Core Insights - The four largest hyperscalers plan to invest over $650 billion in AI infrastructure this year, indicating a significant growth opportunity in the AI sector [1] Group 1: Chip and Memory Makers - Nvidia is the primary supplier of GPUs for AI workloads and maintains a competitive edge with its CUDA software platform, which is essential for AI training [2] - Advanced Micro Devices (AMD) has established a niche in AI inference and has secured significant partnerships with OpenAI and Meta Platforms [2] - Broadcom is capitalizing on the demand for custom AI ASICs and has collaborated with Alphabet to develop Tensor Processing Units (TPUs) for data center infrastructure [3] - Taiwan Semiconductor Manufacturing (TSMC) holds a virtual monopoly on advanced logic AI chip manufacturing, providing it with substantial pricing power [4] - Micron Technology is positioned well in the AI infrastructure boom due to its production of high bandwidth memory (HBM), which is in short supply and driving up prices [5] Group 2: Central Processing Units (CPUs) - AMD is recognized as a leader in the CPU market, which is becoming increasingly vital with the rise of agentic AI [6] - Other potential beneficiaries in the CPU space include Arm Holdings and Intel [6]

Vi e w p o i n t | 26 Feb 2026 15:23:27 ET │ 17 pages China CCL Sector Rapid Growth Momentum on AI-CCL Materials Upgrade CITI'S TAKE During NVDA's 4QFY26 results call, we believe the company painted a rosy picture for increased usage of high-end CCL materials. Fig 1 exhibits the rapid development of CCL materials roadmap from Ampere (A100) in 2020 to Blackwell with a full-year ramp in 2026 and then the next Rubin (Vera Rubin) from 2H26. This high demand of material upgrade should enable advanced CCL supplie ...

Global Technology February 24, 2026 10:16 AM GMT M Global Foundation Global Technology Webcast - NVDA, Semiconductor Capex, Equipment, Memory Join technology analysts Shawn Kim, Mason Wayne, Shane Brett, Suzune Tamura, and Nigel van Putten to discuss Nvidia set-up into results, recent developments in global semiconductor capex, equipment, memory industries and global technology stock ideas. Morgan Stanley & Co. International plc+ | Shawn Kim | | | --- | --- | | Equity Analyst | | | Shawn.Kim@morganstanley.c ...

The rapid advancement of generative AI and large language models is forcing a fundamental redesign of data-center computing architectures. In 2024, NVIDIA introduced the Blackwell GB200 NVL72, enabling a single rack to house 72 GPUs and 36 Grace CPUs, interconnected via NVLink 6 and Quantum X800 InfiniBand / Spectrum X Ethernet to deliver 400 Gb/s scale-out networking. SEMIVISION GTC 2026 Outlook: How NVIDIA Is Redefining AI Infrastructure with LPX, CPO, and Rubin Original Article By SemiVision Research [Re ...

洁净室深度报告 20260227 摘要 国内洁净室市场规模约 2000 亿元，电子领域占比过半，预计未来增速 维持在 7%左右。技术升级推动洁净室等级提升，单位造价显著增加， 2023 年达 0.66 万元/平方米。 洁净室行业竞争格局分层明显，高端市场由少数企业主导，中高端市场 企业数量较多，低端市场竞争激烈。头部企业凭借高可靠性、长期合作 关系和高端项目经验构筑护城河，市场集中度较高。 电子洁净室是集成电路制造的关键基础设施，对空气洁净度、气流组织、 温湿度与静电控制要求严格，集成电路核心制造环节均需在 1~3 级洁净 环境中完成。 全球半导体销售额进入高速增长期，AI 是核心驱动力。台积电预计 2030 年全球半导体市场规模将突破 1 万亿美元，其中 AI 终端市场占比 45%。 AI 芯片市场的结构性增量由逻辑芯片与存储芯片双轮驱动，预计 2026 年逻辑芯片与存储芯片占比分别为 43%与 32%。英伟达在 AI 芯片市场 占据主导地位。 Q&A 洁净室行业的定义、系统构成、分类及核心应用场景分别是什么？ 洁净室是先进制造业的基础性工程，伴随制造业转型升级而发展。洁净室建设 以洁净室为核心载体，整合空 ...

英伟达的下一个 Mellanox 针对 Agentic AI 底时延的 Groq LPU20260228 英伟达或将 Groq 的技术与工程团队吸收，将其 IP 融合进后续产品，而 非以独立产品线形式销售，此策略与 Mellanox 的收购路径一致，旨在 强化其在高性能计算领域的护城河。 Groq LPU 架构主要针对推理侧对极低时延的需求，尤其适用于 batch size=1 的场景，通过片上 SRAM 和确定性时序控制，实现更稳定的低 时延推理表现，与 GPU 在训练和较大 batch size 推理上的优势互补。 LPU 与英伟达 GPU 的整合预计至少需要 18-24 个月，可能在 Finman 那一代产品中实现，更可能采用 chiplet 封装级集成，通过混合键合和 TSV 技术实现 compute die 与 LPU die 之间更低时延的数据交互。 Finman compute die 可能采用台积电 A16 节点，而 LPU die 则可能 选择更成熟的 3nm 或 4nm 节点，以便更容易实现更高的 SRAM 配置 密度和 3D/堆叠式集成，作为第一代方案先行验证落地。 LPU 能力更可能整 ...

Accessible version Global Memory Tech Weekly theme: Singapore investors, upbeat DRAM spot vs NAND, SK Square, Wonik Industry Overview Singapore investor meeting and virtual tour takeaways We met investors in Singapore to discuss memory and Korea tech this week. Some investors alluded to their short / underweighted position (NAND, backend equipment and hardware stocks) but our positive views on DRAM and Korea tech were well discussed during the meetings. Investor feedback on our valuation shift from P/B to P ...

高盛周末宏观电话会：宏观股市观点、AI 采用信号、油价 地缘政治溢价、新兴市场韧性及全球其他地区跑赢大市 20260228 摘要 特朗普政府关税政策调整频繁，虽声称统一加征 15%关税，但实际执行 可能采取差异化安排，对不同国家和产品实施不同税率，实际关税税率 影响差异仅约 0.1 个百分点。 关税退款问题进展较快，预计退款金额约 1,800 亿美元，但发放速度和 执行周期仍存在不确定性。新关税面临法律挑战，但因 122 条款关税期 限限制，司法程序可能无法在期限内完成裁决。 中期选举中，众议院民主党优势明显，但参议院竞争激烈，民主党面临 艰巨挑战。关注得克萨斯州初选结果，可能影响参议院席位归属。 消费者端 AI 采用率快速提升，已广泛融入日常生活，使用频率大幅提高， 推动算力需求爆炸式增长。企业端 AI 推进缓慢，规模化落地面临数据和 系统层面的障碍。 英伟达财报超预期，但股价未明显上涨，投资者关注 2027 年超大规模 资本开支的能见度、非传统客户的持续支出路径以及技术领先性再确认。 Q&A 近期关税问题重回头条，叠加国情咨文演讲，华盛顿当前最关注的议题是什么， 需要重点关注哪些方面？ 关税问题不确定性仍 ...

27 February 2026 | 1:54PM EST Equity Research Americas Technology: Hardware: Expert Network Series: Data Center Strategy for Hyperscalers and AI We hosted an investor meeting with Mark Monroe, former Principal Engineer of Data Center Technology at Microsoft, on February 27th, 2026 as part of our Expert Network Series to discuss current trends impacting the data center and AI infrastructure market. Bottom line: (1) Hyperscalers continue to prefer self-build to best optimize for cost and control, but accelera ...

3 月 GTC 大会热点展望 - 掘金 AIDC 系列电话会议 20260301 摘要 GPU 功率提升推动三次电源架构变革，模块化和 IVR 方案成潜在趋势。 传统分立式方案面临器件堆叠、布线损耗等瓶颈，模块化供电通过立体 PCB 集成电感、电容等，节省空间并提升耐流，IVR 方案则通过芯片内 部集成电源，降低电流需求，GTC 大会或将催化相关技术落地。 三次电源模块化升级对 PCB 工艺提出更高要求，需在设计阶段协同电源 设计，并引入新材料和压合工艺。中富电路已进入 MPS、伟创力等三次 电源供应链，威尔高也在推进。电源 PCB 市场规模相对较小，但与客户 配合迭代设计至关重要，较早布局者有望脱颖而出。 英伟达 Rubin Ultra 代际基本确定导入 HVC（800V）方案，以降低电 流和损耗。麦格米特已在英伟达供应链内，配合进行产品开发与预研， 进度领先大陆厂商。欧陆通等谷歌链电源厂商以及铂科新材等电感材料 企业也有望受益。 GTC 大会重点展示 Quantum 3,400 以及以太网 6,800、6,810 三款 CPO 交换机，量产时间预计在 2026 年底至 2027 年初。台积电良率提 升至 ...