电科数字 20251211 摘要 博飞电子参与国家级太空算力验证项目，提供全国产化解决方案，包括 高性能计算卡和 AI 芯片卡，价值量约 100 万元，目前正与某科研院所 和星网进行验收，目标应用于低空操作和太空算力。 博飞电子为星网提供高性能计算卡、AI 芯片卡和射频卡，这些板卡通过 高速通信架构连接，构成 AI 训练服务器，用于处理太空中的算力需求。 核心竞争力在于高安全性和高可靠性的设计。 博飞电子使用的 AI 芯片（包括 CPU 和 GPU）均为国内主流厂商产品， 已在多个行业验证应用，旨在确保项目的稳定性和可控性，目标是将国 产 AI 芯片应用于太空。 电科数字控股子公司博飞电子聚焦低轨道领域，大股东 32 所负责高空 应用及火箭相关业务，提供核心软硬件产品，如硬件计算机、控制软件 及仿真测试等，保障火箭发射近 300 次。 电科数字全资子公司华信网络获得华为升腾整机钻石经销商资质认证， 具备 384 节点全系列生成产品的销售和服务资格，预计显著提升收入水 平，初期毛利率高于原有水平。 Q&A 博飞电子在商业航天领域的主要产品和应用是什么？ 能否详细介绍一下这三款产品的功能？ 这三款产品可以理解为地 ...

Core Viewpoint - Goldman Sachs has downgraded its forecast for rack-level AI server shipments, projecting a decline in expected volumes for 2025 and 2026 due to product transition impacts and supply-demand uncertainties [2][4]. Group 1: AI Server Market Outlook - Goldman Sachs expects AI training servers to remain the main growth driver in the market, but the growth rate is anticipated to be lower than previously expected due to factors such as product transition, production complexity challenges, demand variability, and tariff risks [7]. - The forecast for rack-level AI server shipments has been revised down to 19,000 units in 2025 and 57,000 units in 2026, with market sizes adjusted to $54 billion and $156 billion respectively [8]. Group 2: Impact on Supply Chain Companies - Goldman Sachs has lowered the target prices for several Taiwanese AI server supply chain companies, including Quanta, with reductions ranging from 7% to 21% [3][11]. - The downgrade reflects a shift from rapid growth to more rational expansion in the AI server industry, indicating that while growth is slowing, AI infrastructure investment remains a key growth driver in the tech sector [11]. Group 3: Performance of Different Server Types - High-performance AI servers are not expected to be completely replaced by rack-level solutions, as some customers prefer motherboard solutions for design flexibility [5]. - AI inference servers are projected to see sales growth of 41% and 39% in 2025 and 2026, respectively, driven by expanding application areas [12].

Core Viewpoint - Goldman Sachs has downgraded its forecast for rack-level AI server shipments, indicating a slowdown in industry growth due to product transition impacts and supply-demand uncertainties [1][3][8]. Group 1: Shipment Forecast Adjustments - The forecast for rack-level AI server shipments in 2025 and 2026 has been revised down from 31,000 and 66,000 units to 19,000 and 57,000 units, respectively [1]. - The revenue forecast for AI training servers has also been adjusted, with expected growth of 30% in 2025 to reach $160 billion and 63% in 2026 to reach $260 billion, down from previous estimates of $179 billion and $248 billion [3][5]. Group 2: Factors Influencing Adjustments - The slowdown in shipments is attributed to several factors, including the transition period for GPU platforms, production complexity challenges, demand variability due to new AI models, and tariff risks affecting ODM manufacturers [4][5]. - The production complexity of full rack systems adds uncertainty to capacity ramp-up, while the release of more efficient AI models raises questions about market demand for intensive computing capabilities [4]. Group 3: Impact on Supply Chain Companies - Goldman Sachs has lowered target prices for several Taiwanese ODM and cooling supply chain companies, including Quanta, Foxconn, FII, Wistron, AVC, and Auras, with reductions ranging from 7% to 21% [1][7]. - Quanta's rating has been downgraded from "Buy" to "Neutral" due to limited upside potential in the current market environment [7]. Group 4: Market Dynamics - The market is transitioning from a phase of rapid growth to more rational expansion, reflecting a shift in the AI server industry [8]. - Despite the slowdown, investment in AI infrastructure remains a key growth driver for the technology sector, although growth will be more moderate than previously expected due to various limiting factors [8].