diff --git a/.sync-state b/.sync-state index 421f4ff..090de86 100644 --- a/.sync-state +++ b/.sync-state @@ -1 +1 @@ -2026-05-26T17:01:06.670054+10:00 \ No newline at end of file +2026-05-27T20:57:32.837434+10:00 \ No newline at end of file diff --git a/links-posts/2663-2026-05-27-technology.md b/links-posts/2663-2026-05-27-technology.md new file mode 100644 index 0000000..b59e6f5 --- /dev/null +++ b/links-posts/2663-2026-05-27-technology.md @@ -0,0 +1,34 @@ +--- +title: "2026-05-27-technology" +created: 2026-05-27 +updated: 2026-05-27 +type: summary +tags: [newsletter, technology] +external: http://go/ui/posts/2663/ +--- + +**科技新闻** + +- [首屆AI原生世代畢業，職場準備就緒](https://www.wsj.com/tech/ai/ai-natives-graduates-job-cuts-6bab8ac9?mod=rss_Technology) + 初級職位減少，但具備AI技能的畢業生備受追捧。 + +- [Mistral AI借Harvey AI合作進軍法律領域](https://www.wsj.com/tech/ai/mistral-ai-takes-aim-at-legal-sector-through-expanded-harvey-ai-partnership-2a8dc703?mod=rss_Technology) + 法國AI公司通過與Harvey AI合作，瞄準利潤豐厚的法律行業，與Anthropic等對手競爭。 + +- [用戶向聊天機器人詢問中國時發現了什麼](https://www.wsj.com/world/china/what-readers-found-when-they-asked-their-chatbots-about-china-229e4203?mod=rss_Technology) + 調查顯示AI對中國的回答存在偏見；好萊塢編劇稱歷史糾紛導致中美合拍項目擱淺。 + +- [FBI網路犯罪報告四大要點：AI犯罪激增](https://www.wsj.com/tech/cybersecurity/internet-crime-fbi-report-fd7c16e8?mod=rss_Technology) + 網路犯罪分子利用AI技術大幅簡化犯罪流程，相關案件數量猛增。 + +- [美國航空選用SpaceX星鏈為500多架飛機提供機上Wi-Fi](https://www.cnbc.com/2026/05/26/american-airlines-spacex-starlink.html) + 美國航空宣布將在超過500架飛機上安裝SpaceX的星鏈系統用於機內網絡服務。 + +- [美光科技市值首次突破1萬億美元，股價飆升19%](https://www.cnbc.com/2026/05/26/micron-stock-trillion-market-cap.html) + AI驅動的全球內存短缺推高了對美光等芯片製造商的需求。 + +- [蘋果股價創新高後面臨六月WWDC重大考驗](https://www.cnbc.com/2026/05/26/apples-surge-to-record-highs-faces-a-major-test-next-month-what-it-must-do-to-pass.html) + 蘋果全球開發者大會於6月8日開幕，需在AI領域再次打動華爾街。 + +- [Dropbox CEO Drew Houston任職19年後將卸任](https://www.cnbc.com/2026/05/26/dropbox-ceo-drew-houston-ashraf-alkarmi.html) + 24歲創立雲存儲公司的Drew Houston計劃卸任CEO，轉任執行董事長。 \ No newline at end of file diff --git a/links-posts/2664-每日新闻简报-2026-05-27.md b/links-posts/2664-每日新闻简报-2026-05-27.md new file mode 100644 index 0000000..b1e22fd --- /dev/null +++ b/links-posts/2664-每日新闻简报-2026-05-27.md @@ -0,0 +1,68 @@ +--- +title: "📰 每日新闻简报 | 2026-05-27" +created: 2026-05-27 +updated: 2026-05-27 +type: summary +tags: [2026, hackernews, international, news, stocks, tech] +external: http://go/ui/posts/2664/ +--- + +# 📰 每日新闻简报 | 2026-05-27 + +## 🌍 国际要闻 + +**BBC World** +1. [伊朗谴责美国空袭"严重违反"停火协议](https://www.bbc.com/news/articles/c4g44yl7q70o?at_medium=RSS&at_campaign=rss) +2. [俄罗斯威胁发动更多基辅打击，并呼吁外国公民撤离](https://www.bbc.com/news/articles/c1e22n55zn4o?at_medium=RSS&at_campaign=rss) +3. [比利时校车与火车相撞致四人死亡](https://www.bbc.com/news/articles/c70vvd17rg7o?at_medium=RSS&at_campaign=rss) +4. [以色列加强空袭，黎巴嫩村庄十余人遇难](https://www.bbc.com/news/articles/cx211rgngp0o?at_medium=RSS&at_campaign=rss) +5. [温度纪录为何不仅被打破，更被大幅超越](https://www.bbc.com/news/articles/c62rrj66p3eo?at_medium=RSS&at_campaign=rss) + +## 💻 科技资讯 + +**TechCrunch** +1. [2026年Startup Battlefield参赛指南：5月27日截止，我们在寻找什么](https://techcrunch.com/2026/05/26/what-were-looking-for-in-startup-battlefield-2026-and-how-to-apply-in-time-for-the-may-27-deadline/) +2. [特朗普政府允许沃尔沃继续在美国销售联网汽车](https://techcrunch.com/2026/05/26/trump-admin-permits-volvo-to-keep-selling-connected-cars-in-the-u-s/) +3. [特朗普政府希望核能初创公司使用钚作为反应堆燃料](https://techcrunch.com/2026/05/26/trump-admin-wants-nuclear-startups-to-use-plutonium-for-their-reactors/) +4. [DuckDuckGo安装量增长30%，用户拒绝"被迫接受"谷歌AI搜索](https://techcrunch.com/2026/05/26/duckduckgo-installs-are-up-30-as-users-reject-being-force-fed-googles-ai-search/) +5. [SpaceX星舰S-1测试后，可重复使用之路前景不明](https://techcrunch.com/2026/05/26/starships-path-to-reusability-looks-murky-after-spacexs-s-1/) + +## 📈 美股行情（2026-05-26 收盘） + +### 主要指数 +| 指数 | 收盘价 | 涨跌 | +|------|--------|------| +| Nasdaq Composite | 26,656.18 | +312.21 (+1.19%) | +| Nasdaq-100 (NDX) | 30,001.32 | +519.67 (+1.76%) | +| Dow Jones | 50,461.68 | -118.02 (-0.23%) | + +### Nasdaq 科技股 +| 股票 | 价格 | 涨跌 | +|------|------|------| +| NVDA | $214.00 | -0.86 (-0.40%) | +| AAPL | $308.15 | -0.18 (-0.06%) | +| MSFT | $415.14 | -0.89 (-0.21%) | +| GOOGL | $388.09 | -0.79 (-0.20%) | +| META | $609.45 | -2.89 (-0.47%) | +| AMZN | $264.43 | -0.86 (-0.32%) | +| TSLA | $434.81 | +1.22 (+0.28%) | +| AMD | $504.92 | +1.03 (+0.20%) | +| INTC | $122.86 | -0.66 (-0.53%) | + +> 市场简析：纳斯达克表现强劲，综合指数和100指数分别上涨1.19%和1.76%，NDX突破30,000点关口，主要由科技权重股提振。道琼斯小幅下跌0.23%，走势分化。科技股方面，特斯拉和AMD逆势微涨，而英伟达、Meta小幅回调。市场整体呈现科技板块领涨态势。 + +## 🏄 Hacker News Top 10 + +1. [西班牙禁止预测市场Polymarket和Kalshi，称其缺乏赌博许可](https://www.reuters.com/business/spain-blocks-prediction-markets-polymarket-kalshi-over-lack-gambling-licences-2026-05-26/) — 686 points, 327 comments +2. [一些有趣的现代像素字体](https://unsung.aresluna.org/a-few-interesting-modern-pixel-fonts/) — 188 points, 43 comments +3. [化学品罐背后的化学原理](https://www.science.org/content/blog-post/methyl-methacrylate-tank) — 153 points, 56 comments +4. [大型科技公司的反劳工剧本已降临维基百科](https://medium.com/@jakeorlowitz/wikipedia-is-doing-the-capitalist-thing-56a393232943) — 141 points, 59 comments +5. [我绕开Adobe和微软，构建了Git追踪的图书制作流水线](https://www.djspeckhals.com/posts/2026-05-22-how-i-bypassed-adobe-and-microsoft-to-build-a-git-tracked-book-production-pipeline/) — 123 points, 32 comments +6. [Rosalind：用Rust编写的基因组学工具包，可在笔记本上运行全基因组分析](https://github.com/logannye/rosalind) — 109 points, 26 comments +7. [我经历过的最糟糕面试](https://www.oliverio.dev/blog/the-worst-job-interview-i-had) — 63 points, 33 comments +8. [C语言数组类型很奇怪](https://anselmschueler.com/blogposts/2025-c-pointers/) — 29 points, 9 comments +9. [Uber一个季度用光了AI预算](https://fortune.com/2026/05/26/uber-coo-ai-spending-tokens-claude-code/) — 19 points, 7 comments +10. [一次不祥的重聚](https://bcantrill.dtrace.org/2026/05/25/a-portentous-reunion/) — 12 points, 3 comments + +--- +*新闻来源：BBC / TechCrunch / HN | 股市数据：NASDAQ / Yahoo Finance | 抓取时间：2026-05-27 09:00 AEDT* \ No newline at end of file diff --git a/links-posts/2665-2026-05-27-victoria.md b/links-posts/2665-2026-05-27-victoria.md new file mode 100644 index 0000000..df953ea --- /dev/null +++ b/links-posts/2665-2026-05-27-victoria.md @@ -0,0 +1,26 @@ +--- +title: "2026-05-27-victoria" +created: 2026-05-27 +updated: 2026-05-27 +type: summary +tags: [local_news, newsletter] +external: http://go/ui/posts/2665/ +--- + +# 突发新闻 +- [能源账单将下降：四大州居民和小企业将受益](https://www.sbs.com.au/news/article/energy-bills-cheaper-as-default-market-offers-drop/7poit1zs2) + 新财年起，澳洲四个州的居民和小型企业的能源账单有望下降，带来电价减免。 +- [IS相关家庭从叙利亚返回澳洲](https://www.sbs.com.au/news/article/second-group-isis-bride-to-arrive-in-australia/f22dtm7sm) + 七名与自称“伊斯兰国”组织有关联的妇女及其子女已抵达澳洲，数周前已有三名妇女被捕。 +- [珀斯“入侵日”集会炸弹案嫌犯首次出庭](https://www.sbs.com.au/news/article/failed-invasion-day-bombing-accused-makes-first-court-appearance/kfi13bota) + 一名被控在珀斯“入侵日”集会投掷炸弹的男子首次出庭，面临多项指控，其中包括西澳历史上首例恐怖主义行为指控。 +- [Vivid Sydney史上最大规模无人机表演因“异常”取消](https://www.sbs.com.au/news/article/vivid-sydney-drone-show-cancelled/jeozjgw9l) + Vivid Sydney史上规模最大的无人机表演因技术“异常”而未能成功举办，具体原因正在调查中。 +- [维州幼教改革适得其反，墨尔本面临幼儿园短缺危机](https://www.theage.com.au/national/victoria/this-program-aimed-to-improve-early-education-instead-it-s-created-a-growing-kinder-crisis-20260514-p5zwrr.html?ref=rss&utm_medium=rss&utm_source=rss_national_victoria) + 一项旨在改善早期教育的项目，却导致墨尔本未来十年可能短缺数千个幼儿园名额，活动中心激增而理事会难以应对需求。 +- [墨尔本可疑火灾殃及企业，被指与黑帮头目Gatto及飞车党有关](https://www.theage.com.au/national/victoria/suspicious-inferno-guts-businesses-in-melbourne-s-west-20260526-p600mq.html?ref=rss&utm_medium=rss&utm_source=rss_national_victoria) + 墨尔本西部一家律师事务所在可疑火灾中被毁，其所有者被曝与知名黑帮人物Mick Gatto及飞车党领袖有联系。 +- [维州“成年人刑罚”新法首案推迟审理](https://www.theage.com.au/national/victoria/first-adult-time-for-adult-crime-plea-adjourned-amid-courtroom-confusion-20260526-p600nm.html?ref=rss&utm_medium=rss&utm_source=rss_national_victoria) + 一名16岁少女成为维州新“成年人刑罚”法案生效后首位被起诉的青少年，该案件因法庭程序问题被推迟审理。 +- [喜剧演员Hughesy痛批维州政府处理犯罪和腐败不力](https://www.theage.com.au/national/victoria/beyond-a-joke-hughesy-blows-up-over-crime-corruption-and-the-victorian-government-20260526-p600se.html?ref=rss&utm_medium=rss&utm_source=rss_national_victoria) + 知名喜剧演员Dave Hughes公开抨击维多利亚州政府在处理犯罪和腐败问题上的表现，认为维州已陷入混乱，呼吁政府变革。 \ No newline at end of file diff --git a/links-posts/2666-486231-for-aud-to-cny-on-5272026.md b/links-posts/2666-486231-for-aud-to-cny-on-5272026.md new file mode 100644 index 0000000..478af1c --- /dev/null +++ b/links-posts/2666-486231-for-aud-to-cny-on-5272026.md @@ -0,0 +1,10 @@ +--- +title: "4.86231 for AUD to CNY on 5/27/2026" +created: 2026-05-27 +updated: 2026-05-27 +type: summary +tags: [aud, bot, cny, rate] +external: http://go/ui/posts/2666/ +--- + +Current rate from 1 AUD to CNY is ¥4.86231 at time 2026-05-26T00:00:00+0000, more info please check \ No newline at end of file diff --git a/links-posts/2667-2026-05-27-world_news.md b/links-posts/2667-2026-05-27-world_news.md new file mode 100644 index 0000000..f146e6a --- /dev/null +++ b/links-posts/2667-2026-05-27-world_news.md @@ -0,0 +1,32 @@ +--- +title: "2026-05-27-world_news" +created: 2026-05-27 +updated: 2026-05-27 +type: summary +tags: [newsletter, world_news] +external: http://go/ui/posts/2667/ +--- + +- [美国对伊朗发动“自卫打击” 伊朗谴责严重违反停火](https://www.cnbc.com/2026/05/26/us-conducts-self-defense-strikes-in-iran-as-trump-seeks-peace-deal.html) + 美军打击伊朗导弹发射点及布雷船；伊朗抨击此举为“严重违反停火”，当时双方正在多哈进行和谈。 + +- [俄罗斯警告美国公民离开基辅 称将对乌首都实施“系统性打击”](https://www.cnbc.com/2026/05/26/russia-ukraine-war-rubio-trump-lavrov-kyiv.html) + 俄方通过鲁比奥渠道告知美方，即将对基辅发动持续、集中打击，敦促美国公民撤离。 + +- [卡尼警告：艾伯塔省若脱离加拿大将重蹈“英国脱欧”式后悔](https://www.cnbc.com/2026/05/26/alberta-leave-canada-vote-mark-carney-brexit-stay-free.html) + 加拿大总理卡尼（前英国央行行长）称艾伯塔省独立公投是“非常危险的虚张声势”，或引发类似脱欧的经济痛苦。 + +- [投资者“不敢建仓”油市 霍尔木兹海峡通行费威胁加剧](https://www.cnbc.com/2026/05/26/iran-oil-price-strait-hormuz-trump-peace-toll-charges-energy.html) + 油轮面临伊朗可能对霍尔木兹海峡收取费用的风险，数据机构称市场恐惧已导致交易员不敢操作。 + +- [欧洲央行誓言“不惜一切”遏制通胀 市场预期下次会议加息](https://www.cnbc.com/2026/05/26/ecb-villeroy-inflation-iran-war.html) + 法国央行行长表示欧央行将采取必要行动，市场定价几乎完全计入加息。 + +- [SpaceX提交IPO文件引爆太空板块牛市 分析师给出“买入”策略](https://www.cnbc.com/2026/05/26/spacex-ipo-filing-sparks-a-bull-market-in-space-stocks-how-to-buy-in.html) + SpaceX上市申请令太空股集体飙升，供应商、替代标的甚至竞争对手均被视为受益者。 + +- [西欧遭热浪袭击 巴黎“酷热难耐”](https://www.bbc.com/news/videos/cy522zrkr7lo) + 一股热穹顶笼罩西欧，巴黎气温远超5月均值，当地居民与游客经历“惩罚性”高温。 + +- [BP股价下跌 董事会以“严重行为问题”罢免董事长曼尼福德](https://www.cnbc.com/2026/05/26/bp-chair-albert-manifold-removed-conduct-governance-issues.html) + 英国石油巨头BP董事会一致决定立即撤换董事长阿尔伯特·曼尼福德，理由是“严重行为与治理问题”。 \ No newline at end of file diff --git a/links-posts/2668-ai-interconnect-and-photonics-research-report-on-m.md b/links-posts/2668-ai-interconnect-and-photonics-research-report-on-m.md new file mode 100644 index 0000000..add6c4c --- /dev/null +++ b/links-posts/2668-ai-interconnect-and-photonics-research-report-on-m.md @@ -0,0 +1,151 @@ +--- +title: "AI Interconnect and Photonics Research Report on Marvell, Lumentum, Coherent, Corning, Micron, and Nokia Valuation and Technology" +created: 2026-05-27 +updated: 2026-05-27 +type: summary +tags: [AI, interconnect, notes, photonics, telegram] +external: http://go/ui/posts/2668/ +--- + +# AI Interconnect and Photonics Research Report + +## Executive summary + +Across these six names, the cleanest conclusion is that **all six are genuinely relevant to AI data-centre buildout**, but they sit at very different points in the stack and deserve very different valuation frameworks. **Marvell** and **Micron** are the most semiconductor-levered beneficiaries. **Lumentum** and **Coherent** are the most direct photonics/optics component plays. **Corning** is the fibre-and-connectivity enabler rather than a chip company. **Nokia** is the systems-and-network-domain participant, with growing AI-cloud optical/IP relevance but a much broader telecom backdrop. citeturn11search11turn21view1turn13search4turn40search6turn38view3turn15search1 + +On competitive position, my ranking for **core technology defensibility in chips and optical communications** is: **Micron** (memory/HBM, scale, process roadmap), **Marvell** (custom AI silicon, switching, DSP/coherent interconnect), **Coherent** (broad photonics vertical integration), **Lumentum** (cloud/datacom laser and optical-component concentration), **Corning** (fibre/connectivity materials leadership, but less chip leverage), and **Nokia** (strong optical/IP systems and Bell Labs heritage, but lower pure-play AI exposure than the others). That ranking is about **technology exposure**, not expected stock returns. citeturn38view3turn43search14turn13search6turn21view1turn40search6turn15search0 + +On valuation, the market is already pricing in a large part of the AI buildout story. By current multiples, **Marvell, Lumentum, Coherent, Corning, and Nokia all look expensive versus their historical business quality and cyclicality**, while **Micron is the one name where valuation looks most defensible because forward earnings are rising much faster than near-term price/sales optics imply**. In other words, the current market is paying for AI scarcity and design-win optionality almost everywhere, but Micron has the strongest **earnings-backed** case among the group. citeturn44search1turn43search1turn43search2turn43search3turn41search5turn44search0 + +My research-only bottom line is: **Micron = best relative risk/reward**, **Marvell = high-quality but needs disciplined entry**, **Corning = reasonable strategic hold on pull-backs**, **Coherent = attractive business but valuation-rich**, **Lumentum = powerful momentum but the most demanding valuation**, **Nokia = improving strategically but probably ahead of itself after the rally**. These are research views only, not trading instructions. citeturn44search0turn44search1turn43search3turn43search2turn43search1turn42search23 + +## Comparative snapshot + +The table below condenses the highest-confidence current picture. Valuation ratios are current market-data snapshots around 27 May 2026; revenues and margins come from the latest annual report or latest reported quarter, depending on availability. “Current profitability” uses the latest clearly disclosed GAAP metrics when available; where a metric was not cleanly disclosed in the accessible source set, I mark it as unspecified rather than guessing. citeturn44search1turn43search1turn43search2turn43search3turn44search0turn41search5 + +| Company | Core AI-interconnect role | Latest FY revenue | Current profitability snapshot | Cash / debt snapshot | Current valuation snapshot | Recent analyst consensus snapshot | Research view | +|---|---|---:|---|---|---|---|---| +| **Marvell (MRVL)** | Custom AI silicon, Ethernet switching, DSP/coherent optics, interconnect | **US$8.19bn FY2026** | FY2026 GAAP gross margin **51.0%**; GAAP net margin c. **32.6%**; GAAP operating margin not cleanly reproduced in accessible summary | FY2026 year-end cash **US$0.95bn**; debt c. **US$4.06bn** at FY-end | Mkt cap **US$182.1bn**; EV **US$184.3bn**; **P/E 67.8x**, **Fwd P/E 54.1x**, **P/S 22.2x**, **EV/EBITDA 70.1x** | WSJ average target **US$157.03** vs current **US$196.33**; consensus **Buy** | **Hold / accumulate only on pull-backs** | +| **Lumentum (LITE)** | EML/CW lasers, optical components, cloud/datacom subsystems, optical circuit switching | **US$1.65bn FY2025** | FY2025 GAAP gross margin **28.0%**; GAAP operating margin **-10.9%**; latest Q3 FY2026 revenue and profitability improved sharply | Q2 FY2026 cash + ST investments **US$1.16bn** | Mkt cap **US$70.9bn**; EV **US$71.0bn**; **P/E 173.2x**, **Fwd P/E 57.5x**, **P/S 28.5x** | Barron’s/WSJ-style target snapshot: average near **US$1,127**; Public/Benzinga-type sources also show **Buy** consensus | **Avoid chasing / tracking position only** | +| **Coherent (COHR)** | Lasers, photonic devices, SiPh/InP components, transceivers, optical materials | **US$5.81bn FY2025** | Latest Q3 FY2026 GAAP gross margin **37.7%**; GAAP net income positive; GAAP operating margin not cleanly reproduced in accessible summary | Balance-sheet cash/debt not fully reproduced in accessible summary; EV suggests modest net debt | Mkt cap **US$74.6bn**; EV **US$75.6bn**; **P/E 159.2x**, **Fwd P/E 51.2x**, **P/S 11.3x**, **EV/EBITDA 57.6x** | MarketWatch/Fintel-style average target about **US$384**; recommendation **Overweight/Buy** | **Wait for better entry** | +| **Corning (GLW)** | Optical fibre, cabling, connectors, data-centre optical infrastructure | **US$15.63bn FY2025** | Latest Q1 2026 GAAP gross margin **36.9%**; GAAP operating margin **15.4%**; GAAP net margin c. **9.0%** | Q1 2026 cash **US$1.76bn**; debt **US$8.97bn** | **P/E 94.4x**, **Fwd P/E 58.2x**, **P/S 10.3x** | MarketWatch/Barron’s average target about **US$204.6**; recommendation **Overweight/Buy** | **Hold / add on weakness** | +| **Micron (MU)** | HBM, DRAM, NAND, data-centre SSDs, memory subsystem scaling | **US$37.38bn FY2025** | Latest Q2 FY2026 GAAP gross margin **74.4%**; GAAP operating margin **67.6%**; GAAP net margin c. **57.8%** | Q2 FY2026 cash + investments **US$16.7bn**; FY2025 debt **US$14.4bn** | **P/E 42.1x**, **Fwd P/E 9.6x**, **P/S 17.4x**, **EV/EBITDA 27.3x** | WSJ median target **US$600**, average **US$684.32**; Reuters notes a high target of **US$1,625** after UBS raised estimates | **Best relative idea in the group** | +| **Nokia (NOK)** | Optical transport, coherent/line systems, IP routing, data-centre switching, AI-cloud networking | **€19.89bn FY2025** | FY2025 reported gross margin **43.5%**; operating margin **4.4%**; net margin c. **3.3%** | Q4 2025 net cash + interest-bearing financial investments **€3.38bn** | Mkt cap about **US$91.9bn**; **P/E 98.2x**, **Fwd P/E 38.9x**, **P/S 3.9x**, **EV/EBITDA 29.9x** | MarketWatch/Barron’s average target about **US$11.6**; Fintel average target ~**US$10.2** | **Trim / don’t chase** | + +Sources: Marvell filings and market statistics citeturn11search0turn22view0turn44search1turn41search2; Lumentum releases and statistics citeturn16view2turn16view0turn43search1turn42search8; Coherent releases, presentation and statistics citeturn16view5turn16view4turn43search2turn42search13; Corning quarterly/annual releases and statistics citeturn40search3turn17view1turn43search3turn42search14; Micron releases, 10-K, statistics, and analyst pages/news citeturn35view2turn36view2turn39view1turn44search0turn41search3turn41news41; Nokia reports and valuation/target pages citeturn45view0turn16view6turn41search5turn42search23turn42search11turn44search4 + +## Technology and competitive positions + +**Marvell** is the most complete AI-networking semiconductor platform in this group. Its disclosed stack spans **custom silicon, Ethernet switches, DSP/coherent interconnect, SerDes, DPUs and broader interconnect IP**. In FY2026, data-centre revenue reached **US$6.10bn**, or **74%** of total revenue, confirming that Marvell is no longer just a diversified infrastructure chip vendor but a data-centre-first supplier. That is strategically important: hyperscalers increasingly want merchant-plus-custom options, and Marvell is one of the few vendors positioned across the **custom-XPU adjacencies**, **switching fabric**, and **optical/electrical interconnect** layers. The trade-off is that scale-up optimism is already capitalised into the stock. citeturn43search14turn23view1turn29view1turn44search1 + +Marvell’s near-term AI interconnect roadmap is especially strong. Recent public announcements included **NVLink Fusion ecosystem participation with NVIDIA**, the **Polariton acquisition to push optical performance scaling to 3.2T and beyond**, and a February 2026 launch claiming **industry-first 1.6T ZR/ZR+ pluggables and 2nm coherent DSPs** for AI scale-across networks. Those releases materially strengthen Marvell’s competitive position in both **inside-the-rack electrical connectivity** and **between-cluster optical transport**. citeturn11search4turn11search12 + +**Lumentum** is a focused optical/photonics supplier rather than a semiconductor generalist. Its Cloud & Networking business is built around **optical and photonic chips, components, modules and subsystems** used in intra-data-centre links, DCI, metro, long-haul and submarine networks. That includes the most commercially relevant AI-era building blocks: **laser chips and laser assemblies, line subsystems, wavelength-management systems, cloud transceivers, optical circuit switches, and co-packaged-optics-enabling components**. The business mix has shifted decisively toward cloud/datacom, which is the right place to be, but concentration risk is correspondingly high. citeturn21view1turn19view0turn33view0 + +Lumentum’s own commentary is notable because management explicitly identified **optical circuit switches**, **co-packaged optics**, and **cloud transceivers** as its “next major growth engines”. Q1 FY2026 commentary also pointed to strong momentum across **data centre, DCI and long-haul** markets, while the FY2025 annual report tied Cloud & Networking growth to **higher unit sales from cloud and AI/ML customers** and the Cloud Light acquisition. That is the right strategic direction, but it also means the stock now depends on sustained hyperscaler capex and clean optical-component ramps. citeturn19view0turn21view1 + +**Coherent** has perhaps the broadest underlying photonics platform in the group. Its portfolio combines **engineered materials, lasers, optoelectronic devices, compound semiconductor capability, VCSELs, InP/SiPh-related assets, optical transceivers and subsystems**, and it is increasingly framing the company around **Datacenter & Communications** plus **Industrial**. The key competitive advantage is vertical breadth: Coherent can participate from active optical chips and laser sources through packaging and systems-level modules. That breadth matters as AI networks push toward higher lane speeds and more complex optical architectures. citeturn13search4turn16view3turn16view5 + +Coherent’s most helpful disclosure for investors is that **Datacenter & Communications represented 72% of revenue in Q2 FY2026**, up from 63% in FY25 Q2 on the investor presentation chart, while management also highlighted a **next-generation 2D VCSEL-array-based solution for AI scale-up networks**. In practice, that means Coherent is moving away from being “just” a diversified photonics company and toward being a datacom/AI-optics compounder — but with meaningful industrial cyclicality still attached. citeturn16view3turn13search11 + +**Corning** is often misunderstood in AI discussions because it is **not a chip company** in the usual sense. Its relevance is in **optical fibre, cable, connectivity hardware, and dense optical infrastructure**, which form the physical layer under AI data-centre and campus buildouts. Corning’s own optical-communications materials explicitly target carriers, data centres and enterprise networks, and management has repeatedly said that **Optical Communications’ enterprise sales are being lifted by new GenAI products**. The competitive moat is manufacturing/process leadership and installed-base relationships, not semiconductor IP. citeturn40search6turn10search16turn40search15 + +Corning’s “Springboard” plan makes the AI linkage concrete. Management upgraded the plan and is targeting **US$20bn annualised sales run-rate by end-2026**, with much of the optical upside tied to high-density connectivity and AI data-centre products. That is strategically attractive because fibre demand is real and persistent; however, Corning still carries exposure to displays, environmental technologies and life sciences, so the stock should not be valued like a pure AI semiconductor name. citeturn40search9turn40search15turn40search3 + +**Micron** sits outside optical communications but squarely inside the AI-capacity problem. Its four reportable segments — **Cloud Memory, Core Data Center, Mobile & Client, and Automotive & Embedded** — show how the business has been reorganised around AI-era demand pools. The key technologies are **HBM, DDR5/LPDDR, GDDR, NAND, data-centre SSDs, advanced packaging and process-node transitions such as 1-gamma DRAM and G9 NAND**. In AI systems, memory bandwidth and capacity are no longer secondary inputs; they are hard constraints. That makes Micron strategically critical. citeturn38view3turn38view2turn37view0 + +Micron’s FY2025 10-K explicitly ties CMBU to **large hyperscale cloud customers and HBM for all data-centre customers**, and CDBU to **data-centre memory and storage solutions**. It also identified products such as the **9550 series SSD**, **6500 ION**, and early **PCIe Gen6 SSD** activity, while the Q2 FY2026 release showed a dramatic acceleration in both cloud memory and core data-centre revenue. Among the six names here, Micron has the strongest evidence that AI demand is already translating into **very large, margin-accretive earnings**, not just roadmap optionality. citeturn38view3turn38view2turn36view2 + +**Nokia** is the least “pure” AI play here, but it has become more relevant than the market used to assume. Nokia’s AI-data-centre case runs through **Optical Networks, IP Networks, pluggables, line systems, data-centre switching and Bell Labs / routing software capability**, now strengthened by the **Infinera acquisition** and the company’s stated push into **AI & Cloud design wins**. The strategic problem for Nokia is that these opportunities coexist with more mature and more cyclical mobile-network businesses. citeturn15search0turn45view0 + +Still, Nokia’s recent disclosures were better than many investors appreciate. In Q1 2026, management said Optical Networks grew **20%**, that it won **AI & Cloud design wins and orders for both pluggables and line systems**, and that IP progress is helping it expand its presence “inside the data centre”. The Q4/full-year 2025 report also said demand from AI & Cloud customers was driving optical and IP book-to-bill above one. Strategically, Nokia is no longer just a telecom-equipment story; the question now is whether that transition is durable enough to justify the stock’s rerating. citeturn16view6turn45view0 + +## Financial performance and revenue mix + +The clearest revenue-mix change in this group is at **Marvell**. Its last four reported quarters show a sustained increase in data-centre concentration: **Q1 FY2026 data-centre revenue US$1.441bn**, **Q2 US$1.491bn**, **Q3 US$1.518bn**, **Q4 US$1.651bn**. By Q4 FY2026, Marvell had also simplified reporting into **Data Center** and **Communications and other**, with Q4 revenue of **US$1.651bn** and **US$0.567bn** respectively. For FY2026, Marvell reported **US$6.10bn data-centre revenue** and **US$2.09bn communications and other**. Latest annual shipment geography was **China US$2.97bn**, **Taiwan US$1.66bn**, **United States US$1.17bn**, and **Other US$2.39bn**. This is a very AI-centric mix, but it also leaves the company highly sensitive to cloud/customer concentration and export-control complexity. citeturn28view3turn28view2turn29view0turn29view1turn23view1 + +**Lumentum** also delivered a significant revenue inflection, but with more concentrated product economics. The last four quarters by product type were: **Q4 FY2025 Components US$424.1m / Industrial Tech US$56.6m** under the old segment lens, then **Q1 FY2026 Components US$379.2m and Systems US$154.6m**, **Q2 FY2026 Components US$443.7m and Systems US$221.8m**, and **Q3 FY2026 Components US$533.3m and Systems US$275.1m**. FY2025 revenue by segment was **Cloud & Networking US$1.411bn** and **Industrial Tech US$234m**. FY2025 region mix was **Americas US$481m**, **Asia-Pacific US$1.001bn**, **EMEA US$164m**, with Hong Kong and Thailand especially important shipment locations. The takeaway is simple: the business is pivoting quickly toward cloud/datacom optical demand, and that is showing up in both growth and margin recovery. citeturn18view2turn19view0turn18view0turn18view3turn21view1turn21view2 + +For **Coherent**, the reporting change matters. The investor presentation showed a five-quarter progression in revenue distribution from **Datacenter & Communications US$905m / Industrial US$530m** in FY25 Q2 to **US$1,208m / US$478m** in FY26 Q2, and total revenue rose further to **US$1.81bn** in FY26 Q3. FY2025 full-year revenue was **US$5.81bn**. That mix shift is strategically bullish because Datacenter & Communications is structurally better aligned with AI than the industrial businesses, but it also means the stock is now partly pricing Coherent as a datacom growth company rather than as a mixed photonics manufacturer. Geography detail for the last four quarters was not cleanly reproduced in the accessible source set and is therefore left unspecified here. citeturn16view3turn16view4turn16view5 + +For **Corning**, the key financial point is less about segment-level decomposition and more about the profitability quality of the optical uplift. Q1 2026 delivered **US$4.144bn** GAAP sales, **36.9%** gross margin, and **15.4%** operating margin; full-year 2025 delivered **US$15.63bn** GAAP sales, **36.0%** gross margin and **14.6%** operating margin. Management also highlighted that Optical Communications enterprise sales grew **81% year on year** in Q2 2025 on demand for GenAI products. Full quarterly segment and geography schedules exist in Corning’s detailed financial statements, but they were not fully reproduced in the accessible summary sources I used here, so I am not inventing a more granular quarterly table. citeturn17view1turn40search3turn10search16turn40search2 + +For **Micron**, the last four reported quarters by business unit are unusually clear and very useful. **Q3 FY2025** revenue was **Cloud Memory US$3.386bn**, **Core Data Center US$1.530bn**, **Mobile & Client US$3.255bn**, **Automotive & Embedded US$1.127bn**. **Q4 FY2025** was **US$4.543bn / US$1.577bn / US$3.760bn / US$1.434bn**. **Q1 FY2026** was **US$5.284bn / US$2.379bn / US$4.255bn / US$1.720bn**. **Q2 FY2026** then surged to **US$7.749bn / US$5.687bn / US$7.711bn / US$2.708bn**. FY2025 revenue by business unit was **CMBU US$13.524bn**, **CDBU US$7.229bn**, **MCBU US$11.859bn**, **AEBU US$4.753bn**; geographic revenue by customer HQ was **US US$24.113bn**, **Taiwan US$5.672bn**, **Mainland China US$2.639bn**, **Other Asia Pacific US$1.913bn**, **Hong Kong US$1.138bn**, **Japan US$895m**, **Europe US$625m**. This is the strongest hard financial evidence in the group that AI demand is moving the entire P&L. citeturn35view2turn36view0turn36view2turn38view2turn39view1 + +For **Nokia**, the latest accessible detail shows FY2025 **reported net sales of €19.889bn** and Q4 2025 revenue of **€6.125bn**, with Q4 segment net sales of **Network Infrastructure €2.407bn**, **Mobile Networks €2.502bn**, **Cloud and Network Services €837m**, and **Nokia Technologies €384m**. In Q1 2026, management said group net sales grew **4%**, Network Infrastructure grew **6%**, and Optical Networks grew **20%**, with important AI-cloud design wins in pluggables and line systems. Because Nokia shifted its operating model entering 2026, a fully like-for-like last-four-quarters segment bridge is harder than it is for Micron or Marvell, but the strategic direction is still clear: AI-related optical/IP demand is now a real contributor. citeturn45view0turn16view6 + +## Valuation and fair-value framework + +The market is currently pricing **Marvell** as a premium AI-infrastructure semiconductor franchise: roughly **22.2x sales**, **54.1x forward earnings**, and **70.1x EV/EBITDA** on current market-data sources. That is a very demanding multiple set even for a company with dominant AI-interconnect positioning. My fair-value framework therefore assumes Marvell deserves a premium, but not an unconstrained one: roughly **14x–18x forward sales** or an equity value band that translates to about **US$174–US$240 per share** on a medium-term base case. On that basis the stock is **not obviously cheap**; it is closer to **fair-to-rich**, with upside dependent on sustained custom-silicon wins and optical attach-rate expansion. citeturn44search1turn11search0turn23view1turn41search2 + +**Lumentum** is the most difficult valuation in the group to support on conventional metrics. Current market-data sources show roughly **28.5x sales** and **57.5x forward earnings**. That is extremely rich for a component supplier, even one riding a powerful datacom super-cycle. The fair-value question is not “is the business improving?” — it clearly is — but whether the market is paying too far in advance for margin normalisation, co-packaged optics, and optical-switching scale. My justified range is materially below the current market cap unless FY2027 revenue and margins move dramatically above current trajectory; in practice I would only call it attractive on a substantial de-rating or after several quarters of proof that revenue above the recent run-rate is sustainable. citeturn43search1turn19view0turn18view3turn42search8 + +**Coherent** looks somewhat less extreme than Lumentum in sales terms, but still rich. Current market-data sources show about **11.3x sales**, **51.2x forward earnings**, and **57.6x EV/EBITDA**. For a business that still has industrial exposure, integration complexity, and reporting-transition noise, that is a premium multiple set. My fair-value framework would be more comfortable closer to **7x–9x forward sales** or **18x–24x EBITDA**, which implies the market is currently assuming a very long datacom runway with limited cyclicality. That may eventually prove correct, but today I think the stock offers a less attractive entry point than the business quality alone would suggest. citeturn43search2turn16view3turn16view4turn42search13 + +**Corning** also screens expensive relative to how the market normally values high-quality industrial/materials businesses. The stockanalysis snapshot shows about **10.3x sales** and **58.2x forward earnings**. That is a rich rating for a company whose AI upside is real but indirect, and whose portfolio still includes slower or less-AI-linked businesses. My fair-value range is therefore more conservative than the market’s current enthusiasm: roughly **6x–8x sales** for a high-quality optical-infrastructure compounder with improving margins. That still supports a constructive long-term view on the franchise itself, but it argues for patience on entry. citeturn43search3turn40search3turn17view1turn42search14 + +**Micron** is the outlier. On trailing optics it still looks expensive — current sources show about **17.4x sales** and **27.3x EV/EBITDA** — but the forward-earnings lens is completely different, because the market is discounting a very large jump in memory profitability. The current **forward P/E of roughly 9.6x** is the lowest and most understandable valuation in the group once you accept the AI memory thesis. My fair-value range is therefore built from forward earnings rather than trailing sales: on a mid-cycle-to-super-cycle framework I think Micron is **closer to fair than overvalued**, and it is the easiest name here to justify fundamentally. citeturn44search0turn35view2turn36view2turn41news41 + +**Nokia** screens cheaper on sales than the U.S. names, but its re-rating has still been aggressive. Current market-data sources show around **3.9x sales**, **38.9x forward earnings**, and **29.9x EV/EBITDA**. For a business still carrying telco cyclicality and margin variability, that is not obviously cheap. The market is now paying for the AI optical/IP pivot and for Infinera-related strategic improvement. My fair-value framework ends up near the broad analyst-target range — roughly the low teens in ADR terms — not far above, which is why I see the shares as **strategically improved but valuation-stretched after the run**. citeturn41search5turn42search23turn42search11turn45view0 + +## Research-only investment views + +**Marvell** is a high-quality strategic asset in AI networking, and I would be comfortable owning it over a **12–24 month** horizon — but only with valuation discipline. For research sizing, I would think in terms of a **2%–4% position** at most, accumulated on weakness rather than at today’s premium. My buy triggers would be: a material pull-back, evidence of another step-up in custom-silicon wins, and continued confirmation that optical/coherent products are broadening the data-centre moat. My trim triggers would be: hyperscaler digestion, export-control pressure on the China/Taiwan mix, or signs that custom-AI programmes are extending timelines. citeturn44search1turn23view1turn31view0 + +**Lumentum** is strategically well placed, but I would treat it as a **tracking position at most** until valuation becomes easier to defend. A sensible research allocation would be **0%–1.5%**, with the default choice being to wait. The purchase case would strengthen only if the company proves it can scale cloud transceivers, optical circuit switches and co-packaged-optics-related products while keeping margins in the improved range. My trim/avoid triggers are simpler: any sign that hyperscaler optical component demand is pausing, that cloud transceiver pricing is worsening, or that the systems mix becomes less favourable. citeturn19view0turn18view3turn43search1turn42search8 + +**Coherent** is one of the most interesting businesses in the set, because the technology base is broad and the datacom mix is improving, but I still prefer it as a **watch-list quality compounder rather than an aggressive entry today**. A **1%–2.5% research position** would be the upper end of what I would consider, and only after a better entry or more evidence that Datacenter & Communications can keep compounding above the industrial drag. I would buy more confidently if the company sustains revenue growth while delevering the income statement and simplifying the equity story. citeturn16view3turn16view4turn43search2turn42search13 + +**Corning** is the most acceptable “infrastructure derivative” of the group if you want AI-network exposure without taking direct silicon-cycle risk. The expected return is probably lower than Micron’s in a strong cycle, but so is the business-model volatility. For research purposes I would consider **1.5%–3%** on pull-backs, not on momentum spikes. The buy trigger is straightforward: proof that GenAI-related optical demand is translating into durable free cash flow and not just transient enterprise optical orders. The trim trigger is equally straightforward: if Optical Communications growth slows materially before the broader Springboard plan has delivered its margin targets, the market multiple could compress sharply. citeturn10search16turn40search15turn43search3turn42search14 + +**Micron** is my preferred name in the basket. It is still cyclical and therefore not “safe”, but it has the most attractive combination of **technology centrality, earnings leverage, and valuation support**. For research sizing I would be comfortable with **3%–5%** over a **12–24 month** horizon, subject to standard cyclicality discipline. My buy triggers are continued HBM/memory leadership, signed long-term agreements, and evidence that supply discipline is holding. My trim triggers are classic memory-cycle signals: aggressive industry capacity additions, price erosion, inventory rebuilds, or a weakening of data-centre demand intensity. citeturn44search0turn41news41turn38view3turn36view2 + +**Nokia** has improved strategically, but I do not think the post-rally risk/reward is as attractive as the narrative sounds. If I wanted exposure, I would keep it modest — **1%–2%** — and only on pull-backs closer to or below the current target-price cluster. The buy case depends on Optical and IP Networks sustaining AI-cloud momentum while the broader company avoids margin leakage. The trim case is that mobile/telecom softness reasserts itself before AI-cloud optical wins become large enough to dominate consolidated results. citeturn45view0turn16view6turn42search23turn42search11 + +## Timeline and representative IP appendix + +The timeline below summarises the most relevant disclosed milestones for AI data-centre interconnects over the last three years and the next 12 months. The milestones are drawn from company filings, earnings releases, and official investor materials. citeturn11search4turn19view0turn13search6turn40search15turn38view3turn45view0 + +```mermaid +timeline + title Major AI-interconnect milestones + 2023 : Lumentum adds Cloud Light to strengthen datacom modules + : Nokia begins integrating Infinera optical assets + 2024 : Micron scales HBM and data-centre memory positioning + : Corning begins seeing stronger GenAI optical demand in enterprise + 2025 : Marvell FY2026 data-centre revenue accelerates sharply + : Lumentum highlights optical circuit switches, co-packaged optics, cloud transceivers + : Coherent shifts revenue mix toward Datacenter & Communications + : Micron reorganises business units around cloud/data-centre demand + : Nokia Q4 sees AI & Cloud demand lift optical/IP order intake + 2026 : Marvell joins NVIDIA NVLink Fusion ecosystem + : Marvell launches 1.6T ZR/ZR+ pluggables and 2nm coherent DSP roadmap + : Marvell acquires Polariton for 3.2T+ optical scaling + : Lumentum posts record Q3 FY2026 revenue on cloud/datacom demand + : Coherent Q3 FY2026 revenue reaches US$1.81bn + : Nokia Q1 2026 reports AI & Cloud design wins in pluggables and line systems + Next 12 months : Watch HBM4 / next-gen packaging at Micron + : Watch co-packaged optics and OCS commercialisation at Lumentum + : Watch 1.6T/3.2T coherent optical roadmaps at Marvell + : Watch DC optics attach-rate and industrial mix at Coherent + : Watch AI fibre/connectivity conversion at Corning + : Watch AI optical and IP growth versus mobile drag at Nokia +``` + +Because none of these companies publicly publish an issuer-certified ranking of their most valuable patent families, the IP list below should be read as **five representative patent-family themes/search links per company that are closely aligned with the AI interconnect thesis**, not as a legally definitive “top five by value” list. The underlying claim — that these firms have meaningful IP depth in the named areas — is supported by their annual reports and product disclosures. citeturn43search14turn21view1turn13search4turn40search6turn38view3turn15search1 + +**Marvell representative patent-family themes**: [coherent DSP](https://patents.google.com/?q=assignee%3A(Marvell)+coherent+DSP), [PAM4/optical DSP](https://patents.google.com/?q=assignee%3A(Marvell)+PAM4+DSP+optical), [Ethernet switch fabric](https://patents.google.com/?q=assignee%3A(Marvell)+Ethernet+switch+SerDes), [co-packaged optics](https://patents.google.com/?q=assignee%3A(Marvell)+co-packaged+optics), [silicon-photonics interconnect](https://patents.google.com/?q=assignee%3A(Marvell)+silicon+photonics+interconnect) + +**Lumentum representative patent-family themes**: [EML / laser chip](https://patents.google.com/?q=assignee%3A(Lumentum)+EML+laser+chip), [wavelength management](https://patents.google.com/?q=assignee%3A(Lumentum)+wavelength+management), [optical circuit switch](https://patents.google.com/?q=assignee%3A(Lumentum)+optical+circuit+switch), [cloud transceiver optics](https://patents.google.com/?q=assignee%3A(Lumentum)+cloud+transceiver+optical), [co-packaged optics laser source](https://patents.google.com/?q=assignee%3A(Lumentum)+co-packaged+optics+laser) + +**Coherent representative patent-family themes**: [VCSEL arrays](https://patents.google.com/?q=assignee%3A(Coherent)+VCSEL+array), [InP photonic devices](https://patents.google.com/?q=assignee%3A(Coherent)+indium+phosphide+photonic), [silicon photonics packaging](https://patents.google.com/?q=assignee%3A(Coherent)+silicon+photonics+packaging), [optical transceiver modules](https://patents.google.com/?q=assignee%3A(Coherent)+optical+transceiver+module), [laser materials / photonic integration](https://patents.google.com/?q=assignee%3A(Coherent)+laser+photonic+integration) + +**Corning representative patent-family themes**: [low-loss optical fibre](https://patents.google.com/?q=assignee%3A(Corning)+low-loss+optical+fiber), [bend-insensitive fibre](https://patents.google.com/?q=assignee%3A(Corning)+bend-insensitive+fiber), [multi-fibre connector systems](https://patents.google.com/?q=assignee%3A(Corning)+multi-fiber+connector), [data-centre cabling density](https://patents.google.com/?q=assignee%3A(Corning)+data+center+fiber+cabling), [glass substrates / optical materials](https://patents.google.com/?q=assignee%3A(Corning)+optical+glass+substrate) + +**Micron representative patent-family themes**: [HBM packaging](https://patents.google.com/?q=assignee%3A(Micron)+HBM+packaging), [3D NAND architecture](https://patents.google.com/?q=assignee%3A(Micron)+3D+NAND+architecture), [DRAM process / EUV / 1-gamma](https://patents.google.com/?q=assignee%3A(Micron)+DRAM+EUV+memory), [data-centre SSD controller/system](https://patents.google.com/?q=assignee%3A(Micron)+data+center+SSD), [advanced memory module packaging](https://patents.google.com/?q=assignee%3A(Micron)+memory+package+module) + +**Nokia representative patent-family themes**: [coherent optical transport](https://patents.google.com/?q=assignee%3A(Nokia)+coherent+optical+transport), [pluggable optics / line systems](https://patents.google.com/?q=assignee%3A(Nokia)+pluggable+optics+line+system), [IP routing silicon](https://patents.google.com/?q=assignee%3A(Nokia)+IP+routing+silicon), [data-centre switching / SR Linux](https://patents.google.com/?q=assignee%3A(Nokia)+data+center+switching), [Bell Labs AI-native networking](https://patents.google.com/?q=assignee%3A(Nokia)+AI-native+network) + +## Open questions and limitations + +A few parts of the request are constrained by what the companies explicitly disclose in public filings and what was reproducible from the accessible source set. The biggest limitations are these: some firms changed reporting structures mid-stream, so **last-four-quarter segment series are cleaner for Marvell, Lumentum, Micron and Coherent than for Corning and Nokia**; **quarterly geography is not always fully reproduced in the accessible summaries**, so where a complete quarterly geography bridge was not available I used the latest annual geography or marked it unspecified; and **no company publishes an official “top five patent families by economic value” list**, so the patent appendix is explicitly framed as representative thematic family coverage rather than issuer-ranked top five patents. Those limitations matter for precision, but they do **not** change the main conclusions on technology position, AI exposure, or relative valuation. citeturn23view1turn21view1turn38view2turn45view0turn40search3 \ No newline at end of file diff --git a/links-posts/2669-美光闪迪西部数据ai存储与hdd供应紧张推高景气估值已非便宜股追高需谨慎评估安全边际.md b/links-posts/2669-美光闪迪西部数据ai存储与hdd供应紧张推高景气估值已非便宜股追高需谨慎评估安全边际.md new file mode 100644 index 0000000..e88d85d --- /dev/null +++ b/links-posts/2669-美光闪迪西部数据ai存储与hdd供应紧张推高景气估值已非便宜股追高需谨慎评估安全边际.md @@ -0,0 +1,73 @@ +--- +title: "美光、闪迪、西部数据：AI存储与HDD供应紧张推高景气，估值已非便宜股，追高需谨慎评估安全边际" +created: 2026-05-27 +updated: 2026-05-27 +type: summary +tags: [AI, notes, storage, telegram, valuation] +external: http://go/ui/posts/2669/ +--- + +下面按“截至美股 2026-05-26 收盘、2026-05-27 盘前数据可见”的口径看。先说结论：这三家都已经不是传统低毛利存储周期里的“便宜股”了，市场已经把 AI 存储短缺和多年合约吃进价格。真正的问题不是“有没有景气”，而是“现在追进去，是否还留有足够安全边际”。 + +**总判断** + +| 公司 | 业务核心 | 当前景气度 | 产能/订单可见度 | 估值判断 | 我的偏好 | +|---|---:|---:|---:|---:|---:| +| Micron / MU | DRAM、HBM、NAND、SSD | 极强 | 2026 供应紧，2027 扩产建设中，但未公开说 2027 全卖完 | 前瞻 PE 低，但 P/S、FCF 估值很贵 | 最稳，但不宜追高 | +| SanDisk / SNDK | NAND、企业 SSD、消费闪存 | 极强且爆发最快 | FY2027 bits 已有超过三分之一被 NBM 锁定，目标继续提高 | 前瞻 PE 看似便宜，股价和利润都像峰值周期 | 弹性最大，风险也最大 | +| Western Digital / WDC | HDD，尤其云/近线盘 | 极强 | 2026 HDD 基本卖完，已有 2027/2028 LTA | 当前估值不便宜，故事兑现度高 | 订单最硬，但股价也最满 | + +**1. Micron：最像“AI 存储核心股”，但现在不是便宜小票** + +Micron 最新 FQ2 FY2026 收入 $23.86B，去年同期 $8.05B；GAAP 净利 $13.79B，non-GAAP EPS $12.20，经营现金流 $11.90B。公司给 FQ3 指引更夸张：收入约 $33.5B，毛利率约 81%，non-GAAP EPS 约 $19.15。也就是说，利润弹性现在处在非常高的位置。([micron.gcs-web.com](https://micron.gcs-web.com/news-releases/news-release-details/micron-technology-inc-reports-results-second-quarter-fiscal-2026)) + +结构上，DRAM 是核心：FQ2 DRAM 收入 $18.8B，占 79%；NAND 收入 $5.0B，占 21%。公司披露 DRAM 价格环比上涨 mid-60%，NAND 价格环比上涨 high-70%，这说明当前盈利主要来自供需紧张和价格跳升，不只是出货量增长。([investors.micron.com](https://investors.micron.com/static-files/e089f8c0-065d-47b8-9d02-bfa863cdb357)) + +产能方面，Micron 的口径是“紧到 2026 以后”，不是“2027 已经全部卖完”。公司说 DRAM 和 NAND 供需都预计在 2026 年以后仍然偏紧；同时已经签了第一个 5 年 SCA 战略客户协议。扩产节奏上，Idaho 第一座 fab 初始 wafer output 预计在 2027 年中，Tongluo 现有 fab 有意义出货在 FY2028，新加坡 NAND fab 初始 wafer output 在 2028 下半年，新加坡 HBM advanced packaging 会在 2027 年贡献 HBM 供应。([investors.micron.com](https://investors.micron.com/static-files/e089f8c0-065d-47b8-9d02-bfa863cdb357)) + +估值上，MU 5/26 收盘 $895.88，市值约 $1.01T；TTM 收入 $58.12B、净利 $24.11B。表面上 forward PE 只有 9.56，但 trailing PE 42.14、P/S 17.38、P/FCF 98.27，RSI 74.86，股价 52 周涨了约 859%。([stockanalysis.com](https://stockanalysis.com/stocks/mu/statistics/)) ([stockanalysis.com](https://stockanalysis.com/stocks/mu/statistics/)) +我的判断：如果相信 FY2027 以后 HBM/DRAM 利润能维持，这不贵；如果毛利率从 80% 回落到传统周期水平，就很贵。MU 是三家里业务质量和战略地位最强的，但现在要按“高景气高估值”处理。 + +**2. SanDisk：盈利最炸裂，合约化转型最激进，但波动会非常大** + +SanDisk 已经从 WDC 分拆出来，2025-02-21 成为独立上市公司。([sandisk.com](https://www.sandisk.com/company/newsroom/press-releases/2026/2026-04-30-sandisk-reports-fiscal-third-quarter-2026-financial-results)) 最新 Q3 FY2026 收入 $5.95B，环比 +97%、同比 +251%；GAAP 净利 $3.615B，non-GAAP EPS $23.41；non-GAAP 毛利率 78.4%。公司 Q4 指引收入 $7.75B-$8.25B，non-GAAP EPS $30-$33。([sandisk.com](https://www.sandisk.com/company/newsroom/press-releases/2026/2026-04-30-sandisk-reports-fiscal-third-quarter-2026-financial-results)) + +分业务看，Datacenter 收入 $1.467B，环比 +233%；Edge $3.663B，环比 +118%；Consumer $820M，环比 -10%。也就是说，这次不是消费 U 盘/存储卡逻辑，而是企业 SSD、AI 数据中心 NAND 需求把公司利润模型重估了。([investor.sandisk.com](https://investor.sandisk.com/static-files/8ea78860-f8e5-4f1c-ada3-c554437d6281)) + +最关键的是 NBM。SanDisk 已签 5 个多年供货协议，其中 Q3 签的 3 个合同最低合同收入约 $42B；5 个协议合计有超过 $11B 金融担保，包含预付款和第三方金融工具；这些协议覆盖 FY2027 超过三分之一 bits，公司还希望继续提高，管理层甚至说可以超过 50%。([fool.com](https://www.fool.com/earnings/call-transcripts/2026/04/30/sandisk-sndk-q3-2026-earnings-transcript/)) +所以答案是：SanDisk 的 2027 不是全部卖完，但已经有超过三分之一 FY2027 bits 被硬性客户承诺锁住，而且未来几个季度可能继续上升。 + +估值上，SNDK 5/26 收盘 $1,589.55，市值 $235.4B；forward PE 9.80，trailing PE 53.51，P/S 17.85，P/FCF 52.78，过去 52 周涨幅超过 4,100%。([stockanalysis.com](https://stockanalysis.com/stocks/sndk/statistics/)) +我的判断：SNDK 是“看 forward EPS 很便宜，看历史周期和股价位置很吓人”。如果 NBM 真的让 NAND 变成更像长期合同制造/平台供应链，它可能仍有空间；但如果 NAND 行业重新过度扩产、价格下行，80% 毛利率不可能长期当常态。 + +**3. Western Digital：订单最硬，HDD 供给最紧，但估值也最不便宜** + +WDC 现在基本是 HDD 公司，SanDisk 分拆后不再并表。([westerndigital.com](https://www.westerndigital.com/company/newsroom/press-releases/2026/2026-04-30-wd-reports-fiscal-third-quarter-2026-financial-results)) Q3 FY2026 收入 $3.337B，同比 +45%；non-GAAP 毛利率 50.5%，non-GAAP EPS $2.72，自由现金流 $978M。Q4 指引收入 $3.65B 左右，non-GAAP 毛利率 51%-52%，EPS $3.25。([westerndigital.com](https://www.westerndigital.com/company/newsroom/press-releases/2026/2026-04-30-wd-reports-fiscal-third-quarter-2026-financial-results)) + +WDC 的优点是供给可见度最明确。Q2 电话会口径里，CEO 表示 2026 年 HDD 基本卖完，有 top 7 客户 firm POs，并且 top 5 里已有两个客户签到 2027、一个签到 2028 的 LTA。([fool.com](https://www.fool.com/earnings/call-transcripts/2026/01/30/western-digital-wdc-q2-2026-earnings-transcript/?utm_source=openai)) Q3 deck 也显示云业务占收入约 89%，nearline exabytes 持续上升，总 exabytes 从 Q3 FY2025 的 166EB 升到 Q3 FY2026 的 222EB。([investor.wdc.com](https://investor.wdc.com/static-files/5b2d41c1-7d45-4575-b9ea-c51424dbffeb)) + +估值上，WDC 5/26 收盘 $524.65，市值 $180.84B；trailing PE 28.73，forward PE 33.79，P/S 15.36，P/FCF 62.25，52 周涨幅约 946%，RSI 70.78。([stockanalysis.com](https://stockanalysis.com/stocks/wdc/statistics/)) +我的判断：WDC 的订单质量最好，但股票并不便宜。它更像“供给锁定、盈利确定性提升，但价格已经给了很高确定性溢价”。新钱追 WDC，要特别小心 2027 以后 HDD 客户议价和 Seagate 竞争。 + +**谁更值得买？** + +我的排序会分两种： + +| 维度 | 排名 | +|---|---| +| 业务质量和长期战略地位 | MU > SNDK > WDC | +| 2026 产能锁定明确度 | WDC > SNDK > MU | +| 盈利弹性 | SNDK > MU > WDC | +| 当前估值安全边际 | MU/SNDK 取决于 forward EPS 是否可信，WDC 最不便宜 | +| 风险调整后偏好 | MU 第一，SNDK 第二，WDC 第三 | + +我不会把它们当“便宜股”买。更准确地说：它们是“forward 看便宜、trailing/销售额/现金流看很贵”的高景气周期股。现在买入的核心风险不是今年业绩差，而是市场已经按 2027 继续紧缺来定价，一旦订单增速、价格涨幅、毛利率任何一项放缓，股价可能先跌，哪怕财报仍然很好。 + +**操作判断** + +现在大仓位追高，我不喜欢。更合理的是等回撤或分批： +MU 可以作为核心观察对象，回撤到 forward PE 仍低、但技术面不再过热时更舒服。 +SNDK 适合高风险仓位，重点盯 NBM 覆盖率是否从三分之一走向 50% 以上，以及 Q4 后 RPO 是否继续扩大。 +WDC 适合等估值消化，除非 2027/2028 订单进一步变成更大规模、价格更固定的公开承诺。 + +最重要的跟踪点：MU 的 2026-06-24 附近下一次财报、SanDisk 下一季 RPO/NBM 覆盖率、WDC 是否把 2027 HDD capacity 进一步锁定，以及三家公司毛利率是否从“异常高”开始松动。 \ No newline at end of file