（联合报系资料照） 张量处理器（TPU）需求强劲，美银证券将 2026 年 TPU 预估出货量从 400 万颗上修至，看好相关供应链，包 括台积电（2330）、京元电子（2449）、日月光投控（3711）、致茂（2360）、颖崴（65 美银发布 AI 特殊应用 IC（ASIC）报告指出，根据供应链在财报季的回馈，将 TPU 今年预估出货量上调到 460 万颗，比去年的 230 万颗增加一倍。 （联合报系资料照） 张量处理器（TPU）需求强劲，美银证券将 2026 年 TPU 预估出货量从 400 万颗上修至，看好相关供应链，包 括台积电（2330）、京元电子（2449）、日月光投控（3711）、致茂（2360）、颖崴（6515）、旺矽（6223）、 世芯 – KY（3661）等。 美银发布 AI 特殊应用 IC（ASIC）报告指出，根据供应链在财报季的回馈，将 TPU 今年预估出货量上调到， 比去年的 230 万颗增加一倍。 美银发布 AI 特殊应用 IC（ASIC）报告指出，根据供应链在财报季的回馈，将 TPU 今年预估出货量上调到 460 万颗，比去年的 230 万颗增加一倍。 其中，训练晶片组由博通主导 ...

张量处理器（TPU）需求强劲，美银证券将 2026 年 TPU 预估出货量从 400 万颗上修至460 万颗，看好相关供应 链，包括台 美银将 2026 年 TPU 出货预期从 400 万颗上修至 460 万颗 看好七档台股 美银发布 AI 特殊应用 IC（ASIC）报告指出，根据供应链在财报季的回馈，将 TPU 今年预估出货量上调到 460 万 颗，比去年的 230 万颗增加一倍。 （联合报系资料照） 美银将 2026 年 TPU 出货预期从 400 万颗上修至 460 万颗 看好七档台股 美银指出，TPU 需求强劲，主要受出本身工作负载快速成长及外部销售推动，源头信息加微WUXL3408尤其在 Meta 的 MTIAASIC 专案因管理阶层变动而延迟后更是明显。 此外，Google 正加速采用自研 ARM 伺服器 CPU（目前为 Axion，后续専案预计在 2027 年下半年推出），搭配 TPU，以取代 x86 CPU。 美银也上调亚马逊 Trainium的预估需求，将 2026 年的预估量由 200 万颗上修至 210 万颗，2027 年的预估由 260 万 颗调高到 300 万颗，因为亚马逊的资本支出 ...

Core Viewpoint - Meta and Nvidia have established a long-term partnership focusing on local deployment, cloud, and AI infrastructure, marking a significant expansion of their technological collaboration [1][5]. Group 1: Partnership Details - Meta will build ultra-large-scale data centers optimized for training and inference to support its long-term AI infrastructure roadmap [3]. - The collaboration will involve the deployment of millions of Blackwell and Rubin GPUs, as well as Nvidia's Grace CPUs, with Nvidia's Spectrum-X Ethernet switches integrated into Facebook's open switching system [3]. - This partnership represents the first large-scale deployment of Nvidia's Grace and aims to enhance the energy efficiency of AI computing [3][6]. Group 2: Strategic Implications - Meta is expected to allocate a significant portion of its projected capital expenditure of up to $135 billion this year towards expanding Nvidia's data centers [7]. - The large-scale adoption of Nvidia's chips validates Nvidia's "full-stack" infrastructure strategy, which includes both CPU and GPU [7]. - Despite the partnership, Meta is also exploring alternatives, including the potential use of Google's Tensor Processing Units (TPUs) in its data centers by 2027 [7][8]. Group 3: Market Reactions - Following the announcement, both Meta and Nvidia's stock prices rose in after-hours trading, while AMD's stock fell over 4% [3].

Core Viewpoint - AMD has significantly expanded its scale as a GPU company, currently valued at approximately $350 billion, but analysts still believe it may struggle to generate substantial investment returns despite its stock performance tripling compared to the S&P 500 index [1][2]. Financial Performance - AMD's revenue grew by 36% year-over-year, increasing from $6.8 billion to $9.2 billion, primarily driven by demand from large enterprise customers, particularly in the data center and server business [10][16]. - The annualized revenue for AMD is nearing $37 billion, indicating a substantial business scale [11]. - The diluted earnings per share (EPS) rose from $0.47 to $0.75, reflecting a 60% year-over-year increase [16]. Product and Market Position - AMD has built a competitive product matrix leveraging its Ryzen processors and AI technology, enhancing its market competitiveness [4]. - The company's strength is evident in its support for global supercomputers, showcasing its advanced GPU products and leading CPU technology [7]. - AMD's data center business is a core segment, contributing nearly 50% of total revenue with quarterly earnings reaching $4.3 billion [18]. Strategic Partnerships - AMD announced a significant collaboration with OpenAI, involving a $6 billion agreement linked to GPU deployment, which may prioritize AMD's products in future computational needs [19]. Market Challenges - Despite the growth, AMD's non-GAAP gross margin remained flat at 54%, indicating limited pricing power amidst increasing competition [14]. - Analysts express concerns over the potential profit margin squeeze due to the unique partnership structure with OpenAI, which could impact AMD's profitability [20]. - The competitive landscape is intensifying with alternatives like Google's TPU, posing challenges to AMD's future growth [24]. Future Outlook - AMD's revenue forecast for the next quarter is approximately $9.6 billion, with a projected gross margin of around 54.5%, indicating a slowdown in growth momentum [23]. - The GPU industry is currently in a multi-player expansion phase, but AMD may require more time to achieve sufficient profit growth as the market transitions to a cost-sensitive phase [26]. - Analysts highlight the risk of a "one-stop replacement" scenario, where AMD's ability to produce competitive chips could lead to significant revenue growth, especially as Intel faces challenges [27].

Core Viewpoint - Micron Technology (MU) is well-positioned in the memory market driven by AI demand, but it needs to demonstrate sustainable long-term growth to investors [1][2]. Financial Performance - Micron recently reported its Q1 FY2026 earnings, showcasing strong performance despite market downturns. The guidance for Q2 FY2026 indicates adjusted earnings per share expected to soar from $4.78 to $8.42, significantly exceeding market expectations, with revenue projected at $18.7 billion, surpassing expectations by $4.47 billion [3][4]. Market Trends and Opportunities - Analysts highlight that the development of AI technology heavily relies on memory, with increasing demand for High Bandwidth Memory (HBM). The transition from HBM3E to HBM4 is expected to drive average selling prices (ASP) higher, while current demand exceeds supply, supporting a favorable industry outlook [7][8]. Market Share and Growth Projections - Micron's market share in HBM is approximately 21%, ranking second in the industry. The global HBM market is projected to grow at a CAGR of 40%, expanding from $35 billion this year to $100 billion by 2028. If Micron maintains its market share, it could generate around $10.29 billion in revenue from HBM alone by 2026 [9][10]. Profitability and Margin Expansion - Micron's gross margin increased significantly by 1110 basis points to 56.8%, with expectations to rise further to 68% in the next quarter. This contrasts with competitors like Broadcom and NVIDIA, which anticipate margin pressures [10][11]. Revenue Composition - In Q1 FY2026, DRAM accounted for 79% of total revenue, with a slight increase in sales volume and a 20% increase in ASP. NAND contributed 20% to revenue, with mid-to-high single-digit percentage growth in sales volume and mid-teens percentage increase in ASP [12]. Valuation Metrics - Micron's forward P/E ratio stands at 12.61, significantly lower than the industry average of 24.35. The forward EV/Sales ratio is 4.49, slightly above the industry average of 3.38, indicating that Micron's valuation remains relatively low compared to its peers [13][14]. Future Growth Drivers - The ongoing demand for HBM is expected to outstrip supply, driven by the growth of AI GPUs and the expansion of the ASIC market. Micron plans to increase capital expenditures by 45%, raising its budget from $18 billion to $20 billion to capitalize on these growth opportunities [15].

台湾经济日报：供应链最新消息指出，随着谷歌张量处理器（TPU）需求大涨，谷歌扩大了对联发科合作定制新一代TPU v7e的订单，订单量比原规划激增数倍。消息称，联发科为谷歌操刀定制的首款TPU v7e将于下季度末进入风险性试产，并再拿下谷歌下一代TPU v8e的订单。2027年台积电提供给联发科谷歌项目的CoWoS产能更将暴增7倍以上。外汇交易员 (@myfxtrader):谷歌正大幅加紧在AI芯片竞赛中与英伟达展开竞争，而Meta正成为其潜在的数十亿美元客户。多年来，谷歌一直将其定制TPU限制在自家云数据中心内使用，并对外出租算力。但据The https://t.co/D2wxUlbHSQ ...

Group 1 - Media reports indicate that MediaTek has partnered with Alphabet's unit to design Tensor Processing Units (TPUs), which are seen as potential competitors to NVIDIA's chips in the AI application field [1] - MediaTek is known for smartphone chips, but faces pressure on gross margins due to uncertain demand, intense competition, and high R&D costs; however, AI-related news has provided some relief for its stock price, which has still declined by approximately 1.4% this year [1] - Morgan Stanley analysts upgraded MediaTek's rating from "Equal Weight" to "Overweight," citing that the growth of Google TPUs should offset headwinds in the smartphone market in the long term [1] Group 2 - UBS analysts raised their 2027 sales forecast for MediaTek's TPUs from $1.8 billion to $4 billion, predicting that these chips will account for 20% of the company's operating profit by 2028, contingent on MediaTek's execution with Google [2] - Recent interest has been fueled by reports that Meta is discussing the adoption of Google TPUs in data centers by 2027; UBS believes MediaTek has further growth potential in additional ASIC projects with Meta [2] - Overall, foreign investors remain optimistic about MediaTek, with 23 firms maintaining a "Buy" rating and 10 firms a "Hold" rating, while no firms have issued a "Sell" rating; analysts from Macquarie Group express a preference for investing in MediaTek and other Google partners over NVIDIA's supply chain [2]

Group 1 - OpenAI has placed a $10 billion order for new AI self-developed chips with Broadcom, aiming to reduce reliance on Nvidia and enhance its AI server capabilities [1] - OpenAI plans to produce its own AI chips for the first time next year, with the chips designed in collaboration with Broadcom, expected to be shipped next year [1] - Foxconn (Hon Hai) has set a target for AI server-related revenue to exceed NT$1 trillion this year, benefiting from increasing ASIC server orders [1] Group 2 - OpenAI has revised its spending forecast for this year to over $8 billion, an increase of approximately $1.5 billion from previous estimates, and expects spending to exceed $17 billion next year [2] - Google is accelerating the deployment of its TPU and has reached an agreement with Fluidstack to install TPUs in its New York data center, receiving up to $3.2 billion in funding support [2] - Foxconn is the largest contract manufacturer for Nvidia's GPU servers and has secured ASIC server orders from major U.S. cloud service providers, with expectations to expand its footprint in the ASIC server market as OpenAI enters the competition [2]

Group 1: Nvidia - Nvidia has faced production issues with the GB200 NVL72 rack, leading to a second reduction in its CoWoS supply and shipments falling below targets [1] - KeyBanc analysts expect Nvidia's Q2 revenue to be $45.1 billion, slightly below market expectations of $45.6 billion, but anticipate a Q3 guidance of $53.5 billion, exceeding the FactSet consensus of $51.8 billion [3] - Market sentiment remains positive due to Nvidia's strong position in the generative AI sector, with investors focusing on its business in China, the impact of U.S. export controls, and feedback on the Blackwell platform and NVLink technology [3] Group 2: AMD - KeyBanc forecasts AMD's Q2 revenue to be $7.51 billion, above market expectations of $7.41 billion, with Q3 guidance expected to reach $8.63 billion, also higher than the consensus of $8.25 billion [4] - Despite maintaining an "equal weight" rating due to uncertainties in the data center GPU business and potential weakness in PC sales, AMD has made progress in the AI market [4] - Investors are expected to focus on customer feedback for the MI355 chip, annual AI-related revenue forecasts, traditional server business performance, and future plans for the MI400 series [4] Group 3: Broadcom - KeyBanc anticipates Broadcom's Q3 revenue to be $15.8 billion, in line with market expectations, while Q4 revenue is projected to reach $17.7 billion, surpassing the consensus of $17 billion [5] - Investors will be monitoring Broadcom's AI business outlook, ASIC order backlog, customer collaborations, and updates related to trade tensions with China and the development of the iPhone 17 in partnership with Apple [5] Group 4: Qualcomm and Monolithic Power Systems - KeyBanc holds a cautious outlook on Qualcomm and Monolithic Power Systems, with Monolithic expected to regain some market share on Nvidia's Blackwell Ultra HGX platform, but overall market share growth is limited due to a decline in enterprise data business [6] - Qualcomm's performance in the June quarter is expected to benefit from short-term gains due to subsidies for Chinese head-mounted devices, but guidance for the September quarter may be lowered as subsidy funds decrease [6] - Overall sentiment towards Qualcomm is negative, with concerns over Apple's in-house baseband chip development and a slowdown in Android smartphone demand impacting future performance guidance [6]