供应链分析
产业链结构、上下游与瓶颈环节研究 · 共 1413 条 · 滚到底自动加载更多
-
DDR2内存Q2/Q3价格大涨指向特定市场信号。
@embgaspbaud 第二季度/第三季度双倍数据速率2(DDR2)内存价格的巨大涨幅指向了这一点 https://t.co/CS770NbwNz
英文原文
@embgaspbaud The enormous Q2/Q3 DDR2 hike pricing points to it https://t.co/CS770NbwNz
-
DDR4短缺引发ODM降级设计，老厂退出加剧级联瓶颈。
“级联瓶颈”的概念简直太酷了。由于DDR4供应冲击和成本问题：许多原始设计制造商(ODM)正在从DDR4向下重新设计至DDR3，再至DDR2。但即使是像华邦电子($WIN)这样的老牌厂商也在退出DDR2市场。这使得瓶颈问题更加恶化。
英文原文
The concept of “cascading bottlenecks” is just super cool. Because of DDR4 supply shocks and costs: Lot of ODMs are redesigning downward from DDR4 -> DDR3 -> DDR2. But even the legacy players themselves like Winbond are withdrawing from DDR2. Making the bottleneck even worse
-
光收发器需求预测从5000万上调至8000-9000万。
@equityclimb1 是的，我错过了这句引言：“行业预测将消耗5000万个光收发器(transceiver)单元。现在我们听到的数字是8000万、9000万个。” https://t.co/WWFP09AP74
英文原文
@equityclimb1 Yeah I missed this quote: "The industry is forecasting 50 million transceiver units to be consumed. Now we are hearing the number 80, 90 million" https://t.co/WWFP09AP74
-
Ayar等巨头选用Sivers激光器，验证其供应链核心地位。
@Meatusmax @adiradir734145 好问题，主要是关于 $SIVE 的第三方评论+行动。Ayar 可能对我而言信息量最大，他们从官网供应链板块移除了 $LITE / $MTSI，仅单独展示 Sivers 作为激光器供应商。Ayar 副总裁也称 Sivers 是“赋能我们光学 I/O 解决方案的关键”，这是相当强烈的看多信号。还有 $JBL 在选定 Sivers 后构建了 1.6T LRO（长距光学），这非常罕见，并在一次壁炉对话中炫耀其技术护城河。$GFS 也选择 Sivers 作为参考激光器并在演示文稿中展示。感觉所有领先玩家都在因某种原因推进 Sivers？我假设这些领先公司做了尽职调查(DD)。
英文原文
@Meatusmax @adiradir734145 Good question, it's mainly third party commentary + actions around $SIVE. Ayar probably speaks volumes to me most removing $LITE / $MTSI from their supply chain section of their site. Then just featuring Sivers by itself for the laser suppliers. VP at Ayar also said Sivers was ... "essential to powering our optical I/O solution", which is pretty high signal commentary. Then you have $JBL that built a 1.6T LRO after selecting Sivers, which is very rare. And bragged about their technological moat at one of the fireside chats. Then $GFS choosing Sivers as the reference laser too and featuring them in their presentations... Feels like all your leading players are moving forward with Sivers for some reason? And I'd assume these leading companies did their DD
-
FAU产能不足或成CPO瓶颈，印证此前对NVDA/TSM供应链的预判。
我知道大家都在关注那些模仿南希·佩洛西的跟单者让 $BE 暴涨 11.83%。或者特朗普…… 但你们知道还有什么很酷的吗？ 共封装光学(CPO) 的 FAU 瓶颈。 我记得今年早些时候说过，FAU/Foci 可能会成为 $TSM / $NVDA CPO 路线图未来的瓶颈…… 这是支持这一观点的媒体评论： “据悉， Largan Precision (3008)、FOCI (3363) 和中国的 TFC 已进入 $NVDA FAU 供应链” “基于今年预计出货的 20,000 台 Spectrum-X 系统，这些系统中安装的光引擎将需要大约 60 万至 70 万个 FAU” “2027 年，需求预计将达到数百万台，而现有供应链产能目前无法满足需求。”（联合报，中文翻译为英文有点偏差） 咳咳，“现有供应链产能目前无法满足需求” 现在有点早，因为大规模生产预计要到 2027 年下半年，但读到这些更新还是很酷。
英文原文
I know everyone's focused on Nancy Pelosi copytraders sending $BE up 11.83%. Or Trump... But you know what else is cool? FAU bottlenecks for CPO. I remember saying earlier this year FAU/Foci might be a future bottleneck around $TSM / $NVDA CPO roadmaps... This was some media commentary supporting that: "It is understood that Largan Precision (3008), FOCI (3363), and China’s TFC have entered $NVDA FAU supply chain" "Based on an estimated 20,000 Spectrum-X systems shipped this year, the optical engines installed in those systems would require approximately 600k-700k FAUs" In 2027, demand is expected to reach several million units, and existing supply chain capacity is currently unable to meet demand." (UDN, mandarin -> english translation is kinda off) Cough cough, "existing supply chain capacity is currently unable to meet demand" Tad early right now since mass production is expected H2 2027, but cool to read updates around this.
-
AAOI避签LTA以保灵活性，虽显意外但有望借短缺获定价权。
@siflower 不完全是，我认为 $AAOI 表示他们避免签署长期协议(LTA)，以免产能过多绑定单一客户而不得不拒绝其他客户。如果签署了带预付款的长期协议，那么在那种情况下可能就不需要如此大规模的自动取款机(ATM)增发。考虑到 $NVDA 的长期协议对 $LITE / $COHR 效果很好，这个决定有点令人惊讶。不过我假设随着时间推移，由于当前激光器/光模块短缺，AOI 会获得更大的定价权。
英文原文
@siflower Not exactly, I think $AAOI stated they avoided LTAs so they wouldn't have too much capacity committed to one customer and have to turn away others. If they signed LTAs with prepayments, then that in scenario there probably wouldn't need to be such sizeable ATMs. Decision was a bit surprising though considering how well $NVDA LTAs turned out well for $LITE / $COHR. I would assume though AOI gets more pricing leverage as time goes on due to current laser/transceiver shortages.
-
英伟达投资Nebius后，NBIS获Groq3 LPX优先量产权。
$NVDA，继其$200亿交易后……宣布Groq3 LPX现已全面投产。 $NBIS 通过 Nebius token factory 成为首个采用者。 - 在运行 Gemma 4 31B 的 Artificial Analysis 基准测试中，达到 3,400 tok/s，支持 10 万 token 上下文 - 响应速度比最近的替代品快 4 倍 引起我注意的是……英伟达对 Nebius 投资$20 亿后的一份旧新闻稿。 其中提到将支持其“早期采用英伟达最新一代”架构。 看来我们在这里看到了这种特殊待遇。
英文原文
$NVDA, following their $20B deal... Announces Groq3 LPX now in full production. With $NBIS the first to adopt through Nebius token factory. - 3,400 tok/s in Artificial Analysis benchmarking running Gemma 4 31B, with 100k token context - 4x faster responsiveness than the nearest alternative Thing that caught my attention.. was a former press release after Nvidia's $2B investment into Nebius. That it would support their "early adoption of Nvidia's latest generation" architectures. Guess we're seeing that special treatment here.
-
小鹏拆分机器人部门融资，人形机器人及物理AI领域融资热潮兴起。
$XPEV 拆分其机器人部门，并以超过 63 亿美元的估值融资逾 9 亿美元。 此前，Unitree 已完成首次公开募股(IPO)，目前市值约为 350-360 亿美元。 此外，还有由 $NVDA / $AMZN 支持的 Agility，其投前估值为 25 亿美元。我认为，Agility 在商业化方面实际上比小鹏汽车走得更远。 无论如何，人形机器人 + 物理人工智能(Physical AI)领域的融资正在真正起飞。
英文原文
$XPEV carves out its robotics unit and raises >$900M at $6.3B+ valuations. This is following’s Unitree IPO, which now has a ~$35-$36B MC. Then there’s $NVDA / $AMZN backed Agility… at a $2.5B premoney valuation. Which I’d argue is actually further ahead in commercialization than Xpeng. Regardless, humanoids + physical AI sector funding is really taking off.
-
美国政府简报显示政策聚焦太空发射、采矿及多晶硅供应链。
最近阅读美国政府的事实简报（每周几份）很有趣。本周早些时候是“太空运输的黄金时代”，$SPCX 像 Faker 一样强力支撑着发射领域。$RKLB 和其他火箭公司作为 2030 年 1000 次发射/再入目标的受益者。本月我们还有：无人机组件关税、美国采矿/关键矿产交易、多晶硅供应链关税等等……所以看到美国政策聚焦于哪些板块非常有帮助。
英文原文
It's been fun to read these US Gov fact sheets (few a week) recently. Earlier this week was "Golden Age of Space Transportation", with $SPCX hard carrying the launch sector like Faker. And $RKLB + other rocket companies as beneficiaries of the 1K launch/reentry goal by 2030. This month we also had: - Drone component tariffs - American Mining / Critical Mineral deals - Polysilicon supply chains tariffs and a lot more... So helpful to see what sectors US policy is focusing on.
-
硅光子+CW主导CPO，激光元件供不应求，2027-28年或现短缺。
@OnurKocerr 不，EML、CW、microLEDs 都是并行路线，硅光子+CW 是 2027-2028 年的主要商业化重点，尤其是共封装光学(CPO)。如果有的话，内存/光互连增加了激光供应商（包括 CW DFB）的总可寻址市场(TAM)。这些已经供不应求。因此，当 2027-2028 年到来时，我们可能会看到另一场类似内存的短缺……鉴于你的前三名未来两年已售罄，且像 $AAOI 这样的独立激光产能上线后正被内部用于光模块而非出售。
英文原文
@OnurKocerr No, EML, CW, microLEDs are all parallel tracks, with silicon phonetics + CW being the main commercial focus for 2027-2028, especially for CPO. If anything, having memory/optical links adds another material TAM increase across laser suppliers including CW DFBs. Which are already in short supply. So we'll likely see another memory style shortage when 2027-2028 comes about... Given your top 3 are already sold out for next 2 years, and independent laser capacity coming online like $AAOI are now being used internally for transceivers rather than being sold.
-
POET获Lumilens订单，但Marvell自有IP或致其被Celestial项目排除。
@tozzi21 $POET 正在向 Lumilens 供货，Lumilens 是另一家顶级前二超大规模云服务商（拥有数十亿美元客户协议）的供应商。 光宝科技 (Lite-On) 是另一家主要参与者。 我只是说，鉴于 Marvell 在 Poet 提供的同一层级拥有自己的封装/集成 IP，他们很可能被 $MRVL Celestial 设计排除在外。
英文原文
@tozzi21 $POET is supplying Lumilens, which is another top-2 hyperscaler supplier (with billions in customer agreements). Lite-On is another major player. I'm just saying they got likely designed out $MRVL Celestial since Marvell has their own packagig/intergration IP in the same layer that Poet is offering.
-
推测SK海力士CPO供应链涉及Ayar/Celestial及上游SIVE等。
既然大家都在愉快地猜测 $SKHY 共封装光学(CPO) 路线图供应链。 我来猜一下：SK 海力士正在评估 Celestial/Ayar。随后将重点评估第一代部署后的 microLEDs（微发光二极管）。 这两家公司其实已经在这方面工作了一段时间。 - Ayar 出现在 SK 海力士官网上，语境是光 <-> 内存链接。 - Marvell 正积极与 SK 海力士合作定制内存解决方案（猜猜现在谁拥有 Celestial）。 Celestial/Ayar 也被 SK 海力士链接的 Nature 论文引用，这在技术相关性上提供了一些佐证： (121): Stojanovic, V. A UCIe optical I/O retimer chiplet for AI scale-up... 这是 Ayar 的引用（Ayar 的 teraphy 光 I/O 芯片组），顺便提一下 Stojanovic 是 Ayar Lab 的联合创始人。 (123): Winterbottom, P. Photonic interconnect for accelerated computing celestial AI photonic fabric module (咳咳，Celestial) 还有更多线索，但我想保持篇幅简短。 关于更多上游受益者： -> 我仍然认为 $SIVE 是最清晰的受益标的，因为他们很可能同时供应给 Celestial/Ayar。 中介层/封装 IP 层... 人们曾猜测 $POET，但 Marvell 可能在宣布后今年就通过垂直整合将他们淘汰了。 鉴于他们在 Celestial 被收购前就拥有自己的 SiPH 中介层技术/集成 IP（例如 2023 年展示过硅光中介层） -> AMS Osram（我不持有），看起来他们正在这一领域共同开发，因为他们今年特意指出 HBM <-> 光学对他们来说是一个 TAM（总可服务市场）增长点。 TLDR：我认为 Ayar/Celestial 是 SK 海力士 CPO 路线图的候选者，然后 microLEDs 可能在 2029 年左右实现第一代后的商业化。（这纯属猜测） 顺便给量子点粉丝一点甜头，这篇论文引用了它 5 次。
英文原文
Since people are having fun speculating on $SKHY CPO roadmap supply chains. I'm gonna do my own guess and say SK Hynix is evaluating Celestial/Ayar. Then doing heavy evaluation into microLEDs past first-gen deployments. Both have been kinda working on it for awhile. - Ayar shown up in SK hynix's own website in the context of optical <-> memory links. - Marvell is actively working with SK Hynix in custom memory solutions (and guess who owns Celestial now). Celestial/Ayar are both also cited in SK Hynix's linked Nature Paper, which helps a bit with technical relevance: (121): Stojanovic, V. A UCIe optical I/O retimer chiplet for AI scale-up. In 2025 IEEE Hot Chips 37 Symposium (HCS) 1–22 (IEEE, 2025) - This is Ayar reference (Ayar’s teraphy optical I/O chiplet), Stojanovic is Ayar Lab's co-founder btw. (123): Winterbottom, P. Photonic interconnect for accelerated computing celestial AI photonic fabric module (cough cough Celestial) There's more breadcrumbs out there, but wanted to keep this relatively short-form. For more upstream beneficaries: -> I still think $SIVE is the cleanest read through since they're likely supplying to both Celestial/Ayar. The interposer/packaging IP layer... People were speculating $POET, but Marvell probably vertically integrated players them out this year after the announcement. Given they've had their own SiPH interposer tech/integration IP before Celestials acquisition (eg. presented a silicon-photonics interposer back in 2023) -> AMS Osram (which I don't own), seems like they're co-developing in this area given they randomly went out of their way this year to say HBM <-> optics was a TAM increase for them. TLDR: I see Ayar/Celestial as candidates for SK Hynix's CPO roadmap, then microLEDs being commercialized past gen-1 maybe 2029. (this is all speculation) Just to throw a bone to quantum dot bros, this got cited like 5 times.
-
光模块需求极高且供应链严重短缺，认为板块跑输荒谬，看好行业前景。
非常积极。我的观点是，近期光模块板块从 $AAOI 到 $SIVE 的跑输表现简直荒谬。 需求可见性……简直高得愚蠢。 AOI：“即使合并 AOI 和 Coherent 的项目，未来三年（至2029年）仍很难满足客户需求。” Elazr 总经理：“整个光供应链正面临严重短缺。”“短缺将持续到未来几年。” Sivers CEO 对磷化铟(InP)激光器需求失衡的说法相同，预计未来3-5年持续。 关于 $LITE、$MTSI 和其他评论可以列举更多。 我们甚至还没触及 1.6T、NPO、共封装光学(CPO)规模化以及光存一体（如 SK Hynix 所示）的拐点。 然而行业已受限于 EML/CW 及 PD/TIA/DSP、收发器等上游组件，随着 CPO 规模化，FAU 等也将成为瓶颈…… 我个人对这一切的发展感到极度舒适，只是有点困惑市场为何不会算未来一两年的账。
英文原文
Very positive. My opinion is that the optical sector underperformance recently from $AAOI to $SIVE has been absurd. The demand visibility... is just way too stupidly high. AOI: “Even combined AOI, Coherent program altogether, is still very tough to meet the customer demand in the next 3 years” (2029) Elazr GM: "entire optical supply chain was facing major shortages." "The shortage will continue to the next few years" Sivers CEO said the same thing about InP laser demand imbalances expected for the next 3-5 years. Can go on and on about $LITE, $MTSI and other comments. We haven't even hit the inflection point with 1.6T, NPO, CPO scale out/up, and optics with memory (as seen with SK Hynix). Yet the industry is already bottlenecked by EML/CW and all your other upstream components from PDs/TIA/DSPs, transceivers, and soon FAU + others when CPO scales… I'm personally extremely comfortable watching this all play out, but just a little confused that markets don't know how to math a year or two ahead.
-
AI资金回流高流动性板块，半导体供应链多处出现涨价与瓶颈。
我很高兴杠杆交易者们从AI股票撤出，转回高流动性板块和生物科技（如$MRNA），这样我们或许能更快看到复苏。 除此之外，一些要点摘要： - $EWY上涨，主要得益于SK海力士的回购（他们利润太丰厚了），以及三星晶圆代工约10-15%的涨价。 - $SKHY发布共封装光学(CPO)路线图？看来他们在做光子中介层与内存连接的技术，我得深入看看。 这可能会带来更大的潜在市场(TAM)扩张，以及更多供应链受益者（如激光器、光子集成电路(PICs)、封装）。 - $YMTC继长鑫存储(CMXT)成功IPO后，计划下季度寻求IPO。 中国IPO市场正从宇树科技到内存制造商全面爆发。 美国应该采取类似做法，而不是让公司在万亿估值时IPO。 - Tyntek涨停，鉴于光子领域光电二极管需求失衡，其订单可见度已至2028年。 Elazr总经理曾表示整个光学供应链面临严重短缺……我接下来可能会更关注高端光电二极管瓶颈。 - 摩根士丹利预测DDR4在Q3环比涨价50%，2026年Q4涨价10%+。（南亚/2408，旺宏/2344应该很高兴）。 - 据报道，中国推迟对台湾出口锗、石英基材料和钕磁铁的许可。 所以我可能会一边吃杯面一边研究那里的供应链中断问题。 - 高端PCB钻头短缺（仍在持续），凯美(5498)称AI数据中心、服务器和光通信正在推动PCB、HDI和封装基板的需求，同时钨价上涨。 尽管全球五大钻头制造商正在积极扩产，供应仍远低于整体市场需求。 只是最近发现的一些有趣内容，当然$MRNA的事也挺有意思。
英文原文
I'm glad the leverage bros moved from AI stocks back to Hyerliquid and biotech like $MRNA, so maybe we'll see a faster recovery. Aside from that just some TLDR stuff: - $EWY up, primarily from SK Hynix buybacks (they're just too profitable), and Samsung Foundry roughly ~10-15% price hikes. - $SKHY unveils CPO roadmap?? Apparently they're doing something with photonic interposer that links to memory, so I gotta look a bit deeper into this Probably going to be more TAM expansion, more supply chain beneficiaries (eg. lasers, PICs, packaging) off this. - $YMTC pursuing IPO next quarter following CMXT's successful IPO. The Chinese IPO market is just going brrr from Unitree to memory makers. US needs to do something similar rather than IPOing companies at trillion dollar valuations already. - Tyntek limit up there's now order visibility through 2028, given photodiode demand imbalance for photonics. Elazr GM did say the entire optical supply chain was facing major shortages... I'll probably look more at high-end PD bottlenecks next. - MS forecasts a 50% Q/Q price increase for DDR4 for Q3, and 10%+ increase for Q4 2026. (Nanya/2408, Windbond/2344, should be happy). - China reportedly delays clearance from germanium, quartz-based materials, and neodymium magnets To taiwan. So I'll probably have some fun looking into supply chains disruptions there over cup noodles. - High-end PCB drill shortage (still ongoing), Kawei (5498) said AI DCs, servers and optical communications are driving demand for PCBs, HDI and package substrates, while tungsten prices are rising. Though the world's five largest drill manufacturers are aggressively expanding capacity, supply remains far below overall market demand Just stuff I found interesting recently, obviously $MRNA stuff is too.
-
AI光通信供应链严重短缺，光子学或重演存储超级周期。
我们从 $LITE、$MTSI、$AAOI 的财报中均看到激光器/TIA/DSP等存在极端的供需失衡... 现在，Elazr 总经理在其投资者大会上表示： 光模块供应仍远远落后于市场需求，预计 AI 光通信供应链（AI optical supply chain）的紧张局面将持续数年。 - “整个供应链都缺货” - “供应仍远远落后于市场需求” - “短缺将持续到未来几年” “无论是 PCB、载板、激光芯片，还是相关封装产能，只要是 AI 供应链环节，几乎全部短缺。” 当需求可见度延伸至数年且供应链存在瓶颈时： 很难不认为光子学（Photonics）正在重演进入 2027 年的存储超级周期。
英文原文
We got $LITE, $MTSI, $AAOI all pointing to extreme demand imbalances with lasers/TIA/DSP/etc from earnings... And now Elazr GM at their investor conference stated: Optical supply is still far behind the market demand, the expected AI optical supply chain are likely to continue for years. - "the entire supply chain is out of stock" - "the supply is still far behind the market demand" - "the shortage will continue to the next few years" "Whether it is PCB, carrier board, laser chip, or even related packaging capacity, as long as the AI supply chain link, almost all are in short supply." When demand visibility extends for years with the supply chain bottlenecked: It's getting difficult not to see photonics echoing the memory supercycle entering 2027.
-
宇树科技上市估值533亿，显示人形机器人纯玩需求超预期。
宇树科技(688836)现已上市。在上涨+492.18%后，其估值约为533亿美元。我们终于有了一个主要的人形机器人上市公司基准。作为参考，由$NVDA和$AMZN支持的Agility通过$CCXI预计将于第四季度以25亿美元投前估值上市。$TSLA市值超过1万亿美元，但Optimus被包裹在一个更大的公司内。但我认为宇树科技表明，市场对纯人形机器人玩家的需求比公开市场预期的要大得多……
英文原文
Unitree (688836) is now public. And is valued at ~$53.3B, after rising +492.18% We finally have a major public company benchmark for humanoids. For reference, Agility (backed by $NVDA, $AMZN) via $CCXI is expected to go public at $2.5B premoney valuation Q4. $TSLA is $1T+, but Optimus is wrapped inside a much larger company. But I think Unitree showed that the demand for pure play humanoid players is much larger… than people expected in public markets.
-
对比人形机器人零部件供应商的全球巨头与中国厂商的营收节奏。
@kevs_World02 哈默纳科(Harmonic)/LeaderDrive/舍弗勒(Schaeffler)似乎无处不在。还有一些更小众的零部件供应商，比如 $VPG，可能用于 Optimus/Figure。但我觉得规模差异巨大，因为舍弗勒声称他们正与约45家人形机器人玩家合作。不过，向 Agibot/Unitree 供货的中国供应商可能会率先实现营收，我很快会对该领域做更多研究。
英文原文
@kevs_World02 Harmonic/LeaderDrive/Schaeffler seem to be in pretty much everywhere. There's some more niche component suppliers like $VPG for maybe Optimus/Figure. But I think the scale is vastly different since Schaeffler claimed they were wokring with ~45 humanoid players. Chinese suppliers to Agibot/Unitree probably hit revenue first though, I can do more research into that area soon.
-
对比CCXI与TSLA机器人应用路径，关注中美自动化竞争。
@HowlStWoofer $CCXI 正优先瞄准商业化，因为他们已进入 $AMZN 和 $MELI 的仓库进行人类自动化作业。我认为 Agility 的机器人无需做出后空翻般的动作来完成配送或包装。亚马逊仓库工人也是如此。这是两种不同的应用场景，我认为波士顿动力(Boston Dynamics)在突破机器人能力极限方面是最接近的同行。$TSLA 也瞄准通用目的，但我认为他们目前还难以实现。无论如何，观察美国能否追赶并超越中国将非常有趣。
英文原文
@HowlStWoofer $CCXI is targeting commercialization first, since they're in places like $AMZN and $MELI warehouses for human automation. I don't think Agility's Robots need to do backflips to do deliveries or packaging. Same with Amazon warehouse workers. Two different use cases, think Boston Dynamics is the closest peer in terms of breaking limits on what robotics can do. $TSLA is also targeting general purpose, but I don't quite think they can pull that off yet. Regardless, seeing if the US can catch up and leapfrog China will be very interesting.
-
宇树科技机器人性能突破，IPO在即，中国领跑硬件竞赛。
OpenAI 的发布让我想起了越南（股市）的闪回。随着每一个新模型，性能都在不断提升……但现在主角是人形机器人。宇树科技(Unitree)宣称其人形机器人： - 现在跑得比尤塞恩·博尔特还快。 - 现在跳得比任何人类都高。 随着其明天 IPO，似乎中国在硬件竞赛中处于领先地位。
英文原文
I'm getting Vietnam flashbacks to the launch of OpenAi. Where with every model, it just keeps getting better... but now with hunanoids. Unitree has claimed that their humanoids: - Can now run faster than Usain Bolt. - Can now jump higher than any human. And with their IPO tomorrow, it seems like China is leading the hardware race.
-
AI半导体板块分化，光学股复苏节奏各异，预期风险偏好回升将带动反弹。
@pkcapital_ 从 $AVGO 到 $TSM 来看，AI 半导体主题(AI semi theme)正在下跌。高贝塔(higher beta)个股似乎受到的打击最重。从我观察到的光学(optical)个股趋势来看，$LITE / $COHR 恢复最快，其次是规模更大但仍属高贝塔的个股如 $AAOI / $AXTI（恢复了一半）。而像 $SIVE / $POET 这类个股耗时最长。但我认为，当市场风险偏好(risk-on)回升或财报(earnings)推动重新评级(re-ratings)时，会出现更强劲的反弹。
英文原文
@pkcapital_ The AI semi theme is down if you look at $AVGO to $TSM. Higher beta names seem to have been hit the hardest. From trends I've seen with optical names, $LITE / $COHR recovered fastest, then larger still high beta names like $AAOI / $AXTI (halfway). Then your $SIVE / $POET type names have taken the longest. But I do see a sharper recovery when markets are more risk-on or when earnings forces re-ratings.
-
日本企业定价保守，需SNDK或三星引领行业激进提价。
@RancicYaro4064 日本公司试图匹配行业费率，但它们并没有像环比增长 70% 那样激进地提价…… 恕我直言，行业内需要有一家 $SNDK 来引领方向，也许三星可以在这里更激进一些。
英文原文
@RancicYaro4064 Japanese companies try to match industry rates, but they don't go out of their way to hike aggressively like 70% Q/Q... There just needs to be a $SNDK in the industry to lead the way imo, maybe Samsung can be more aggresive here
-
MLCC交货期延长且扩产推迟，AI服务器瓶颈加剧。
所以只是在追踪 MLCC（多层陶瓷电容器）周期： ComponentNews 将服务器 MLCC 归类为“严重”短缺阶段，ETNews 报道： - 根据 DigiKey 的出货数据，三星（$005930）部分高容量 MLCC 的交货期已达约 40 周。 今年早些时候，更广泛的报道约为 20 周，因此 AI 服务器 MLCC 瓶颈仍在扩大。 - 村田（$6976）6 月约为 24 周。 7 月约为 30 周。现在有些已达约 36 周。 这很有趣： “新产能扩张计划从 2026 年第四季度推迟至 2027 年” 未说明是哪项扩张……也许是村田？但如果产能扩张据报道受阻，短期瓶颈应该会收紧。 交货期是追踪供需失衡的好方法。
英文原文
So just tracking the MLCC cycle: ComponentNews classified server MLCCs as “severe” shortage stage, with ETNews reporting: - Samsung has reached ~40 week lead times for some high capacitance MLCCs per DigiKey’s shipment data. Earlier this year, broader reports were ~20 weeks, so the AI server MLCC bottleneck keeps growing. - Murata was ~24 weeks in June. July was ~30 weeks. And now some at ~36 weeks. This was interesting: “The expansion of new production capacity is being postponed from Q4 2026 to 2027” Doesn’t say which expansion… maybe Murata? But if capacity expansion reportedly gets stalled, the bottleneck should tighten short term. Lead times are a good way to track demand imbalances.
-
MLCC供应紧张类似NAND周期，但幅度可能不及，需观察。
@Berlinergy 呃，我个人主要是增加了多层陶瓷电容(MLCC)的头寸，而不是增加NAND的敞口。由于MLCC的价格/供应紧张情况类似于早期NAND涨价周期……如果你的无源元件玩家想跟随 $SNDK 的路径。只是说它显示出相似性，但我不确定我们会看到如此极端的情况。需要再等一两个季度看看。
英文原文
@Berlinergy Uhh I mainly just increased MLCC positions rather than NAND exposure personally. Since MLCC prices/tightening resembles the early NAND hike cycle… if your passive component players want to tag along $SNDK path. Just saying it shows resemblance, but I’m not sure we’ll get something as extreme. Will need to wait a quarter or two.
-
引用群联CEO观点，认为NAND供应瓶颈将持续数年，反驳看空论调。
关于 $SNDK、$SKHY、$MU、Kioxia 以及你们关注的 NAND 玩家，这里有一些更多关于需求失衡可见性的信息。 群联电子(Phison) CEO：“2027 年的产能限制将比 2026 年更为严峻。” “上游供应商从工厂建设和设备投资到实际生产需要长达 4 年的时间。 随着需求爆炸式增长且供应扩张有限，NAND 短缺可能会持续数年。” 我完全不同意那些声称存储（尤其是 2026 年的 NAND 瓶颈）已经彻底结束的人的观点...
英文原文
Just some more demand imbalance visibility for $SNDK, $SKHY, $MU, Kioxia, and your NAND players. Phison CEO: "2027 capacity constraints will be even more severe than in 2026." "Upstream suppliers need up to 4 years from plant construction and equipment investment to actual production. With demand growing explosively and supply expansion limited, the NAND shortage is likely to last for years." Don't quite think I agree with the people claiming it's all over for memory, esp. NAND bottlenecks in 2026...
-
AI服务器MLCC挤占消费级产能，看好太阳诱电等通用MLCC厂商。
我认为我们正看到AI服务器对多层陶瓷电容器(MLCC)的需求产生了一个有趣的影响……即消费级/通用MLCC的产能正在收紧。 如果你像我一样对$MU / 三星 / SK海力士 -> 传统DRAM的越南闪回有印象： 我的论点是，我们可能会在下半年末或2027年初看到消费级/广泛MLCC出现同样的情况。 7月6日： “AI导向的高端MLCC生产的挤出效应已开始蔓延至汽车和消费市场。” 7月28日： “AI需求推动日本和韩国MLCC供应商月度出货量创纪录；消费级订单溢出持续激增” 8月12日： 随着[MLCC]产能从消费级转向AI导向产品……（Trendforce） 8月12日： 中国电子市场渠道调查显示22µF和47µF MLCC（使用更广泛）缺货。随后发现并指出某些制造商已停止接单。 （供参考：华新表示高量智能手机/PC产品如……22µF，因此是更广泛的MLCC） 显然，三星电机/村田是AI服务器MLCC扩产的主要受益者。 但像太阳诱电（6976，披露：我持有仓位）这样在整体MLCC中拥有较大市场份额但AI服务器份额较少的公司，将是这一现象的主要受益者。 有趣的是，我们才处于2026年下半年，所以下一个主要瓶颈应该会很有趣！
英文原文
I think we're seeing AI server MLCC demand cause an interesting effect... with tightening capacity for consumer/general-purpose MLCCs. If you get Vietnam flashbacks like I do to $MU / Samsung / SK Hynix -> Legacy DRAM: My thesis is we'll see the same thing here maybe late H2, early 2027 with consumer/broad MLCCs. From July 6th: "the crowding-out effect from AI-oriented high-end MLCC production has begun spilling over into both the automotive and consumer markets." July 28th: "AI Demand Pushes Japanese and Korean MLCC Suppliers to Record Monthly Shipments; Consumer-Grade Order Spillovers Continue to Surge" Aug 12: As [MLCC] capacity shifts from consumer grades to AI-oriented products... (Trendforce) Aug 12: China's electronics market channel checks found 22µF and 47µF MLCCs (used more broadly) out of stock. Then found and said certain manufacturers had stopped taking orders. (for reference: Walsin said high-volume smartphone/PC products such as... 22µF, so it's broader MLCCs) Obviously Samsung Electro-Mechanics/Murata are main beneficiaries of AI server MLCC ramp. But companies with large market share of overall MLCCs like Taiyo Yuden (6976, disclosure: I have positions) with less AI server share, would be a major beneficiary of this. Fun part is, we're only in H2 2026, so the next major bottleneck should be fun!
-
InP短缺致价格大涨，验证博主此前对供应链瓶颈及涨价的预测。
“严重的磷化铟(InP)短缺引发史上最大涨幅”——$AXTI。 供应商直言：“即使你有钱，也可能买不到材料”（哈哈）。 - 行业消息源透露，InP衬底价格从去年第四季度开始上涨，这是连续第四次衬底涨价。 - 预计最新的第四季度涨幅将超过10%。 - 同时，由这些衬底制成的外延片(epitaxial wafers)价格已上调两次，市场正走向第四季度的第三次连续涨价。（我也写过关于$IQE的论点，该公司同样受益）。 文章指出VPEC、Landmark、IET受益（以台湾为中心的报道）。 我去年就预测到了这一点……瓶颈是真实的，但我对接下来的涨价幅度有更高的预期。 很高兴它开始应验了。
英文原文
"Severe InP Shortage Triggers Largest Price Hike Ever" - $AXTI. Suppliers have bluntly stated that "even if you have the money, you still may not be able to get the material" (lol). - Industry sources revealed that InP substrate prices began rising in Q4 last year and was the fourth consecutive substrate price hike. - The latest Q4 hike is expected to be >10%, - Meanwhile, prices for epitaxial wafers made from those substrates have already been raised two times, and the market is moving toward a 3rd consecutive increase in Q4. (I also wrote a thesis on $IQE, which benefits here too). Article states VPEC, Landmark, IET benefits (Taiwan centric reporting). I did predict this last year... The bottleneck is real, but I do have higher expectations for price hikes that comes next. Glad it's starting to play out.
-
推测Stripe将借监管游说，通过强制KYC和许可制构建AI路由护城河。
@MinuteMoola 只是猜测，我将观察 Stripe 如何构建 OpenRouter 的护城河(Moat)…… 通过游说国会，要求所有前沿模型(Frontier Models)的 API 访问必须像开立银行账户一样进行严格的 KYC(了解你的客户)/DD(尽职调查)。 然后要求其他所有公司必须获得许可证才能将推理请求(Inference Requests)路由给客户。 理由大概是：“我们关心用户安全。如果大语言模型(LLMs)能生成核弹指南或生物武器，路由推理请求的公司必须设置护栏(Guardrails)！”
英文原文
@MinuteMoola Just a guess, I'm going to watch Stripe create OpenRouter's moat... By lobbying Congress to make API access across all frontier models require heavy KYC/DD like opening a bank account. Then require everyone else to get licenses to route inference requests to customers. With something along the lines of: "We care about user safety. If LLMs can generate nuke guides, bio weps, companies routing inference requests must have guardrails!"
-
AI供应链多环节瓶颈显现，内存、基板、组件及先进封装需求强劲，行业加速扩张。
一些我感兴趣的要点摘要： - $SNDK 投资者日显示，预计至2030年调整后毛利率达80%，运营利润率约75%，调整后自由现金流利润率约50%。 长期协议(LTA)已占2028年产能的2/3。最低合同收入达930亿美元（当前市值约2390亿美元）... 当你的收入/目标预计在未来4年持续至2030年时，很难将其视为周期性股票。 - $CRWV 签署合同，将使用6年前的 $NVDA A100 GPU 直至2029年。 对于 Nebius/Iren 等新型云服务商(Neoclouds)而言，这是利好，因为它反驳了如 Burry 的折旧看空论点。 - 据 UBS 估计，传统 DRAM 毛利率（如美光）到2027年将达到前所未有的95%，超越 HBM 毛利率。 如果预测准确，像南亚(Nanya)/华邦(Winbond)这样的传统/标准 DRAM 玩家应该会“起飞”(go brrrr)。 - 据报道，Anthropic 实现了14倍以上的同比增长，Q2环比收入增长约2.4倍至>115亿美元，预计2028年收入达190-2000亿美元。 你的前沿实验室保持疯狂的增长速度，如果不这样反而令人担忧。 - $NVDA 据报道正洽谈向 SB Energy（软银子公司）投资30亿美元，与 Apollo、BlackRock、Blackstone、Brookfield、Goldman 等共同推动 >5000 亿美元的算力融资。$NVDA Feynman 预计2028年下半年转向 TSMC A16 + SoIC + 定制 HBM + 共封装光学(CPO)。 每天都有更多英伟达新闻。 - $MSFT Maia 300 讨论了 $TSM 2027年 >30万颗的产能，并计划扩展至100万+。最快9月发布。 $MRVL 对此消息可能更高兴。关于当前情况： - 可能是微软当前 ASIC 扩产的受益者(GUC)。 - 对于亚马逊阵营，像 Alchip（我持有股份）可能在2026年下半年随 $AMZN ASIC 项目扩产... 所以关注超大规模云厂商 ASIC 扩产时间表及其受益者可能是个好主意。 - $TSM 先进封装副总裁表示“行业未来几年可能不仅面临内存短缺，还将面临 ABF 基板供应紧张”... 强调内存和 ABF 基板在未来几年都是瓶颈。 甚至上游 ABF 基板设备供应商也很高兴，例如使用真空层压设备的 Eternal Precision 表示订单激增，工厂满负荷运行，价格上涨20%+。 - $AMAT 预计2026年先进封装收入增长 >70%（此前预期 >50%），并表示客户讨论已延伸至2030年。 （不太熟悉这家公司，但发现其从2025 Q4 $68亿 -> $70.1亿 -> 79.1亿 -> $91.2亿 -> 2026 Q4 $102.5亿的预测增长率很有趣） - Google 在 OCP APAC 表示传统 48V 已无余量。$NVDA 详细说明了2026年下半年兼容 MGX 的 800VDC 机架（时间线，Delta / Lite-On 受益）。 - 除 $SNDK 外，南亚 LTA 覆盖50%产能。CXMT 上月签署多年 DRAM 协议，因此整个内存行业似乎都在效仿三大巨头的剧本。 - 探针卡(Probe cards)仍是瓶颈，MPI(6223)表示其探针卡产能因需求超过供应而保持满负荷。 本周早些时候已覆盖 CW 激光瓶颈（$AAOI, $SIVE, $LITE），那是另一个有趣的点。 - 据 Nichidenbo 称，部分 MLCC/组件交货期已达36周。 你的三星电机、太阳诱电、村田等玩家听到这个消息应该很高兴。 TLDR: AI 供应链疯狂加速(go brrr)。
英文原文
Just some TLDRs of stuff I found interesting: - $SNDK 80% adjusted gross margins projections through 2030, ~75% operating margins and ~50% adjusted FCF margins from investor day. LTAs already 2/3rd of 2028 output. Minimum contracted revenue reaches $93B (MC is currently ~$239B)... Hard to be a cyclical stock when your revenue/targets are expected to continue 4Y later into 2030. - $CRWV signs contracts for 6Y old $NVDA A100 GPUs through 2029. For Neoclouds like Nebius/Iren, this is positive, since it's a counterargument for eg. Burry depreciation short thesis - conventional DRAM gross margins eg. Micron is estimated to reach an unprecedented 95% by 2027, surpassing HBM GMs per UBS read through for legacy/standard dram players like Nanya/Winbond should go brrrr if projections are correct. - Anthropic reportedly achieved 14x+ YoY growth and roughly 2.4x sequential revenue growth q2 to >$11.5B,. estimating growth to $190–200B in 2028r evenue numbers. Your frontier labs keep growing at stupidly fast paces, it would be worrisome if they didnt. - $NVDA reportedly in talks to invest $3B in SB Energy (Softbank subsidiary), creates a >$500B compute financing push with Apollo, BlackRock, Blackstone, Brookfield, Goldman, and others. $NVDA Feynman reportedly moves to TSMC A16 + SoIC + custom HBM + CPO in H2 2028 Just more nvidia news every day - $MSFT Maia 300 discussed $TSM capacity for >300k units in 2027, with expansion to 1m+. Unveils as soon as September. Likely $MRVL should be more happy from this news. For what's happening right now: - maybe GUC for Microsoft current ASIC ramp. - For the Amazon party, stuff like Alchip (I do own shares), likely is ramping now with $AMZN ASIC program H2 2026... So might be a good idea to look at hyperscaler ASIC ramp timelines + their beneficiaries. - $TSM VP of Advanced Packaging stated "the industry is likely to face not only memory shortages but also tight ABF substrate supply over the next few years"... Emphasis on few years for memory + ABF substrates for bottlenecks. Even upstream abf substrate equipment providers are happy, eg. Eternal Precision which uses vacuum lamination equipment stated orders surged, their plants have been running at full capacity, and 20%+ price hikes. - $AMAT expects advanced packaging revenue to grow >70% in 2026, versus prior >50%, and said customer discussions now extend all the way to 2030 (not too familiar with this company, but found their growth rate from 2025 Q4 $6.8B ->$7.01B -> 7.91B -> $9.12B -> $10.25B Q4 2026 projections pretty interesting) - Google said at OCP APAC said conventional 48V is running out of headroom. $NVDA detailed an 800VDC MGX-compatible rack H2 2026 (timeline, Delta / Lite-On beneficaries) - Aside from $SNDK, Nanya LTAs cover 50% of capacity. CXMT signed multi-year DRAM agreements last month, so entire memory industry seems to be following same playbook as ur big 3. - Probe cards remain a bottleneck, MPI(6223) said their probe card capacity remains fully utilized because demand exceeds supply. Already covered the CW laser bottleneck with $AAOI, $SIVE, and $LITE earlier this week, but that's another fun one. - some MLCC/component lead times have hit 36 weeks per Nichidenbo. Your Samsung Electro-Mechanics, Taiyo Yuden, Murata, players should be very happy to hear this. TLDR: AI supply chains go brrr.
-
美政府劝苹果弃购中国存储，CXMT/YMTC涨价打破廉价印象。
美国政府敦促 $AAPL 不要购买中国存储芯片。 $MU、$SKHY 和三星听到这个消息很高兴…… 颇具讽刺意味的是，长鑫存储(CXMT)/长江存储(YMTC)也在涨价，所以这并非“充斥市场的廉价中国存储芯片”。 在人工智能繁荣之前，苹果曾经拥有极强的供应商议价能力。
英文原文
US Government urges $AAPL not to buy Chinese memory chips. $MU, $SKHY, and Samsung are happy to hear this... Hilariously enough, CXMT/YMTC is price hiking too, so it's not exactly "cheap Chinese memory that floods the market". Apple once upon a time actually had extreme supplier leverage before the AI boom.
-
质疑MU和SNDK的长期协议及高毛利预测，认为其周期极长。
@sellihca_espn 我的记忆包（持仓）相当沉重，谢谢关心。纯粹出于好奇，你认为与 $MU 签署的 16 份以上长期协议（Long Term Agreements, LTAs）...以及 $SNDK 到 2030 财年 80% 的毛利率预测是周期性大宗商品吗？（他们的长期协议已覆盖 2028 年大部分产量）。在我看来，这周期相当长...
英文原文
@sellihca_espn My memory bags are decently heavy, thank you for asking. Just out of curiosity, you think 16+ long term agreements with $MU. And $SNDK 80% gross margin projections through fiscal 2030 is a cyclical commodity? (their LTAs already cover majority of 2028 output). This is quite a long cycle if you ask me...