$ANET
相关推文
按时间倒序
-
谷歌巨额资本支出加速AI基建，利好相关供应链。
$GOOGL 公布了财报，其资本支出(CapEX)规模巨大。 1750亿-1850亿美元 vs. 1200亿美元。 这对AI基础设施建设是利好。 跟随资金流向下游： - $AVGO、联发科、$TSM（设计/晶圆代工） - SK海力士、三星、$MU、$SNDK（存储） - $ANET、$LITE、$COHR、$VRT（网络、光子学、能源） 等等，因为谷歌表示支出主要在于：“包括服务器、数据中心和网络在内的AI基础设施”。 从 $META 到 $GOOGL 的超大规模云服务商正在大幅增加资本支出。 这仅仅表明： AI基础设施建设正在加速，并由全球最盈利的公司资助。 做多AI供应链。
英文原文
$GOOGL reported earnings and their CapEX spend was enormous. $175-$185 Billion vs. $120 Billion. This is bullish for AI buildout. Follow the money flow down to: - $AVGO, Mediatek, $TSM (design / foundry) - Sk hynix, Samsung, $MU, $SNDK (memory) - $ANET, $LITE, $COHR, $VRT (network, photonics, energy) etc. as Google said spend was: "primarily on AI infrastructure including servers, data centers and networking" Hyperscalers from $META to $GOOGL are increasing their capex spend enormously. This just goes to show: AI buildout is accelerating and is funded by the most profitable companies in the world. Long AI supply chains.
-
日企Nitto Boseki垄断AI关键材料T-glass，巨头争抢供应至2026年。
Nitto Boseki - $NBCLF | 3110: AI极端瓶颈的故事： > 1923年开始纺纱/玻璃纤维 > 做到极致 > 为网球拍创造“T-Glass”（T-玻璃） > 等待100年，所有人都需要它 > $TSM 用于 CoWoS（芯片级封装） > $APPL 用于 M-Series Ultra/Max > $NVDA 用于 Blackwell + Vera Rubin > $TSLA 用于 FSD 汽车内的芯片 > $AMD 用于 MI300/MI400 AI 加速器 > $ANET 用于 800G 到 1.6T 和 3.2T > 苹果、英伟达和台积电争夺供应 > 已售罄至 2026 年 > 苹果、英伟达和高通派团队驻厂，字面意义上向这家纱线制造商乞求玻璃供应 > 目前在半导体用高端 T-glass 市场占据 80-90% 份额 这家价值 42.3 亿美元的垄断企业，全球科技公司都在争夺： 一家随机的日本纱线制造商。 披露：我持有头寸。
英文原文
Nitto Boseki - $NBCLF | 3110: The story of an extreme bottleneck for AI: > Spin Yarn/Glass in 1923 > Be good at it > Create "T-Glass" for tennis rackets > Wait 100Y, everyone needs it > $TSM for CoWoS > $APPL for M-Series Ultra/Max > $NVDA for Blackwell + Vera Rubin > $TSLA for chips inside FSD cars > $AMD for MI300/MI400 AI accelerators > $ANET for 800G to 1.6T and 3.2T > Have Apple to Nvidia to TSM fighting over supply > Sold out through 2026 > Apple, Nvidia, and Qualcomm sends teams live on site to literally beg this Yarn Maker for glass supply > Now is 80-90% market share of the high-end T-glass used in semiconductors The $4.23 Billion monopoly that every tech company in the world is fighting over: A random Japanese Yarn Spinner. Disclosure: I own positions.
-
分析Google TPU v7供应链，建仓Lumentum以博弈TPU生态扩张。
对 $GOOGL TPU v7 Ironwood 供应商的分析。 以下是受 Google TPU 建设影响最大的公司列表。 + 我正在建仓的 TPU 相关股票。 [关键] 设计/IP： - 博通 [ $AVGO ]：共同设计并实现 Google 的 TPU ASIC（专用集成电路） [关键/高] 半导体晶圆代工： - #1 台湾半导体 [ $TSM ]：TPUv7 在 TSMC 3nm 工艺制造 - #2 三星电子：次要存储及晶圆代工合作伙伴 [关键/高] 存储： - #1 SK 海力士：为 TPUv7 Ironwood 提供 HBM3E - #2 三星电子：~TPUv7 特定报告强调 SK 海力士 + 三星。 [高] 光网络： - Lumentum [ $LITE ]：Google 广泛使用光电路交换 (OCS) - Coherent [ $COHR ]：OCS 参与者但较弱 [高] 电源管理 IC： - Monolithic [ $MPWR ]：这是一个投机性观点，即 Vicor 将被 $MPWR 取代，源自财报中提及 TPU [中] 热管理： - Vertiv [ $VRT ]：Vertiv 供应作为液冷系统核心的 CDU（冷却分配单元），将冷却液泵送至 TPU 芯片的冷板 - Modine [ $MOD ]：更投机性地认为他们提供大型冷水机组和空气处理单元 (AHU) [中] 互连： - TTM Technologies [ $TTMI ]，$ANET，Unimicron，Ibiden ______ Google TPU v7 “Ironwood” 的建设代表了一个旨在打破 $NVDA GPU 垄断的平行硅生态系统的构建。 实质性影响最集中在博通（作为硅架构师和商业载体）、存储综合体（SK 海力士/三星）以及光网络/电源领域（Lumentum/Vertiv），这是基于公开证据创建的，但很大程度上取决于采用率、供应商份额和竞争反应的实际表现。 从这项供应链研究中，我正在 $LITE 建立新头寸，以防 TPU 成为推理领域的主导 ASIC。 Lumentum 是 Google 致力于 OCS 的主要受益者，并构成了 TPU 吊舱中使用的 “Apollo” OCS 交换机的核心。 TPU v7 集群的爬坡直接转化为 Lumentum 光开关模块的出货量。由于 OCS 是 Google 超大规模方法独有的定制架构，Lumentum 在此处面临的 commoditization（商品化）压力小于标准收发器市场。 然而，如果 Anthropic、Meta、Apple 和其他公司购买 $GOOGL ASIC 导致 TPU v7 规模扩大，该供应链中的所有公司都将受益。
英文原文
Analysis of the $GOOGL TPU v7 Ironwood Suppliers. Here's the list of what comapnies are the most materially impacted by the Google's TPU buildout. + the TPU stock I'm taking a position on. [Critical] Design/IP: - Broadcom [ $AVGO ]: co-designs and implements Google’s TPU ASICs [Critical/High] Semiconductor Fab: - #1 Taiwan Semi [ $TSM ]: TPUv7 is fabbed at TSMC 3nm - #2 Samsung Electronics: Secondary memory & foundry partner [Critical/High] Memory: - #1 SK Hynix: HBM3E for TPUv7 Ironwood - #2 Samsung Electronics: ~TPUv7-specific reporting emphasizes SK hynix + Samsung. [High] Optical Networking: - Lumentum [ $LITE ]: Google uses extensively uses Optical Circuit Switching (OCS) - Coherent [ $COHR ]: OCS player but weaker [High] Power Management ICs: - Monolithic [ $MPWR ]: This is speculative that Vicor will be replaced by $MPWR, from earnings mentioning TPU [Medium] Thermal Management: - Vertiv [ $VRT ]: Vertiv supplies the CDUs that act as the heart of the liquid cooling system, pumping coolant to the cold plates on the TPU chips - Modine [ $MOD ]: More speculative that they provide provides the massive chillers and air handling units (AHUs) [Medium] Interconnects: - TTM Technologies [ $TTMI ], $ANET, Unimicron, Ibiden ______ The buildout of the Google TPU v7 "Ironwood" represents the construction of a parallel silicon ecosystem designed to break the monopoly of $NVDA GPU. The material impact is most concentrated in Broadcom (as the silicon architect and commercial vehicle), the Memory Complex (SK Hynix/Samsung), and the Optical/Power sectors (Lumentum/Vertiv) and was created from public evidence but is largely dependent on adoption, vendor shares, and competitive responses actually play out. From this supply chain research, I'm initiating a new position in $LITE, in the event the TPU becomes the dominant ASIC for inference. Lumentum is primary beneficiary of Google’s commitment to OCS and form the core of the "Apollo" OCS switches used in TPU pods. The ramping of TPU v7 clusters translates directly to unit volume for Lumentum’s optical switch modules. And because OCS is a bespoke architecture unique to Google’s hyperscale approach, Lumentum faces less commoditization pressure here than in the standard transceiver market. However all companies in this supply chain are set to benefit if the TPU v7 scales up from Anthropic, Meta, Apple, and others buying the $GOOGL ASIC.