证券代码：300693 证券简称：盛弘股份 编号：2025-013 深圳市盛弘电气股份有限公司 投资者关系活动记录表 | 随着智算中心单机柜功率密度的提升，供电系统优化的 | | | --- | --- | | 关注点将转向更高电压等级，以解决导体截面积过大，更多 | | | 的材料和更大的体积的问题。同时，电压等级的升高也会带 | | | 来远距离输电的优势，这将会引发数据中心设计架构上颠覆 | | | 性的变革。传统的交流400V | UPS及直流的240、336V的HVDC虽 | | 然已经过多年的市场考验，已证明其技术可行性，市场接受 | | | 度较高，产业链较为成熟。但对于高密度、高效率的智算中 | | | 心而言，由于智算服务器之间连接的光缆已经占用过多机柜 | | | 的走线空间，探索更高电压的应用可以减少电源线占用的空 | | | 间，带来潜在的性能提升和成本节约，这也更符合可持续发 | | | 展的要求。 | | | 新一代的800V | HVDC，相比传统交流供电以及之前低电压 | | 的直流供电，可以提供系统效率能效提升；占地空间节约； | | | 铜耗降低等等特点，凭借这些优势，在海 ...

Group 1 - The article discusses the limitations of Variational Autoencoders (VAE) in the diffusion model paradigm and explores the potential of using pretrained semantic encoders to enhance diffusion processes [1][7][8] - The shift from VAE to pretrained semantic encoders like DINO and MAE aims to address issues such as semantic entanglement, computational efficiency, and the disconnection between generative and perceptual tasks [9][10][11] - RAE and SVG are two approaches that prioritize semantic representation over compression, leveraging the strong prior knowledge from pretrained visual models to improve efficiency and generative quality [10][11] Group 2 - The article highlights the trend of moving from static image generation to more complex multimodal content, indicating that the traditional VAE + diffusion framework is becoming a bottleneck for next-generation generative models [8][9] - The computational burden of VAE is significant, with examples showing that the VAE encoder in Stable Diffusion 2.1 requires 135.59 GFLOPs, surpassing the 86.37 GFLOPs needed for the core diffusion U-Net network [8][9] - The discussion includes the implications of the "lazy and rich" business principle in the AI era, suggesting a shift in value from knowledge storage to "anti-consensus" thinking among human experts [3]

Investment Rating - The investment rating for the company is "Outperform the Market" [5][24]. Core Views - The company has shown steady operating performance in the first three quarters, with revenue reaching 6.132 billion yuan, a year-on-year increase of 20.39%, and a net profit of 704 million yuan, up 15.57% year-on-year. However, impairment losses have affected profit growth [8][19]. - The company is experiencing a recovery in domestic delivery and maintaining rapid growth in external business. In the first half of 2025, revenue from grid automation was 1.726 billion yuan, up 2.21% year-on-year, while revenue from power plant and industrial automation reached 2.003 billion yuan, a 31.25% increase year-on-year [19][20]. - The company is accelerating its overseas expansion, achieving significant breakthroughs in multiple countries, including Thailand, Malaysia, South Korea, and Indonesia, and winning SVG projects in Laos, Congo, and India [20]. - The company has a leading position in solid-state transformer technology, with multiple key projects delivered. The efficiency of its solid-state transformer products has been improved to 98.5% through several iterations [20][22]. Financial Performance and Forecast - The company is expected to achieve net profits of 828 million yuan, 1.005 billion yuan, and 1.205 billion yuan for the years 2025, 2026, and 2027, respectively, representing year-on-year growth rates of 16%, 21%, and 20% [3][24]. - The projected revenue for the company is 8.15 billion yuan in 2025, with a growth rate of 17.3% compared to the previous year [4][26]. - Key financial metrics include a projected PE ratio of 28 for 2025, a net profit margin of 11.0%, and a return on equity (ROE) of 17.7% [4][26].

Summary of Sifang Co., Ltd. Conference Call Company Overview - **Company**: Sifang Co., Ltd. - **Industry**: Power and Energy Solutions Key Points Business Performance - In the first three quarters of 2025, Sifang Co. achieved a new contract signing growth of approximately 20% year-on-year, with a target of 10 billion new contracts for the year [2][5][6] - The revenue growth rate reached over 30% in Q3 2025, with net profit growth exceeding 20% [3] - The gross profit margin has slightly declined due to changes in business structure, but overall profitability remains stable [3] Segment Performance - **Grid Automation**: Revenue growth of about 15% year-on-year [7] - **Power Plant and Industrial Automation**: Revenue growth of approximately 25% [7] - **New Energy**: Revenue growth of 40%-50%, driven by demand for booster stations [2][7] - **International Business**: New orders reached 410 million yuan, a significant increase from 150 million yuan in the same period last year [6] Strategic Focus - The company emphasizes the importance of grid transformation and safety, predicting continued growth in grid investment [4][10] - Data center business is a strategic priority, with expectations for commercialization of medium-voltage direct current distribution or SST (Solid State Transformer) by 2027 [4][11] - The company aims for international business to account for 30% of total revenue by 2030, focusing on Southeast Asia, the Middle East, Europe, and South America [4][29] Product Development - SST is viewed as a critical strategic layout, with significant potential in medium-voltage direct current distribution [8][17] - The company is developing distributed phase-shifting devices and static synchronous compensators, with expected revenue growth exceeding 100 million yuan [14] - The company has made breakthroughs in offshore wind power projects and digital twin technology in large base projects [14] Market Trends - The demand for distributed phase-shifting devices is expected to grow, with an estimated market of around 200 units in 2025 [19] - The company is adapting to different market demands, with variations in voltage requirements between domestic and international markets [24] International Strategy - The company has successfully localized its operations, enhancing competitiveness through local teams and partnerships [15][27] - The gross margin for international business is generally higher than domestic, particularly in primary systems [16] Future Outlook - The company is optimistic about the growth of the new energy sector, with a focus on the integration of renewable energy into data centers [21][28] - The storage business is expected to grow significantly, although specific targets for 2026 are still under planning [22][25] Challenges and Considerations - The company acknowledges the need for continuous improvement in core technologies related to SST applications in data centers [23] - There are ongoing considerations regarding the integration of high-voltage cascading storage solutions and their market acceptance [30][31] Conclusion Sifang Co., Ltd. is positioned for robust growth in the power and energy sector, with strategic focuses on international expansion, innovative product development, and adapting to market demands. The company is optimistic about future opportunities, particularly in new energy and data center applications.

Core Insights - The article discusses the transition from Variational Autoencoders (VAE) to a new model called SVG developed by Tsinghua University and Kuaishou's Keling team, which shows significant improvements in training efficiency and generation speed [1][3]. Group 1: Model Comparison - SVG achieves a 62-fold increase in training efficiency and a 35-fold increase in generation speed compared to traditional VAE methods [1]. - The main issue with VAE is semantic entanglement, where features from different categories are mixed, leading to inefficiencies in training and generation processes [3][5]. - The RAE model focuses solely on generation performance by reusing pre-trained encoders, while SVG aims for both generation and multi-task applicability through a dual-branch feature space [5][6]. Group 2: Technical Innovations - SVG utilizes the DINOv3 pre-trained model for semantic extraction, which effectively captures high-level semantic information, addressing the semantic entanglement issue [8]. - A lightweight residual encoder is added to DINOv3 to recover high-frequency details that are often lost, ensuring a comprehensive feature representation [8]. - The distribution alignment mechanism is crucial for matching the output of the residual encoder with the semantic features from DINOv3, significantly enhancing image generation quality [9]. Group 3: Performance Metrics - Experimental results indicate that removing the distribution alignment mechanism leads to a significant drop in image generation quality, as measured by the FID score [9]. - In training efficiency, the SVG-XL model achieves an FID score of 6.57 after 80 epochs, outperforming the VAE-based SiT-XL model, which has an FID of 22.58 [11]. - The SVG model's feature space can be directly applied to various tasks such as image classification and semantic segmentation without the need for fine-tuning, achieving competitive accuracy metrics [13].

Core Viewpoint - The article discusses the transition from Variational Autoencoders (VAE) to new models like SVG developed by Tsinghua University and Kuaishou, highlighting significant improvements in training efficiency and generation speed, as well as addressing the limitations of VAE in semantic entanglement [1][4][10]. Group 1: VAE Limitations and New Approaches - VAE is being abandoned due to its semantic entanglement issue, where adjusting one feature affects others, complicating the generation process [4][8]. - The SVG model achieves a 62-fold improvement in training efficiency and a 35-fold increase in generation speed compared to traditional methods [3][10]. - The RAE approach focuses solely on enhancing generation performance by reusing pre-trained encoders, while SVG aims for multi-task versatility by constructing a feature space that integrates semantics and details [11][12]. Group 2: SVG Model Details - SVG utilizes the DINOv3 pre-trained model for semantic extraction, effectively distinguishing features of different categories like cats and dogs, thus resolving semantic entanglement [14]. - A lightweight residual encoder is added to capture high-frequency details that DINOv3 may overlook, ensuring a comprehensive feature representation [14]. - The distribution alignment mechanism is crucial for maintaining the integrity of semantic structures while integrating detail features, as evidenced by a significant increase in FID values when this mechanism is removed [15][16]. Group 3: Performance Metrics - In experiments, SVG outperformed traditional VAE models in various metrics, achieving a FID score of 6.57 on the ImageNet dataset after 80 epochs, compared to 22.58 for the VAE-based SiT-XL [18]. - The model's efficiency is further demonstrated with a FID score dropping to 1.92 after 1400 epochs, nearing the performance of top-tier generative models [18]. - SVG's feature space is versatile, allowing for direct application in tasks like image classification and semantic segmentation without the need for fine-tuning, achieving an 81.8% Top-1 accuracy on ImageNet-1K [22].

Core Insights - The article discusses the limitations of traditional Variational Autoencoder (VAE) in training diffusion models, highlighting issues such as low representation quality and efficiency [2][4][8] - A new framework called SVG (Self-supervised representation for Visual Generation) is proposed, which integrates pre-trained visual feature encoders to enhance representation quality and efficiency [3][12] Limitations of Traditional VAE - VAE's latent space suffers from semantic entanglement, leading to inefficiencies in training and inference [4][6] - The entangled features require more training steps for the diffusion model to learn data distribution, resulting in slower performance [6][8] SVG Framework - SVG combines a frozen DINOv3 encoder, a lightweight residual encoder, and a decoder to create a unified feature space with strong semantic structure and detail recovery [12][13] - The framework allows for high-dimensional training directly in the SVG feature space, which has shown to be stable and efficient [16][22] Performance Metrics - SVG-XL outperforms traditional models in generation quality and efficiency, achieving a gFID of 6.57 in just 80 epochs compared to SiT-XL's 1400 epochs [18][22] - The model demonstrates superior few-step inference performance, with a gFID of 12.26 at 5 sampling steps [22] Multi-task Generalization - The latent space of SVG inherits the beneficial properties of DINOv3, making it suitable for various tasks such as classification and segmentation without additional fine-tuning [23][24] - The unified feature space enhances adaptability across multiple visual tasks [24] Qualitative Analysis - SVG exhibits smooth interpolation and editability, outperforming traditional VAE in generating intermediate results during linear interpolation [26][30] Conclusion - The core value of SVG lies in its combination of self-supervised features and residual details, proving the feasibility of sharing a unified latent space for generation, understanding, and perception [28] - This approach addresses the efficiency and generalization issues of traditional LDMs and provides new insights for future visual model development [28]

Core Viewpoint - The company is focusing on the AIDC (Artificial Intelligence Data Center) sector, aiming to develop new products and solutions to meet emerging demands in the energy supply field [1] Group 1: Company Strategy - The company established the AIDC division in June 2025 to enhance its product offerings and address new market needs [1] - The company plans to leverage its existing technology to develop new power supply products such as HVDC (High Voltage Direct Current) and SST (Solid State Transformer) [1] - The company aims to accelerate product development through in-depth market research and active communication with downstream customers [1] Group 2: Long-term Vision - The company aspires to become a comprehensive energy solution provider for AIDC, focusing on continuous innovation in product efficiency, stability, and intelligence [1] - The company intends to expand its business scope to provide overall energy solutions for data centers and intelligent computing centers [1] - The company hopes to enhance its brand influence and market competitiveness in the AIDC sector [1]

Core Viewpoint - The company is actively developing its capabilities in AIDC (Automatic Identification and Data Capture) and aims to become a comprehensive energy solution provider in this field by 2025 [1] Group 1: Company Developments - The company established an AIDC division in June 2025 to focus on research and development of new products to meet emerging demands [1] - The company plans to leverage its existing technology in power products to explore new directions in AIDC power supply, such as HVDC (High Voltage Direct Current) and SST (Solid State Transformer) [1] Group 2: Market Strategy - The company is conducting in-depth market research and engaging with downstream customers to accelerate product development and implementation [1] - The long-term goal is to enhance the company's brand influence and market competitiveness in the AIDC sector by continuously innovating in product offerings and expanding into energy solutions for data centers and intelligent computing centers [1]

Core Insights - The report by Ge Wu Zhi Sheng indicates that Shenzhen Shenghong Electric Co., Ltd. (referred to as "Shenghong Co." or "盛弘股份") is a leading player in the low-voltage power quality market, driven by technological innovation and deep industry engagement [1] Market Overview - The overall market size for low-voltage power quality in China is projected to reach 12.8 billion yuan in 2024, with expectations to grow to 15.28 billion yuan by 2028, driven by the dual trends of "scale expansion + structural upgrade" [1] - Key growth areas identified include industrial projects, data centers, and new energy generation [1] Competitive Landscape - Shenghong Co. is the only company in the industry with revenue from low-voltage power quality products exceeding 500 million yuan, solidifying its position in the first tier of the market [1] - The company has established itself as a preferred brand in high-end application fields such as semiconductors, data centers, and new energy generation [2] Product Development - Key products such as SVG, APF, and AVC are experiencing significant growth and are replacing traditional reactive power compensation devices [2] - Shenghong Co.'s products cover critical sectors including semiconductors, new energy generation, petrochemicals, hospitals, shipping, and rail transportation, demonstrating a strong presence across various application scenarios [2] Technological Advancements - Silicon carbide (SiC) technology is emerging as the latest trend in low-voltage power quality technology development in China [3] - Shenghong Co. has successfully developed and scaled SiC power quality products, achieving a 2% efficiency improvement and a 50% reduction in size compared to traditional silicon-based solutions [3] - The company is positioned as a preferred partner for national and provincial key projects, focusing on technological research and development to support the transition to a high-quality new power system [3]