公众号记得加星标⭐️，第一时间看推送不会错过。 来源 ：本文编译自SEMI 。 细分市场增长 预计逻辑与微电子领域将引领设备扩张，2026 年至 2028 年期间总投资额将达到 1750 亿美元。预计 代工厂将成为这一增长的主要驱动力，而这主要得益于 2 纳米以下工艺的产能扩张。关键推动因素包 括环绕栅极 (GAA) 架构和背面供电等先进技术，这些技术对于提升芯片性能和功耗效率，以应对日 益苛刻的人工智能工作负载至关重要。更先进的 1.4 纳米工艺技术预计将于 2028-2029 年实现量 产。此外，人工智能性能的提升预计将推动边缘设备（包括汽车电子、物联网应用和机器人技术）的 大幅增长。除了先进工艺之外，所有节点和各种电子设备的需求预计将大幅增长，从而推动成熟工艺 设备的投资。 预计内存细分领域将位居第二，三年支出总额达 1360 亿美元，标志着该细分领域开启新一轮增长周 期。预计 2026 年至 2028 年，DRAM 相关设备投资将超过 790 亿美元，其中 3D NAND 投资将达 到 560 亿美元。人工智能训练和推理推动了各类内存需求的全面增长。人工智能训练需要更大的数 据传输带宽和极低的延迟，显 ...

Core Insights - The article discusses the rapid growth of High Bandwidth Memory (HBM) driven by the increasing demand for artificial intelligence (AI) and the acceleration of GPU development by companies like NVIDIA [1][2][5] - HBM is a high-end memory technology that is difficult to implement, and customization is crucial for its continued benefit from the widespread application of GPUs and accelerators [1][2] Market Trends - According to Dell'Oro Group, the server and storage components market is expected to grow by 62% year-over-year by Q1 2025 due to the surge in demand for HBM, accelerators, and network interface cards (NiC) [1] - AI server sales have increased from 20% to approximately 60% of the total market, significantly boosting GPU performance and HBM capacity [2] Competitive Landscape - SK Hynix leads the HBM market with a 64% sales share, followed by Samsung and Micron [1][2] - Micron plans to begin mass production of the next-generation HBM4 with a 2048-bit interface in 2026, expecting a 50% quarter-over-quarter revenue growth in HBM by Q3 FY2025, reaching an annual revenue of $6 billion [2] Technological Challenges - The demand for HBM is increasing rapidly, with manufacturers facing challenges due to the accelerated release cycles of GPU technologies, which are now updated every 2 to 2.5 years compared to the traditional 4 to 5 years for standard memory technologies [3][4] - The complexity of HBM5 architecture poses challenges for standardization and widespread adoption, as it requires a balance between high memory bandwidth and increased capacity for next-generation AI and computing hardware solutions [5][6] Future Developments - Marvell Technology is collaborating with major HBM suppliers to develop a custom HBM computing architecture, expected to be released in the second half of 2024, which will integrate advanced 2.5D packaging technology and custom interfaces for AI accelerators [4][6] - The HBM memory bandwidth and I/O count are expected to double with each generation, necessitating innovative packaging technologies to accommodate the increased density and complexity [4][6]

Core Viewpoint - SK Hynix has significantly improved its financial stability due to explosive growth in high bandwidth memory (HBM) sales, leading to a substantial increase in cash and cash equivalents while reducing debt [1][2]. Financial Performance - As of the end of the year, SK Hynix's cash and cash equivalents reached 16.9623 trillion KRW, an increase of 2.8060 trillion KRW from the previous year-end [1]. - The cash and cash equivalents grew by over 75% compared to the same period last year, which was 9.6680 trillion KRW [1]. - The company's sales revenue for the first half of the year reached 39.8711 trillion KRW, a 38% increase year-on-year [1]. - Operating profit surged to 16.6534 trillion KRW, marking a 99% increase compared to the previous year [1]. - Cash generated from operating activities amounted to 21.7281 trillion KRW, an 86% increase from 11.6824 trillion KRW in the same period last year [1]. Debt Management - Despite executing 11.2490 trillion KRW in capital investments, nearly double the 5.9670 trillion KRW from the same period last year, SK Hynix managed to reduce its debt significantly [2]. - As of mid-year, the company's debt stood at 21.8410 trillion KRW, down by 842.7 billion KRW from the previous year-end [2]. - Compared to the same period last year, debt decreased by 3.3869 trillion KRW from 25.2279 trillion KRW [2].

Group 1 - The core viewpoint is that the rapid development of artificial intelligence (AI) is driving unprecedented demand for specialized hardware, with a significant increase in hardware investment expected due to the explosive growth in AI model usage [2] - Despite concerns such as competition from China and product delays, the fundamental reality is that AI applications will see substantial growth, benefiting companies like NVIDIA, which derives a significant portion of its revenue from AI applications [2] - The two companies dominating the core AI processor market are NVIDIA and Broadcom, with AI-driven advantages for companies like Broadcom and Marvell expected to come primarily from networking rather than just processor products [2] Group 2 - Marvell's optical business is anticipated to perform strongly, with the analyst acknowledging previous underperformance but highlighting significant opportunities in the optical sector [3] - The acceleration of inference workloads is a key driver of computing demand, indicating sustained AI spending until 2026, supported by new product cycles such as NVIDIA's Blackwell [3] - There are substantial opportunities in memory and storage, particularly in high bandwidth memory (HBM), which is specifically designed for AI, with companies like Micron gaining market share in this competitive field [3]

Core Viewpoint - OpenAI aims to deploy over 1 million GPUs by the end of this year, significantly increasing its computational power and solidifying its position as the largest AI computing consumer globally [2][4]. Group 1: GPU Deployment and Market Impact - Sam Altman announced that OpenAI plans to launch over 1 million GPUs, which is five times the capacity of xAI's Grok 4 model that operates on approximately 200,000 Nvidia H100 GPUs [2]. - The estimated cost for 100 million GPUs is around $3 trillion, comparable to the GDP of the UK, highlighting the immense financial and infrastructural challenges involved [5]. - OpenAI's current data center in Texas is the largest single facility globally, consuming about 300 megawatts of power, with expectations to reach 1 gigawatt by mid-2026 [5][6]. Group 2: Strategic Partnerships and Infrastructure - OpenAI is not solely reliant on Nvidia hardware; it has partnered with Oracle to build its own data centers and is exploring Google's TPU accelerators to diversify its computing stack [6]. - The rapid pace of development in AI infrastructure is evident, as a company with 10,000 GPUs was considered a heavyweight just a year ago, while 1 million GPUs now seems like a stepping stone to even larger goals [6][7]. Group 3: Future Vision and Challenges - Altman's vision extends beyond current resources, focusing on future possibilities and the need for breakthroughs in manufacturing, energy efficiency, and cost to make the 100 million GPU goal feasible [7]. - The ambitious target of 1 million GPUs by the end of the year is seen as a catalyst for establishing a new baseline in AI infrastructure, which is becoming increasingly diverse [7].

Core Viewpoint - Micron Technology plans to invest a total of $200 billion in domestic semiconductor manufacturing and R&D, aiming to strengthen the U.S. semiconductor industry and create approximately 90,000 jobs [1][2]. Group 1: Investment Plans - Micron will invest $150 billion in domestic memory manufacturing and $50 billion in research and development [1]. - An additional $30 billion will be allocated for the construction of a second memory manufacturing facility in Boise, Idaho, and for the expansion and modernization of the existing facility in Manassas, Virginia [1]. - The overall expansion strategy includes building two high-volume wafer fabs in Idaho and up to four fabs in New York [1]. Group 2: Production Timeline - Micron's first wafer fab in Idaho is expected to start producing DRAM chips by 2027, with the second fab enhancing DRAM output to meet growing market demand [2]. - Site preparation for the New York super fab is anticipated to begin later this year, pending environmental approvals [2]. Group 3: Strategic Goals - The investment aligns with Micron's goal of producing 40% of its DRAM domestically, which is crucial for various sectors including personal computing, automotive, and artificial intelligence [1][2]. - The introduction of advanced HBM packaging capabilities in the U.S. is part of the strategy to support the growth of high-bandwidth memory essential for AI [2]. Group 4: Government Support and Funding - Micron expects to qualify for the Advanced Manufacturing Investment Credit (AMIC) and has already received $275 million in direct funding from the CHIPS Act for its Manassas facility expansion [3]. - The company anticipates receiving up to $6.4 billion in additional direct funding from the CHIPS Act to support the construction of these facilities [3]. Group 5: Workforce Development - Micron has committed to investing $325 million to develop the next generation of workforce to support the U.S. semiconductor industry, including initiatives for semiconductor curriculum development and partnerships with educational institutions [3].

Core Viewpoint - Micron Technology plans to invest a total of $200 billion in domestic semiconductor manufacturing and R&D, aiming to strengthen the U.S. semiconductor industry and create approximately 90,000 jobs [1][3][4]. Group 1: Investment Plans - Micron will invest $150 billion in domestic memory manufacturing and $50 billion in research and development [1]. - The additional $30 billion investment will be used to build a second memory manufacturing facility in Boise, Idaho, and to expand and modernize the existing facility in Manassas, Virginia [1]. - The overall expansion strategy includes constructing two high-volume wafer fabs in Idaho and up to four fabs in New York [1]. Group 2: Production Timeline - Micron's first wafer fab in Idaho is expected to start producing DRAM chips by 2027, with the second fab enhancing DRAM output to meet growing market demand [2]. - Site preparation for the New York super fab is anticipated to begin later this year, pending environmental approvals [2]. Group 3: Strategic Importance - The investment aligns with Micron's goal of producing 40% of DRAM domestically, which is crucial for various sectors including personal computing, automotive, industrial operations, wireless communications, and artificial intelligence [1][4]. - The U.S. Secretary of Commerce highlighted that Micron's plans will help restore memory chip production in the U.S. and maintain leadership in critical industries such as AI, aerospace, and defense [4]. Group 4: Workforce Development - Micron has committed to investing $325 million to develop the next generation of workforce, supporting semiconductor curriculum development and partnerships with universities and community colleges [4].

Core Viewpoint - SK Siltron's sales to SK Hynix surpassed those to Samsung Electronics for the first time in Q1 2023, attributed to SK Hynix's success in high bandwidth memory (HBM) and server DRAM, while Samsung faced challenges [1][2]. Group 1: Sales Performance - In Q1 2023, SK Siltron sold wafers worth 124.4 billion KRW to Samsung (Company A) and 128.8 billion KRW to SK Hynix (Company B), marking a 27% decline in sales to Samsung and a 32% increase to SK Hynix compared to the previous year [1]. - This shift is significant as it marks the first time since SK Hynix's acquisition by SK Group in 2017 that sales to SK Hynix exceeded those to Samsung [1]. Group 2: Market Dynamics - The decline in Samsung's sales is attributed to weakened product competitiveness and a reduction in foundry capacity due to poor customer orders, leading to a 50% decrease in capital investment compared to the previous year [2]. - Conversely, SK Hynix has been maximizing production capacity for its 10nm fifth-generation process and has seen strong sales for products like HBM3E, while also investing in new factories [2]. Group 3: SiC Business Challenges - SK Siltron's SiC wafer subsidiary reported Q1 sales of 6.147 billion KRW with an operating loss of 63.4 billion KRW, a significant decline from the previous year's sales of 21.3 billion KRW and an increase in losses from 26.3 billion KRW [3]. - The SiC business, seen as a next-generation power semiconductor opportunity, is facing challenges in technology competitiveness and market conditions, leading to speculation that it may become a liability for SK Group [3].