与「CPO」相关的搜索结果

半导体\CPO\芯片\封测，这些 AI 概念都指什么？今天一条视频给你讲明白 这视频太干了，有种知识自己进入大脑的感觉了😂 原以为炒股是赌博，没想到炒股是读博🧐

玻璃桥对下一代光通信的影响 效率为先，原有接入800G/1.6T的8根光纤阵列单元(FAU)被取代，提升接入效率。 对光模块整体价值量的影响，要达到51.2Tbps/102.4Tbps的吞吐量需求，都会配置超过16个光模块给一个芯片，cpo/npo/光芯片的需求还会上升。 玻璃桥外的光纤数量不变，光纤的需求量和光模块同步提升。 玻璃桥采用化学方法蚀刻，替代原有工艺。

玻璃基板，康宁 $GLW 光纤，康宁 $GLW 玻璃光学互连CPO，康宁 $GLW 英伟达投资扩产，康宁 $GLW 康宁 $GLW 简直是下一代AI算力的底座，中了狗屎运。 黄仁勋也是190买的， 长个俩三倍到个5000亿美金不过分吧。

怎么我一买进美股就跌了 我成本337的MRVL 要哭了 刷到白毛股神（Serenity）的开源 Serenity skill，正好在GitHub上， 感觉可以站在前辈的视野去分析一下MRVL，就直接丢进我目前在用的EasyClaw AI给我整了个挺详细的报告，按白毛股神那套“找最难扩产瓶颈”的思路拆了一遍， 我挑重点说说： MRVL卡在哪： 它同时卡住了AI数据中心两个硬活：定制AI加速芯片（ASIC） + 光互联（DSP、CPO、硅光）。 客户认证周期长、切换成本高，不是随便谁都能挤进来的。 Amazon Trainium、Google Axion都有它份，Celestial AI和XConn的收购又把光互联护城河加厚了。 业绩看： 今年营收一路超预期，FY27目标接近110亿刀，订单能见度强，每季度guidance都在往上调。数据中心业务占75%左右，增长很猛。但高位接盘的现实： 股价现在大概在280-308区间（我接的时候接近高点），估值不便宜（forward PE 70-90x），已经price in很多好消息。微软那边有传闻可能部分转给Broadcom，这是最大风险点。AI给的操作建议（挺实在，没画大饼）：如果你还信AI互连和定制ASIC是真瓶颈，就别全卖，留核心仓位。 但也别死扛：反弹一点先减20-30%，止损位设好。 想加仓等明显回踩 + 新催化（比如Google新chip官宣、CPO量产进展、Q2财报微软收入没大掉）。 整体仓位别太重，分散风险。 风险（AI直说）： 微软ASIC流失可能是真雷；毛利率还在下滑；高估值没缓冲； AI资本开支如果边际放缓就危险。 看完以后心情好多了，至少知道自己在干嘛 白毛股神这套卡点思维还是蛮适合普通人的，每次想要买或者不懂的时候， 可以跟这个skill对话一下了解具体情况 让自己不至于那么心慌 工具链接扔评论区了 👇

2028 后限制 AI 算力硬件的不再是晶体管，而是封装！是怎么把多颗 die 封进一个 240mm 巨块并喂进 4kW 的电，然后再把热和光导给出去⚡️配图来自日本 3DIC Research Lab 「先进封装技术研讨会」我转译成了中文，中间那块剖面图，是一个超大尺寸的 3D 异构集成封装，当然这个路线图略显激进😜 - Co-Package Optics（FAU / EIC / PIC, 2λ/4λ）：光直接进封装，铜互连到头 → 这是 CPO 从"会发生"被画成 2028 默认路线的又一独立信源； - >4kW Cooling, Direct-to-Chip → Direct-to-Silicon, 直接液冷：单封装功耗破 4 千瓦，液冷从"贴芯片"进化到"贴硅"； - Vertical Power Delivery, High-voltage DC, PMIC / IVR / DTC：供电从机柜压进封装内部，垂直供电 + 高压直流 + 就近稳压 → 电已经压到封装级，催生封装内供电器件； - D2D / UCIe / micro-bump 3D / Cu direct 3D：chiplet 标准化互连、铜直连 3D 堆叠； - >10 reticle Interposer（Panel-level）/ 240×240mm² Substrate：中介层超 10 个光罩、改用面板级工艺，超大多层基板 → 封装设备/材料（日韩台供应链），基板成新瓶颈； 这么大封装最易翘曲，良率/可靠性是隐性门槛。但传统基板已撑不住 >200×200mm² 的 mega-size 封装，需要全新 building block！

I think something to highlight also is not all my ideas are green, especially on short term timeframes! My core three themes are Neoclouds (Energy), Memory, and Photonics. And I'm glad I chose the literal top performers for each segment from $NBIS to $EWY leaps to $SIVE. However, I still have pretty large losses following the false analyst report on CPO delays that $NVDA refuted: That nuked 3 of my TW CPO longs from Foci, Msscorp, Xintec and others. (which are heavily red). Shunsin / Win Semi are holding up much better though. Some of the other ones in Japan that I've mentioned like Towa, Harmonic Drive, NCI, etc. are performing much better following short-term volatility. As for Korea yes I have PTSD for from Auros and now Foosung that are both both red (I didn't own 093370 though, just got PTSD watching price action). Idk if I'll touch Korea again, just way too volatile. But I still think those will end up green eventually following Sk hynix/samsung qualifications + HVM, and the future japanese supply chain shutdown. I also did change some of my previous long ideas like $XLU following the Iran War nuking all chances of rate cuts from 3-4 down to 0, and that didn't play out too well. Three software names I mentioned previously like $TTD, $META ended up red. $SNAP, I also changed my thesis after noticing the endless SBC accounting methods. $RDDT idea was finally in the green from $130 -> $170 after a long time. But I think the vast majority of my ideas like $MRVL, $NBIS, $ARM, $INTC, $MU, $LITE, SK Hynix, Samsung, for larger cap. Down to $AXTI, $LITE, $AAOI, $RPI, $IQE, and others for smaller cap positions directionally play out pretty well. With random stuff like $SIMO, $HPS.A, $TSEM, $AEHR, $LPK, $SOI, $ALRIB, and so on all ended up turning out aight too. The blended average is kinda overwhelmingly green since majority of my long ideas are triple digit YTD. But I've definitely missed a few. Also entry point is really important too... I mentioned $AAOI at $30 or $AXTI at ~$13, but not everyone has the same entry point. So if someone bought AOI at $220 and it dropped to $160, I'd feel bad. But regardless, I'd prefer to judge how ideas play out on medium term timeframes over a few months rather than a few weeks.

朋友们，千万别碰高位半导体。 存储、CPO、PCB 这些，现在看起来全是时代红利，实际上很可能就是高位出货，诱惑你上去接盘。 泡沫一旦破了，从高处摔下来真的会死得很惨。 老登股就不一样了。 白酒、医药、消费、旅游、银行、券商、保险、房地产，空间上已经相对安全了，剩下只是时间问题。 这个时候就应该加仓老登股，做时间的朋友。 科技的风险我来面对🤪

New reports that $AMD is scrambling for CW laser supply. And is negotiating large-scale purchase orders for CW Lasers to ensure its production capacity is not constrained by $NVDA (Trendforce) Obvious CW laser beneficiaries: - $SIVE (AMD went to GFS for CPO, Sivers reference laser level) - $AAOI (Rosenblatt analyst checks) Lumentum/Coherent are kinda booked out way into 2028 as well. Lumentum is especially constrained for CW capacity already from existing EML contracts (so they probably are buying from Sumitomo/Furukawa and co). Maybe Macom and Japanese giants still have spare capacity. (disclosure, own aaoi/sivers). I predicted this last year and said hyperscalers should go more upstream to secure capacity... at laser levels, epiwafer levels, or even inp substrate levels. To not get bottlenecked by Nvidia.