带来的激光雷达需求，以及具身智能业务顺利推 进，Q4 有望实现单季度盈亏平衡。目标价 48.17 港币，维持"买入"评级。 ADAS 业务：EMX 取得更多市场订单，Q3 末有望迈入大规模交付 2Q25 公司实现激光雷达出货量 12.4 万颗（去年同期 11.8 万颗），公司受小 鹏（转向纯视觉）、问界（转向华为方案）等不利影响较 Q1 大幅 ...

Investment Rating - The investment rating for the company is "Buy" and is maintained [5]. Core Views - The company focuses on self-research to create cost advantages, resulting in an excellent price-performance ratio, with products covering a price range of 50,000 to 200,000 CNY. For Q2 2025, revenue reached 14.23 billion CNY, with a gross margin of 13.6% and a net profit attributable to shareholders of 160 million CNY. The domestic new car cycle is expected to drive continuous sales growth, while collaboration with Stellantis opens up global sales opportunities [2][4][6]. Summary by Sections Company Performance - In Q2 2025, the company achieved a sales volume of 134,000 units, representing a year-on-year increase of 151.7% and a quarter-on-quarter increase of 53.2%. Revenue for the same period was 14.23 billion CNY, up 165.5% year-on-year and 42.0% quarter-on-quarter. The average revenue per vehicle was 106,000 CNY, with a year-on-year increase of 6,000 CNY [6]. Profitability - The gross margin for Q2 2025 was 13.6%, an increase of 10.9 percentage points year-on-year, while it decreased by 1.2 percentage points quarter-on-quarter. The improvement in gross margin was attributed to scale effects, cost reduction, product structure optimization, and income from other businesses. The net profit attributable to shareholders was 160 million CNY, marking a turnaround from loss, with a profit margin of 1.1% [6]. Future Outlook - The company is expected to sell 620,000, 910,000, and 1,210,000 vehicles in 2025, 2026, and 2027, respectively. The company plans to launch 2-3 new products globally each year, with the B10 and B01 already launched in 2025. The company aims to achieve localized manufacturing in Europe by 2026 [4][6].

招商证券指出，根据乘联会数据观察，7月淡季特征明显，受6月车企冲刺半年销量后有透支，叠加部分 地区有补贴暂时中断影响。但上周五中央再次下放国补第三批资金680亿，预计8月下旬市场有望回暖， 逐步向旺季过渡。关于汽车产业链投资策略，该机构看好下半年有爆款，销量增长强劲公司，尤其是中 报预期较好公司：a）吉利汽车：下半年销量及业绩强劲，预计中报业绩正面，当前估值低；b）小鹏汽 车：即将推出重磅新车全新一代小鹏P7，产品线向上；c）理想汽车：市场悲观预期已充分，关注i8销 售策略调整后或加快订单转化以及9月走量车i6上市；d）小米汽车、零跑：下半年新车带动效应突出， 产品终端需求强劲。 8月7日午后，港股三大指数集体上涨。盘面上，科网股涨跌不一，苹果概念涨幅居前，创新药概念回 调，新消费概念股分化，泡泡玛特再创历史新高。智驾含量领先的港股通汽车ETF（159323）震荡走 低，持仓股中，京东方精电、天齐锂业、东风集团股份、金力永磁、舜宇光学科技、赣锋锂业等涨幅居 前，理想汽车、地平线机器人、亿华通等跌幅居前。 公开信息显示，港股通汽车ETF（159323）标的指数为港股通汽车指数（931239.CSI）。该指数高度 ...

Core Viewpoint - The article highlights the performance of various electric vehicle manufacturers in July, noting a decline in deliveries for Li Auto and NIO, while XPeng achieved a record high in sales [3]. Group 1: Sales Performance - Li Auto's delivery volume decreased by 39.7% year-on-year and 15.3% month-on-month to 30,731 units [3] - XPeng's delivery volume increased by 229.4% year-on-year and 6.1% month-on-month to 36,717 units [3] - NIO's delivery volume increased by 2.5% year-on-year but decreased by 15.7% month-on-month to 21,017 units [3] Group 2: New Model Launches - Li Auto launched its second pure electric model, the i8, priced between 321,800 to 369,800 yuan, featuring the VLA driver model [4] - The L90 model from the LeDao brand was launched with a price range of 265,800 to 299,800 yuan, highlighting its spaciousness and family-friendly features [4] Group 3: Delivery Timelines and Policies - Tesla's delivery timelines for the domestic Model 3 and Model Y remain stable, with limited-time subsidies available until August 31 [5] - Li Auto's delivery timelines for the L6 remain at 1-3 weeks, while the L8 and L9 have extended timelines [5] - XPeng's delivery timelines for various models have shortened, indicating improved supply chain efficiency [5] Group 4: Industry Trends - The issuance of new smart connected vehicle operation licenses indicates a potential turning point for the commercialization of Robotaxi services [6] - The industry may shift from price competition to technology upgrades and cost reductions, influenced by the "anti-involution" strategy [6]

Core Viewpoint - On July 30, the A-share market experienced fluctuations with the three major indices showing mixed results, while stock ETFs saw a net inflow of 7.5 billion yuan, indicating investor interest in certain sectors despite market volatility [2][5]. Summary by Sections ETF Fund Flows - On July 30, stock ETFs had a net inflow of 7.5 billion yuan, with significant inflows into the ChiNext index and Hong Kong-related ETFs, covering sectors such as technology, finance, pharmaceuticals, and the internet [2][5]. - The total number of stock ETFs in the market reached 1,159, with a total scale of 3.83 trillion yuan as of July 30 [4]. Top Performing ETFs - Among the ETFs, 38 saw net inflows exceeding 100 million yuan, with the top three being: - E Fund ChiNext ETF: 1.542 billion yuan - E Fund Hong Kong Securities ETF: 976 million yuan - Fuguo Hong Kong Internet ETF: 811 million yuan [5][6]. - The leading sectors for net inflows included: - Hong Kong Technology: 2.72 billion yuan - ChiNext Index: 1.98 billion yuan - Hong Kong Finance: 1.71 billion yuan - Hong Kong Pharmaceuticals: 1.21 billion yuan - Hong Kong Internet: 990 million yuan [5][6]. Underperforming ETFs - Conversely, 20 stock ETFs experienced net outflows, with notable losses in broad-based and industry ETFs, including: - Four ETFs tracking the CSI 300 Index, with a total outflow exceeding 2.1 billion yuan - Four pharmaceutical-related ETFs, with total outflows over 1 billion yuan - Four CSI A500 ETFs, with outflows nearing 700 million yuan [9]. Market Outlook - Analysts from YinHua Fund and JiaShi Fund expressed optimism regarding the A-share market's resilience due to stable economic fundamentals and supportive policies, anticipating a positive impact on market sentiment [9]. - The Hong Kong market is expected to benefit from valuation recovery and liquidity improvements post-September 2024, with strong performance anticipated in technology and innovative pharmaceutical sectors [9].

Core Viewpoint - The WAIC is expected to further strengthen the trends in intelligent driving and robotics industries [1] Passenger Vehicle Sector - The recent issuance of the third batch of "old-for-new" national subsidies is expected to improve consumer sentiment in the passenger vehicle sector [3] - New models such as the Leapmotor B01 have been released, with attention on the upcoming Li Auto i8, indicating a structural market trend driven by new product expectations and premium brands [3] - The "anti-involution" trend is stabilizing the pricing system of existing models, making new product capabilities and branding crucial for competition [3] Intelligent Driving Sector - The L4 intelligent driving industry is reaching a turning point in costs and technology this year, with ongoing evolution in technology iterations, industry structure, and new business models [4] - The WAIC is anticipated to catalyze developments in the sector, with Shanghai issuing multiple Robotaxi operation licenses [4] Robotics Sector - The robotics sector has shown significant strength since mid-July, driven by strong industrial trends and event catalysts [4] - Key events include the capitalized listings of Zhiyuan and Yushun, bulk orders for humanoid robots from China Mobile, and Tesla's Optimus development updates [4] - The market maintains a high recognition of the long-term industrial trends, with Tesla's Q2 earnings call indicating a positive outlook for the development of the Gen3 prototype by the end of 2025 [4] Commercial Vehicle and Heavy Truck Sector - Q2 performance has significantly benefited from the export growth of heavy trucks and large/mid-sized buses [4] - The acceleration of domestic subsidies since May has strengthened the fundamentals of these two core sectors, with expectations for further growth in Q3 due to a low base [4]

Core Viewpoint - The report highlights a significant shift in public fund allocations, with a notable increase in Hong Kong stock investments and a rebalancing of sector allocations, particularly favoring communication and financial sectors while reducing exposure to traditional consumer and renewable energy stocks [1][2][3]. Group 1: Fund Allocation Trends - Public funds have increased their allocation to Hong Kong stocks, with the proportion rising from 8.66% at the end of 2023 to 19.91% by the end of Q2 2025, marking a new high [2]. - The number of Hong Kong stocks held by actively managed equity funds has increased to 360, with a total market value of 326.5 billion yuan [2]. - The allocation to the ChiNext board has rebounded, increasing by 2.35 percentage points to 18.93%, while the main board has seen a decline for the fourth consecutive quarter, dropping by 2.71 percentage points to 65.39% [3][4]. Group 2: Sector and Industry Focus - The communication sector has seen a significant increase in allocation, with a market value share rising by 2.40 percentage points, while the financial sector also gained traction with increases of 1.13 percentage points for banks and 0.76 percentage points for non-bank financials [5][6]. - The healthcare and financial sectors have become the main battlegrounds for southbound capital, with their market value shares increasing by 6.01 and 2.26 percentage points, respectively [2]. - Notable individual stocks that have gained favor include biotech firms such as Innovent Biologics and 3SBio, while tech giants like Alibaba and Tencent have faced significant reductions in holdings [2][6]. Group 3: Market Dynamics and Sentiment - The market is experiencing a recovery in risk appetite, but there remains a cautious approach, with funds focusing on sectors with stable growth prospects or potential policy improvements [7]. - The consumer sector, particularly food and beverage, has seen significant reductions in holdings, reflecting a broader market concern regarding the recovery of consumer spending [7][8]. - The report indicates that the market's response to tariff policies has become less sensitive, with expectations that risks related to tariffs may not escalate further unless unexpected policy changes occur [8].

Core Viewpoint - The article discusses the results of a driving test conducted by Dongche Di, highlighting that Tesla outperformed Chinese car manufacturers in autonomous driving capabilities despite not using local training data. This has sparked discussions about the state of autonomous driving technology in China and the implications for local car companies [1][16][30]. Summary by Sections Test Overview - The test included various challenging driving scenarios, referred to as "corner cases," which are not commonly encountered by users but are realistic situations that drivers may face [5][11]. - The testing process faced criticism regarding the control variables, but it is noted that achieving perfect standardization in such tests is costly and complex [14][15]. Performance Analysis - Tesla's superior performance is attributed to its foundational data and algorithms developed in the U.S., which are better suited for the simpler driving conditions found on American roads compared to the more chaotic and variable conditions in China [17][22]. - Chinese manufacturers have focused on developing systems for urban driving scenarios, which may have compromised their performance in high-speed situations [18][20]. Implications of the Test - The test results indicate a regression in the autonomous driving capabilities of Chinese car manufacturers, particularly in high-speed scenarios [15][26]. - The article suggests that the algorithms designed for urban driving may not translate well to high-speed driving, leading to hesitation and slower response times in critical situations [20][22]. Official Involvement - The test was not solely initiated by Dongche Di but was influenced by official bodies, including the Ministry of Public Security and the Ministry of Science and Technology, indicating a broader regulatory interest in the safety and efficacy of autonomous driving technologies [28][30]. - The official stance emphasizes the need for stricter management of car manufacturers and clearer communication regarding the limitations of autonomous driving systems to prevent consumer misconceptions [31][32]. Future Considerations - The article posits that this test serves as a national safety warning rather than just an industry evaluation, suggesting that similar assessments may continue to occur to ensure the risks associated with autonomous driving are manageable [33].

Core Viewpoint - The article discusses the recent performance of Tesla in autonomous driving tests in China, highlighting that despite the lack of local training data, Tesla achieved superior results compared to Chinese competitors, sparking debates about the capabilities of domestic car manufacturers [1][18]. Group 1: Testing and Performance - The testing conducted by Dongche Di was scrutinized, with many focusing on minute details rather than the broader implications of the results [3][34]. - The tests revealed that Chinese car manufacturers have experienced a regression in their autonomous driving capabilities, particularly in high-speed scenarios [17][29]. - Tesla's performance is attributed to its foundational model being based on simpler driving conditions in the U.S., allowing it to excel in high-speed environments where reaction time is prioritized over complex decision-making [29][30]. Group 2: Challenges in Domestic Autonomous Driving - Chinese road conditions present unique challenges that require a different approach in the architecture of autonomous driving systems, necessitating continuous adjustments based on real-world data [21][24]. - The article emphasizes the "game theory" capabilities of Chinese autonomous systems, which are designed to handle unpredictable scenarios, but this complexity can lead to hesitation in critical situations [25][27]. - The testing highlighted that while Chinese systems are adept at navigating complex urban environments, they may falter in simpler high-speed contexts due to their advanced decision-making algorithms [29][30]. Group 3: Official Involvement and Industry Implications - The testing initiative was not solely driven by Dongche Di but was influenced by official bodies aiming to regulate and clarify the capabilities of autonomous driving technologies [34][39]. - The involvement of government agencies indicates a shift towards ensuring that autonomous driving technologies are developed responsibly and that consumer trust is managed effectively [38][40]. - The article suggests that this testing serves as a national safety awareness campaign rather than a mere industry evaluation, aiming to educate consumers about the limitations of current autonomous driving technologies [40].