公众号记得加星标⭐️，第一时间看推送不会错过。 以下为发改委发布全文： 2025年9月至今，受需求"爆发式"增长、产能"断崖式"紧缺等因素影响，全球存储器市场缺口扩大， 存储芯片价格持续上涨，近1个月多以来，涨幅呈现扩大态势，建议关注存储芯片对下游价格的影 响。 一、近期存储芯片价格大幅快速上涨 调研反映，截至今年1月，存储芯片两大主要产品DRAM和NAND闪存价格均创2016年有数据以来最 高。以主流型号为例，1月DARM（DDR4 8Gb 1G*8）合约平均价格为11.5美元，比上月上涨约 24%，比2025年9月上涨约83%；NAND闪存（128Gb 16G*8 MLC）合约平均价格为9.5美元，比上 月上涨约65%，比2025年9月上涨近1.5倍。 （来源 ：综合自发改委检测中心 ） *免责声明：本文由作者原创。文章内容系作者个人观点，半导体行业观察转载仅为了传达一种不同的观点，不代表半导体行业观察对该 观点赞同或支持，如果有任何异议，欢迎联系半导体行业观察。 2月28日，国家发展改革委价格监测中心发布一篇名为《存储芯片价格持续上涨并向下游传导》的文 称，2025年9月份至今，受需求"爆发式"增长、产 ...

Core Viewpoint - TSMC is undergoing a significant global expansion, driven by geopolitical factors, AI computing power dominance, and technological advancements, with a projected revenue exceeding 3.8 trillion NTD (approximately 850 billion RMB) in 2025 [2] Group 1: Global Expansion and Investments - TSMC has secured a total of 151.42 billion NTD in government subsidies from various countries for 2024 and 2025, indicating strong international support for its expansion efforts [2] - The company is establishing a comprehensive manufacturing ecosystem in the U.S. with plans for six wafer fabs, two advanced packaging plants, and one R&D center, with an investment of up to 165 billion USD [4] - TSMC's Arizona facility has already achieved profitability in 2025, generating 16.14 billion NTD, marking a turnaround from previous losses [4][5] Group 2: Regional Developments - In Japan, TSMC's Kumamoto plant is transitioning to produce 3nm chips, reflecting a strategic shift to meet increasing demand in the automotive and AI sectors [6][7] - The Dresden plant in Germany is focused on 28/22nm technologies, with a production capacity of 40,000 300mm wafers per month, catering to local automotive and industrial clients [8] - TSMC's operations in mainland China focus on mature and mid-range nodes, with the Shanghai and Nanjing fabs supporting the local electronics industry [9] Group 3: Technological Leadership - TSMC is advancing its most cutting-edge technologies in Taiwan, with plans for ten new fabs, including facilities for 2nm and 1.4nm processes, reinforcing its position as a leader in semiconductor manufacturing [10][12] - The company is also expanding its CoWoS packaging capabilities in Taiwan to meet the growing demand for AI chips, aiming to increase monthly production capacity to over 125,000 units by the end of 2026 [11] Group 4: Strategic Rationale - TSMC's aggressive expansion is driven by the exponential growth in AI demand, which is seen as a foundational shift rather than a temporary trend [12] - The company is diversifying its manufacturing locations to mitigate geopolitical risks, transforming from a concentrated "national treasure" to a globally distributed "capacity oasis" [13] - TSMC maintains a competitive edge through high yield rates and strong ecosystem ties, ensuring its dominance in advanced process technologies [14]

Core Insights - GeForce 3, released in February 2001, marked a pivotal moment in GPU history as the first truly programmable GPU supporting DirectX 8.0 pixel and vertex shaders, allowing graphics programmers to write programs that run on the GPU [4][11] - Prior to GeForce 3, most GPUs were fixed-function accelerators, requiring the CPU to handle all special graphics effects, which limited the capabilities of early graphics processing [3][4] - GeForce 3's architecture included a "light-speed memory architecture" that significantly improved effective memory bandwidth, providing advantages at higher resolutions despite similar rasterization performance to its predecessor, GeForce 2 Pro [6][8] Product Evolution - The introduction of GeForce 3 Ti500 in 2001 addressed some performance shortcomings and established the "Ti" suffix, which continues to be used in NVIDIA's graphics cards today [7] - The original Xbox, released in late 2001, solidified GeForce 3's role as a foundational technology for future gaming consoles, showcasing NVIDIA's contributions beyond just graphics to include audio hardware and memory controllers [8] - Although GeForce 3 was not an immediate commercial success, it laid the groundwork for the more popular GeForce 4 series, which enhanced Direct3D support and introduced new features like 3D textures and improved shader capabilities [9][10] Long-term Impact - The evolution from GeForce 3 to the GeForce 8 series, which introduced the Tesla microarchitecture and unified shader design, reflects a significant shift towards fully programmable GPUs, enabling broader applications beyond gaming [10][11] - The rise of general-purpose GPU (GPGPU) computing, initially used for scientific calculations and later for cryptography and AI, can be traced back to the programmability established by GeForce 3 [10][11] - GeForce 3's legacy is evident in the current dominance of NVIDIA in the AI data center market, demonstrating how early innovations in gaming graphics have had far-reaching implications across various industries [11]

Core Viewpoint - Micron Technology is facing legal challenges from local and national groups regarding its semiconductor manufacturing facility in Clay, New York, primarily focused on environmental concerns and the adequacy of public input in the approval process [2][4][5]. Group 1: Project Overview - Micron is constructing a semiconductor manufacturing facility, which is expected to be the largest in New York's history, with potential public subsidies amounting to $25 billion, including $6.1 billion from the federal CHIPS Act and $5.5 billion from New York State [2]. - The project aims to create 9,000 jobs and is part of a broader initiative to revitalize domestic semiconductor manufacturing, which has been deemed critical for national security [2][3]. Group 2: Environmental Concerns - Local residents and advocacy groups are concerned about the environmental impact of the facility, citing the significant water and energy consumption, as well as the generation of hazardous waste associated with semiconductor manufacturing [3][6]. - The project is expected to draw 48 million gallons of water daily from Lake Ontario, sufficient to supply over 585,000 households, raising concerns about local infrastructure strain [6]. Group 3: Legal Actions and Community Response - Lawsuits have been filed against Micron and state agencies, arguing that the environmental review process was rushed and did not adequately consider public opinion [4][5]. - Advocates are pushing for a legally binding community benefits agreement to ensure that the project provides tangible benefits to the local community, including job creation and environmental protections [8]. Group 4: Company Commitments and Community Engagement - Micron has committed to investing hundreds of millions in education, workforce training, and affordable housing over the next 20 years, and claims that 80% of the construction workforce will be local [4][8]. - The company has also pledged to develop new wetlands to compensate for those that will be destroyed during construction [4]. Group 5: Historical Context and Future Implications - There is skepticism among community members regarding the actual benefits of the project, referencing past industrial projects that failed to deliver on environmental promises [7][8]. - The ongoing legal and community challenges highlight the tension between economic development and environmental stewardship in the semiconductor industry [5][6].

Core Viewpoint - Rapidus aims to start mass production of 2nm chips in the second half of 2027, with significant backing from the Japanese government and private sector, indicating a strong commitment to advancing Japan's semiconductor industry [2][3][4]. Group 1: Company Overview - Rapidus is collaborating with over 60 companies, primarily focusing on high-performance computing (HPC), AI semiconductors, and robotics, with around 10 companies already receiving quotes [2]. - The company has secured approximately 267.6 billion yen in funding, with the Japanese government contributing 100 billion yen, making it the largest shareholder [2][3]. Group 2: Future Plans - Rapidus plans to expand its investment to over 7 trillion yen, targeting profitability by 2030 and an IPO around 2031 [4]. - The company intends to construct a second factory in Chitose, Hokkaido, starting in 2027, which will produce 1.4nm chips by 2029, aiming to compete with TSMC [4]. Group 3: Government Involvement - The Japanese government will hold approximately 40% of shares through its investment, with provisions to convert non-voting shares into voting shares under certain conditions [3]. - The government plans to invest an additional 150 billion yen by 2026, potentially increasing its voting power to 60% [3].

Core Insights - The report by Morgan Stanley highlights that the optical communication market is entering a second phase driven by artificial intelligence, with a projected market size of approximately $30 billion by 2025 and over $65 billion by 2028, reflecting a compound annual growth rate of about 30% [2] - An additional $23 billion in opportunities related to new optical communication technologies could expand the long-term market size to around $90 billion [2] Market Size and Structural Demand Drivers - The demand for optical fiber is shifting from traditional telecom and enterprise cycles to cloud-based AI data centers, which require higher bandwidth and faster upgrade cycles [5] - The share of cloud data centers in the overall optical fiber market is expected to grow from 28% in 2021 to 63% by 2025, reaching 78% by 2030 [5] - The report identifies two structural drivers: the limitations of power and the need for high-density deployments, which will catalyze the construction of distributed data centers [6] Market Segmentation and Growth Opportunities - By 2028, the optical transceiver market is expected to reach approximately $50.3 billion, making it the largest sub-market within optical communications [10] - The report predicts about $23 billion in new optical module market opportunities, segmented into five categories, with significant contributions from copper-to-fiber transitions and co-packaged optics [10][11] Technology Transition: Copper to Fiber - The transition from copper to fiber is characterized by short-distance connections, with copper historically dominating due to lower costs and higher reliability [12] - The report notes that the performance of copper is limited by signal integrity at high speeds, necessitating a shift to fiber [12] - The cost of fiber installation is at least double that of copper, highlighting the economic challenges of this transition [13] Co-Packaged Optics (CPO) and Optical I/O - CPO is seen as a key driver in the transition from copper to fiber, addressing bottlenecks in packaging technology [15] - The adoption of CPO faces resistance due to reliability concerns, manufacturing costs, and the need for a standardized ecosystem [16] Optical Circuit Switching (OCS) - OCS aims to reduce latency and power consumption by keeping traffic in the optical domain, with a projected market size growth from $19.8 million in 2024 to $5.7 billion by 2028 [20] - The economic sustainability of OCS depends on competition from advancements in packet switching technology and the predictability of traffic patterns [21] Data Center Interconnect (DCI) and ZR Optical Modules - DCI demand is driven by power limitations, with a shift towards pluggable optical modules due to cost and power advantages [27] - The report emphasizes that pluggable optical modules represent about 20% of system costs, with average selling prices ranging from $2,000 to $3,000 [27][28] Out-of-Band Management (OOBM) - OOBM is identified as a significant cost, potentially accounting for 2-5% of the Ethernet switch market, with a projected market opportunity of $3-4 billion by 2028 [25] - The report highlights the collaboration between Ciena and Meta on a PON-based redesign that reduces active network components and lowers power consumption [25]

Core Viewpoint - The current state of 5G technology is disappointing, with limited improvements over 4G, and the control of 5G networks remains largely with Ericsson and Nokia, despite efforts for more open standards [2][3] Group 1: 5G and Open RAN - The Open RAN movement aimed to replace proprietary interfaces with industry-standard ones but has not significantly impacted market dynamics [2] - The focus has shifted towards open-source initiatives, with the U.S. Department of Defense collaborating with the Linux Foundation on a project called OCUDU to integrate open-source code into 6G networks [2][3] - Nvidia is promoting the idea of "open" and "open-source" as part of its 6G project, which includes major industry players [3] Group 2: Challenges in Open Source - The term "open" in telecommunications is often misused, and open-source software is frequently confused with "free" software [5] - Ericsson and Nokia's profits largely come from technology licensing, which contradicts open-source principles, hindering smaller companies from innovating on these platforms [6] - The current systems supporting 5G services are primarily proprietary, limiting flexibility for smaller developers [6] Group 3: Nvidia's Aerial Platform - Nvidia has developed an open-source RAN reference platform called Aerial, allowing developers to integrate AI-native waveforms [7] - However, Aerial requires developers to work primarily within Nvidia's CUDA framework, which limits deployment on other CPU architectures [7] Group 4: Hardware Dependency and AI Integration - There are concerns about whether open-source software can fully address hardware dependency issues in RAN [8] - Nvidia's preference is for using GPUs for Layer 1 functions, while Ericsson is adapting its software to run on Nvidia's Grace CPU [10] - The integration of GPUs into RAN is seen as costly due to their high energy consumption, despite their performance advantages [11] Group 5: Industry Dynamics and Future of 6G - The participation of Ericsson and Nokia in OCUDU raises questions about their motivations, especially as they face challenges from potential new competitors in an open-source 6G environment [13] - The evolution of AI will significantly impact 6G, and premature standardization could limit future functionalities [14]

Core Viewpoint - Two-dimensional (2D) semiconductors have been researched as complementary materials to silicon transistors, promising smaller, faster, and more energy-efficient processors. However, recent findings indicate that the common testing methods may exaggerate the performance of these transistors, making it difficult to assess their true potential for commercial applications [2][5][6]. Group 1: Research Findings - A study from Duke University reveals that the prevalent "back-gate" testing structure used in evaluating 2D semiconductors can lead to inflated performance metrics due to a phenomenon known as "contact gate control" [2][5]. - The research emphasizes that most high-performance 2D transistor designs are incompatible with commercial technology, thus complicating the evaluation of these materials for future transistor applications [2][6]. - The study highlights that as device sizes shrink, contact resistance becomes a dominant factor in overall performance, making the understanding of contact characteristics increasingly important [6][7]. Group 2: Experimental Methodology - Researchers typically use a simplified "back-gate" structure for testing, where all components are fabricated on the same silicon wafer, allowing for streamlined preparation and rapid experimentation [5]. - The study introduces a new symmetrical double-gate transistor structure that enables direct measurement of the impact of contact gate control on performance, allowing for a more accurate comparison [6][7]. - Results show that in larger devices, contact gate control can enhance performance by approximately twofold, and in devices scaled down to future technology requirements, this effect can increase performance by up to six times under specific conditions [6][7]. Group 3: Future Directions - The research team plans to further reduce device sizes, aiming for a contact length of 15 nanometers, and to explore alternative contact metals to lower contact resistance [7]. - The overarching goal is to establish clearer design specifications for integrating 2D semiconductors into future transistor technologies, acknowledging the need for honest assessments of how device structures influence measurement results [7].

Core Viewpoint - Kioxia Holdings has raised its full-year performance forecast significantly, indicating a transition from recovery to a new growth phase driven by AI demand, with projected sales of 2.1797 trillion to 2.2697 trillion yen (up 28% to 33% year-on-year) and operating profit of 709.5 billion to 799.5 billion yen (up 57% to 77% year-on-year) [2] Group 1: Company Performance and Market Position - Kioxia is the only large storage manufacturer in Japan, focusing solely on NAND flash memory, and holds a 15.3% share of the global NAND market, ranking third after Samsung and SK Hynix [3][4] - The company's stock price has surged dramatically, increasing approximately 15 times from its re-listing price of 1,455 yen to over 20,000 yen, with its market capitalization exceeding 10 trillion yen [2] Group 2: Historical Context and Challenges - Kioxia's history traces back to Toshiba, which faced a crisis in the mid-2010s, leading to the sale of its profitable storage business, which was rebranded as Kioxia in 2019 [4] - The company faced significant challenges prior to its re-listing, including a severe downturn in the semiconductor market, leading to substantial losses in the 2023 fiscal year [5] Group 3: Market Dynamics and AI Demand - The shift in AI market demand from training to inference has created a surge in the need for enterprise SSDs, which are based on NAND technology, positioning Kioxia favorably as competitors focus on HBM investments [6][8] - Kioxia's delayed production of its K2 plant has turned into a strategic advantage, allowing it to capitalize on the increased demand for NAND products as competitors struggle with supply [8] Group 4: Technological Advancements - Kioxia's competitive edge lies in its unique technology, particularly in the development of 3D NAND, where it emphasizes horizontal miniaturization over merely increasing stacking layers, achieving high capacity at lower costs [9][10] - The company is also exploring next-generation memory technologies, such as SCM and MRAM, aiming to re-enter the DRAM market with innovative solutions like OCTRAM, which promises lower power consumption and high capacity [11][12] Group 5: Future Outlook and Strategic Focus - Kioxia's rise in the semiconductor sector demonstrates the potential for Japanese companies to remain competitive globally by adapting to market changes and investing in future technologies [14] - The company must build a robust financial structure to withstand market fluctuations and seize future growth opportunities while navigating the risks associated with the semiconductor industry's cyclical nature [13][14]

Core Viewpoint - The demand for computing power is experiencing exponential growth due to the iteration of AI large models, while the bottleneck of high-speed interconnection in data centers is becoming a key constraint for AI performance breakthroughs [2] Group 1: Industry Trends - The traditional electrical signal transmission is limited by energy consumption and distance, making it difficult to support the massive data flow required for AI model training [2] - The industry consensus indicates that 2026 will be a critical turning point for the large-scale commercialization of silicon photonics technology, recognized as the first year of silicon photonics chip commercialization [2] - The shipment volume of 800G and 1.6T optical modules is expected to double significantly by 2026, with silicon photonics technology penetration in this market projected to reach 50%-70% [3] Group 2: Company Developments - Tower Semiconductor is actively expanding its silicon photonics foundry capacity, planning to double its manufacturing capacity by 2025 and further expand in mid-2026 [4] - Tower's CEO stated that the company is in a leading position in the silicon germanium and silicon photonics technology fields, with significant growth potential in revenue and profit due to strong demand from data centers [6] - Tower's Q4 2025 revenue reached a record high of $440 million, a 14% year-over-year increase, with a net profit of $80 million, exceeding market expectations [7] Group 3: Competitive Landscape - GlobalFoundries announced the acquisition of Advanced Micro Foundry in Singapore, becoming the largest pure silicon photonics chip foundry by revenue [15] - GlobalFoundries plans to utilize AMF's 200mm platform to meet the demand for long-distance optical communication and expand to a 300mm platform as market demand grows [16] - UMC is accelerating its layout in silicon photonics by signing a technology licensing agreement with imec to obtain advanced silicon photonics processes [21][22] Group 4: Technological Innovations - TSMC is developing a compact universal photonic engine (COUPE) technology to support the explosive growth of data transmission driven by AI [24][26] - TSMC's silicon photonics strategy revolves around three key platforms, aiming for significant reductions in power consumption and latency compared to copper interconnects [28] - Intel has been a pioneer in commercializing silicon photonics technology, having shipped over 8 million EICs, and continues to maintain its strategic layout in data center interconnects [32] Group 5: Market Outlook - The silicon photonics market is expected to see rapid growth, with the number of optical chips for 100 GbE and higher projected to increase from 36.6 million in 2024 to 80.5 million by 2029 [39] - The demand for silicon photonics chips is anticipated to grow the fastest, from 9.6 million in 2024 to 45.5 million by 2029 [41] - The competition in the silicon photonics foundry market is intensifying, with major players like GlobalFoundries, Tower, TSMC, UMC, Samsung, Intel, and ST all making significant investments and strategic moves [41]