Core Viewpoint - TSMC reported record profits in Q2, driven by strong demand for AI chips, with a projected revenue growth of nearly 30% for the year despite potential tariff impacts [2][3][11]. Financial Performance - TSMC's Q2 revenue reached NT$933.79 billion, a year-on-year increase of 38.6%, with net income of NT$398.27 billion, up 60.7% [2][4]. - Earnings per share (EPS) for Q2 was NT$15.36, reflecting a 60.7% increase compared to the previous year [5][11]. - The gross margin stood at 58.6%, with an operating margin of 49.6%, indicating strong operational efficiency [5]. Market Outlook - TSMC's chairman, C.C. Wei, indicated that AI chip demand remains robust, with non-AI applications also expected to recover moderately [3][11]. - The company has revised its full-year revenue growth forecast from 24-26% to nearly 30%, outperforming market expectations [12][22]. - Despite potential tariff uncertainties, TSMC's overall outlook remains positive, with no significant changes in customer behavior observed [3][4]. Capital Expenditure and Dividends - TSMC plans to maintain its capital expenditure forecast at US$38-42 billion for the year, ensuring sustainable dividend policies [4][13]. - The company anticipates a total cash dividend of at least NT$18 per share this year, increasing to at least NT$20 next year [4]. Advanced Technology and Production Capacity - TSMC is set to begin mass production of its 2nm process technology in the second half of the year, which is expected to provide significant revenue contributions [15][17]. - The company is expanding its manufacturing capabilities in Arizona, with plans for six wafer fabs and two advanced packaging facilities to support AI and HPC applications [18][19]. Industry Context - Analysts view TSMC's strong performance as a positive signal for the semiconductor market, particularly amid tariff and policy challenges [21][22]. - Concerns remain regarding potential impacts from tariffs and the performance of major clients like Apple, which could affect TSMC's outlook for the latter half of the year [22][23].

Core Insights - Super Micro Computer, Inc. (SMCI) has announced the certification of its BigTwin Server for immersion cooling, utilizing 4th and 5th Gen Intel Xeon Scalable Processors, enhancing performance and efficiency in data centers [1][6] Group 1: Certification and Collaboration - The Supermicro BigTwin server has undergone rigorous testing and is now recognized as a certified immersion server, ensuring compatibility with Intel's guidelines and the Open Compute Project (OCP) specifications [1][4] - Supermicro's long-standing collaboration with Intel combines advanced processor technologies with high-performance solutions, ensuring that the BigTwin server remains fully functional when immersed in specified liquids [2][4] Group 2: Efficiency and Environmental Impact - Immersion cooling technology significantly lowers Power Usage Effectiveness (PUE), with data centers using Supermicro immersion servers achieving PUE values close to 1.05 or lower, thus reducing energy costs and environmental impact [3][8] - The immersion cooling method allows for denser compute configurations without increasing thermal load, improving overall energy efficiency and reducing the need for traditional air-based cooling systems [3][5] Group 3: Industry Standards and Future Outlook - Supermicro's involvement in the OCP Community has been crucial in advancing immersion cooling standards, promoting compatibility, efficiency, and scalability across data center deployments [4][5] - The growing demand for immersion-certified servers is driven by the need for efficient cooling solutions in data centers, particularly for AI and HPC applications, which require powerful processors [8]

Core Viewpoint - Cornelis Networks is reintroducing its Omni-Path interconnect technology with the CN5000 series switches and NICs, aiming to compete with Nvidia's InfiniBand technology, particularly in the AI market, by offering higher performance at a lower cost [1][2][7]. Summary by Sections Overview of Omni-Path - Omni-Path was developed by Intel in 2015 as a lossless interconnect technology, primarily for high-performance computing (HPC) applications, and was initially deployed in several supercomputing platforms [1][2]. CN5000 Series Details - The CN5000 series includes switches and NICs that support 400Gbps bandwidth, with the ability to support over 500,000 endpoints in a cluster, and performance that scales almost linearly [2][4]. - The CN5000 switch options include a 1U high, 48-port switch with a total switching capacity of 19.2Tbps and a Director switch with up to 576 ports and a total bandwidth of 230.4Tbps [4][5]. Performance Comparison with InfiniBand - Cornelis claims its system offers up to 2x message transmission rates, 35% lower latency, and 30% faster simulation times compared to Nvidia's 400Gbps Quantum-2 InfiniBand [7]. - However, the CN5000 switch has fewer ports (48) compared to Nvidia's Quantum-2 (64), which may impact scalability in large deployments [7][9]. Scalability and Network Design - To connect 128,000 GPUs at 400Gbps, approximately 13,334 CN5000 switches would be needed, compared to about 10,000 Nvidia switches [9][10]. - The CN5000 Director switch can reduce the number of switches required for large deployments to 733, thus simplifying wiring [10]. Future Developments - Cornelis plans to launch the CN6000 series with 800Gbps support next year, which will be compatible with Ethernet, allowing for interoperability with other Ethernet devices [13][16]. - The company is also a supporter of the Ultra Ethernet initiative, which aims to modernize Ethernet protocols for HPC and AI applications [15][16]. Market Positioning - Cornelis aims to position its products as cost-effective alternatives to Nvidia's offerings, with a focus on performance and efficiency in AI and HPC environments [7][12].

Core Insights - The global wafer foundry industry is expected to see a revenue decline of approximately 5.4% in Q1 2025, totaling $36.4 billion, influenced by international market conditions and seasonal factors [2][3] - The second quarter is anticipated to show revenue growth for the top ten foundries, driven by China's old-for-new subsidy policy, pre-launch inventory for new smartphone models, and stable demand for AI HPC [2] Company Performance - TSMC remains the market leader with a revenue of $25.5 billion in Q1 2025, a decrease of 5% quarter-over-quarter, holding a market share of 67.6% [3][4] - Samsung's foundry revenue fell by 11.3% to $2.89 billion, with a market share of 7.7%, primarily due to weak mobile chip demand and ongoing inventory adjustments [5] - SMIC reported a revenue increase of 1.8% to $2.25 billion, benefiting from customer pre-orders and domestic consumption subsidies, ranking third in the industry [6] - UMC's revenue decreased by 5.8% to $1.76 billion, maintaining its fourth position, with stable wafer shipments offsetting seasonal impacts [6] - GlobalFoundries experienced a revenue decline of 13.9% to $1.58 billion, while HuaHong Group ranked sixth with stable performance [7] - Vanguard's revenue increased by 1.7% to $363 million, moving up to seventh place, while Tower's revenue decreased by 7.4% to $358 million [7] - Nexchip's revenue grew by 2.6% to $353 million, ranking ninth, while PSMC's revenue slightly decreased by 1.8% to $327 million, placing it tenth [8] Market Trends - TSMC's advanced process technologies, particularly in AI and HPC, are driving revenue growth, with 3nm, 5nm, and 7nm processes contributing 73% of wafer sales [4] - AI chip demand is projected to be a significant growth driver for TSMC, with a forecasted 100% increase in AI accelerator chip sales in 2025 [4] - SMIC anticipates a revenue decline of 4% to 6% in Q2 2025, indicating a challenging market environment ahead [6]

Core Insights - The global wafer foundry industry experienced a 5.4% quarter-over-quarter decline in total revenue for Q1 2025, amounting to $36.403 billion, but showed a significant year-over-year growth of 24.8% [1] - The traditional seasonal slowdown in the wafer foundry industry was somewhat mitigated this year due to uncertainties in international conditions prompting downstream customers to stock up inventory and China's ongoing trade-in policy [1] Industry Performance - Despite facing seasonal challenges, the wafer foundry industry demonstrated resilience, with expectations for revenue growth in Q2 2025 driven by the launch of new smartphone models and stable demand from AI and HPC sectors [4] - The top ten global wafer foundries accounted for 97% of the market share in Q1 2025, with a total revenue of $36.403 billion [2] Company Rankings - TSMC led the market with a revenue of $25.517 billion in Q1 2025, despite a 5.0% decline from Q4 2024, maintaining a market share of 67.6% [2] - Samsung's revenue fell by 11.3% to $2.893 billion, resulting in a market share of 7.7% [2] - SMIC reported a slight revenue increase of 1.8% to $2.247 billion, raising its market share to 6.0% [2] - Vanguard's capacity utilization was above average due to customer stockpiling, while Tower Semiconductor faced significant seasonal impacts and did not benefit from China's subsidy policies [4]

如果您希望可以时常见面，欢迎标星收藏哦~ 来源：内容来自联合报 。 集邦科技今日发布首季晶圆代工市占调查，美国新关税政府引发的提前备货及中国旧换新补贴影 响，急单涌进，抵销淡季效应，首季营收季减5.4%，优于预期，台积电也挟AI及HPC订单需求稳 定，首季全球市占再向上推升，达到67.6%，稳坐全球第一。 集邦调查，第1季主要晶圆代工厂营收合计仅季减约5.4%，为364亿美元，优于预期。展望第2季营 收表现，随着关税引发的提前备货告一段落，整体动能逐步放缓，唯中国旧换新的补贴政策拉货潮 有望延续，加上下半年智慧手机新品上市前备货陆续启动，以及AI HPC需求稳定，将成为带动第 2季产能利用率和出货的关键，预期前十大晶圆代工厂营收将呈现季增。 集邦指出，第1季各晶圆代工业厂仍由台积电以67.6%市占率稳居第一，虽其晶圆出货虽因智慧手 机备货淡季而下滑，但因AI及HPC需求和电视的关税避险急单仍然强劲，抵销手机端需求下滑不 利因素，营收255亿美元，季减5%。 第二名的三星代工部门因美国先进制程禁令限制中国客户投产，以及其客户组成关系，获得中国消 费补贴的红利有限，第1季营收季减11.3%，为28.9亿美元，市占 ...

Summary of Key Points from the Conference Call Industry Overview - The conference call focused on the Taiwan technology sector, particularly semiconductor companies such as TSMC, MediaTek, and UMC, as well as emerging companies like MPI and WinWay [1][3][4]. Core Insights and Arguments TSMC (Taiwan Semiconductor Manufacturing Company) - Investor sentiment regarding AI demand has improved, with a decreasing likelihood of significant AI order cuts in the near term due to better assembly yields from downstream ODMs [3][5]. - TSMC's CoWoS shipments and capacity are expected to grow by 52% and 58% year-over-year in 2026, with capacity increasing from 660k wafers in 2025 to 1,000k wafers in 2026 [5]. - The company's capital expenditure (capex) outlook has been trimmed to US$40 billion for 2026, down from US$45 billion, reflecting potential slower adoption of 2nm technology [5][16]. - TSMC is projected to achieve a 20% revenue compound annual growth rate (CAGR) over the next several years, driven by increasing silicon content and AI demand [14][16]. MediaTek - There are concerns regarding MediaTek's AI ASIC business, with potential revenue expectations for 2026 around US$1 billion [7][18]. - Despite near-term uncertainties, there is optimism about MediaTek's long-term growth in ASICs and its expansion into new markets [7][19]. - MediaTek is expected to see revenue and earnings grow by 16% and 21% CAGR from 2025 to 2027, driven by market share gains and new total addressable markets (TAM) [19][20]. UMC (United Microelectronics Corporation) - UMC has been downgraded to a Sell rating due to risks associated with order cuts and unfavorable foreign exchange trends [8][22]. - The company faces intense pricing pressure from aggressive capacity expansion by mainland Chinese foundries, which is expected to impact its profitability [8][22]. - UMC's share price has increased by 10% year-to-date, but the outlook remains cautious due to competition and potential order cuts in non-AI applications [8][22]. MPI and WinWay - MPI is positioned as a leading probe card provider, with expectations of revenue and earnings CAGR of 19% and 28% from 2024 to 2027, driven by market share gains and increased self-sufficiency [25][27]. - WinWay, a socket provider, is expected to see revenue and earnings accelerate at 23% and 37% CAGR from 2024 to 2027, supported by demand from the AI/HPC segment [30][31]. - Both companies are trading below their historical average P/E ratios, indicating potential upside in their valuations [27][31]. Other Important Insights - Investor sentiment is cautiously optimistic about AI demand, with some investors shifting to a more positive stance as geopolitical tensions and supply chain issues ease [3][5]. - The overall market dynamics for the semiconductor industry are influenced by the ongoing technology migration and increasing complexity of chips, which is driving demand for advanced testing solutions [10][11][31]. - The conference highlighted the importance of understanding the competitive landscape and potential new business opportunities for companies in the semiconductor sector [9][10]. This summary encapsulates the key points discussed during the conference call, providing insights into the current state and future outlook of the Taiwan technology sector, particularly in the semiconductor industry.

先进封装成为下１个技术帝国的边疆要塞，不是偶然，而是3股力道推动出来的必然结果。 第1股力道是算力井喷，但制程进展放缓，芯片必须被切割、堆叠、重组。陆行之表示，你能做到 5奈米，不代表你能塞进20倍算力，光罩极限挡住了芯片的面积，只有Chiplet 能绕过这道墙， Ncidia Blackwell 就是这样诞生的。 第2股力道则是应用百变，芯片不再单一适配，系统设计走向模组化。陆行之说，1种芯片搞定所 有应用的时代已经结束，AI训练、自驾决策、边缘运算、AR装置……每1个应用都需要不同组合 的矽，先进封装加Chiplet，就是设计弹性与效率的平衡解答。 如果您希望可以时常见面，欢迎标星收藏哦~ 来源：内容来自 自由时报 。 半导体的改变正在加速，先进封装，不再是边角料。知名分析师陆行之表示，棋盘中央如果说先进 制程是矽时代的权力中枢，那么先进封装，正在成为下１个技术帝国的边疆要塞。 陆行之在脸书上贴文指出，10年前，这条路线曾被误解，甚至被忽视，但10年后的今天，它已悄 悄从「非主流的Plan B」变成「主流赛道的Plan A」。 第3股力道则是资料搬运成本飙升，能耗变成第１瓶颈。在AI 芯片里，搬资料的耗能 ...

Core Insights - MiTAC Computing Technology Corporation has launched a comprehensive suite of next-generation servers at COMPUTEX 2025, powered by Intel Xeon 6 processors, aimed at AI, HPC, cloud, and enterprise applications [1][2] Product Offerings - The MiTAC G4527G6 is a 4U AI server featuring dual Intel Xeon 6767P processors, supporting up to 8 GPUs and 32 DDR5-6400 RDIMM slots with a maximum memory of 8TB, designed for AI training and advanced analytics [3] - The MiTAC G4520G6 is optimized for compute-intensive workloads, featuring dual Intel Xeon 6700P processors, 8 high-performance GPUs, and energy-efficient power supplies [4] - The MiTAC M2710G6 and M2510G6 are high-density cloud servers, with the M2710G6 supporting Intel Xeon 6900P series processors and up to 128 computing cores per node, while the M2510G6 is a cost-efficient option for cloud service providers [5] - The MiTAC R1520G6 is a 1U dual-socket server optimized for memory-intensive computing, featuring Intel Xeon 6700P series processors and 10 NVMe U.2 drive bays [6] - The MiTAC R2513G6 is a storage-optimized 2U system supporting up to 24 x 3.5" SATA drives, designed for storage-intensive environments [7] - The MiTAC R2520G6 is a versatile 2U dual-socket platform supporting 32 DDR5-6400 RDIMM slots and flexible storage options, ideal for ERP and business intelligence applications [8] Strategic Collaboration - MiTAC and Intel have maintained a collaborative relationship for over five decades, combining Intel's processing power with MiTAC's system design expertise to deliver scalable solutions for modern data centers [2][9] Industry Positioning - MiTAC Computing specializes in AI, HPC, cloud, and edge computing, ensuring high-quality server solutions through rigorous methods and a commitment to performance at all system levels [11][12]

Customer Success & Technology - DDN emphasizes customer success through its HPC experience, focusing on scale, end-to-end configurations, and tuning to ensure successful deployments [2] - DDN's partnership with Nvidia is 30 years old, with Nvidia using DDN for testing, including 4,000 Blackwell GPUs in their lab [2] - DDN highlights the importance of two pillars: Exascaler for large-scale GPU deployments and Infinia as a data intelligence platform [9][10] - DDN's solutions are designed for simple scalability, allowing customers to easily increase capacity by adding more DDN units [11] AI Market & Deployment - AI is permeating across various sectors, including cloud, industries, and daily life, driving significant changes in the next 10 years [3] - DDN is experiencing high demand and is actively hiring to meet customer needs and maintain delivery value [3] - DDN estimates supporting over 1 million GPUs before summer [3] Key Verticals & Partnerships - DDN serves various markets, including NCP/AI providers, healthcare (San Jud), financial services (HFT, fraud detection), energy (oil and gas), automotive (4GM, Tesla), defense, and public sector (NASA) [3] - DDN collaborates with technology partners like Super Micro, Nvidia, and others to deliver solutions at massive scale [6] - DDN has AI cloud program partners, including GCP and Scaleway, offering DDN solutions [6] Product Performance & Innovation - DDN's systems are running at 100% flawlessly [4] - Xia is running a quarter of a thousand GPUs based on 150 terabytes of Exascaler and close to 600 petabytes of Infinia for production [7] - DDN values customers who push the limits of their products, driving innovation and improvements [8]