觅信聚合

我的关注

2026-07-06 14:50

原文：
I’d like to give a special thank you to Bloomberg’s piece on $META compute.

As well as the 2 back to back bear posts on CPO delays.

But I believe thematically photonics, memory, energy, and physical AI will outperform in the long run. https://t.co/vghsG5H2w5

翻译：
我想特别感谢彭博社（Bloomberg）关于 $META 计算的报道。

以及关于CPO延迟的两篇连续的空头帖子。

但我相信从长远来看，光子学、内存、能源和物理人工智能等主题将会表现出色。 https://t.co/vghsG5H2w5

2026-07-06 12:36

原文：
Wow, $WULF signs a $19B DC lease with Anthropic today.

Probably a very positive tailwind for the Neocloud/Colo sector. https://t.co/82tncsG3us

翻译：
哇，$WULF 今天与 Anthropic 签署了一份 190 亿美元的数据中心租赁协议。
这可能是对 Neocloud/Colo 行业的一个非常积极的推动力。 https://t.co/82tncsG3us

2026-07-06 12:04

原文：
Retail: buys memory, optical, upstream semi names.

Media: “retail bubble drives AI meme stock bubble higher”.

Institutions that do the exact same thing, with the exact same names:

Media: “Institutions make record profits by betting on what goes inside AI Server Racks”. https://t.co/lWDCl0Sxbe

翻译：
零售投资者：购买内存、光学和上游半导体股票。

媒体报道：“零售泡沫推动AI概念股上涨”。

机构投资者做着同样的事情，投资同样的股票：

媒体报道：“机构通过押注AI服务器机架内的组件获得创纪录的利润”。 https://t.co/lWDCl0Sxbe

2026-07-06 10:19

原文：
I find it hard for any bears to defend against this humanoid bull case. https://t.co/aWWPYAvSIa

翻译：
我觉得任何空头都很难为自己辩护，因为这个类人牛市的理由很有说服力。 https://t.co/aWWPYAvSIa

2026-07-06 09:30

原文：
Kelper upgrades $XFAB to "Buy" citing:

- Melexis demand strengthening “week-after-week, day-after-day” (their lead customer).
- AI giving XFAB a structural growth leg with SiC, GaN, and photonics offsetting auto cyclicality.
- Scarce Western speciality foundries like $SILEX commanding higher multiples.

"X-Fab remains the cheapest specialty foundry peer on every multiple, trading at 4.8x 2027 EV/EBITDA versus a peer median of 14.5x"

I treat this as good signal for the core fundamentals improving like auto players.

But in terms of unpriced long term growth vectors, I'm personally looking at XFAB from a different angle in terms of photonics growth with photonixfab. Given $NVDA as a potential key driver.

But it's hard for traditional firms to price these types of active developments into valuation models until there's volume contracts.

翻译：
Kelper 将 $XFAB 升级为 "买入"，理由是：

- Melexis 的需求不断增强（他们的主要客户），每周每天都在增长。
- 人工智能为 $XFAB 提供了结构性增长动力，碳化硅、氮化镓和光子学技术可以抵消汽车行业的周期性波动。
- 西方稀缺的特色代工厂，如 $SILEX，具有更高的估值倍数。

"X-Fab 仍然是最便宜的特色代工厂同行，每个倍数都很低，2027 年的 EV/EBITDA 为 4.8 倍，远低于同行中位数的 14.5 倍"

我认为这是核心基本面改善的良好信号，例如汽车行业。

但是，从长期增长的角度来看，我个人正在从不同的角度看待 $XFAB，特别是光子学增长和 photonixfab。考虑到 $NVDA 可能成为关键驱动力。

但是，传统公司很难在没有大量合同之前将这些活跃的发展纳入估值模型中。

2026-07-06 08:26

also mentioned NASDAQ listing completion targeted in the next few quarters (probably H2 2026 or Q1 2027 is my est. timeframe).

This would fund M&A efforts, since it's impossible with the fundraising environments in local Swedish markets.

As for becoming the next $LITE:

M&A makes their lasers more valuable, so downstream IP acqusition -> into contract manufacturing like $FN, and others to make the full 1.6T pluggable or optical engines. Is how they get there, since laser array ASP scaling that people are modeling off of, wouldn't command a $60B+ valuations.

There's going to be a lot of bridge architectures like NPO/pluggables, etc and noise around certain architectural delays in the meantime.

But markets misunderstand laser companies like $LITE, $SIVE, $AAOI and others are used across different architectures compared to if you just look at certain passive optical components.

So markets see "CPO delay headlines" algos sell off laser companies that benefit from other architectures.

Being included in the pluggable 1.6T ramp to CPO scale out (Which Sivers is included in), helps bridge revenue waiting gaps until scale up inflection point H2 2027.

I'm personally holding long term, since I haven't seen a ~$1.4B company mapping to this many hyperscalers before.

TLDR:
- Waiting on volume ramps from different architectures to play out across their hyperscaler supplier mapping
-> 1.6T LRO/CPO scale out late H2 2026 start into high volume ramp 2027
-> H2 2027 CPO scale up volume ramp
- Waiting on NASDAQ listing likely H2 2026/Q1 2027 for M&A efforts to fully take off, unless Sivers get more creative with equity financing in the meantime.

翻译：
我个人认为 $SIVE 可以成为下一个 $LITE。

在过去的几个月里，我们已经看到：

1. 与 O-Net 合作，将 ELS 推向大规模生产
2. $JBL 1.6T LRO 大规模生产信号，使用 Sivers 的可插拔产品具有“相对戏剧性的护城河”
3. $GFS SCALE 参考级激光器用于超大规模计算，具有可插拔、NPO、CPO 功能
-> -> 其中 $AMD 和其他公司已转向 GFS 采购 CPO
4. Ayar 加入

2026-07-06 08:26

$NVDA NVLink 的 CPO 行列
-> -> 从他们的网站上删除了 Lumentum/Macom，可能使 Sivers 成为他们的主要激光器供应商
-> -> AlChip 可能赢得了 Trainium 的订单，来自亚马逊的价格定位（Ayar 的客户）
-> -> GUC 的机架级设计与 Ayar 合作
-> -> 通过 AMD、Alchip、MediaTek 和 NVIDIA 的投资，获得了 5 亿美元的资金用于大规模生产
5. ~ $AEVA 计划在 2026 年下半年开始大规模生产
6. $POET 计划在 2026 年下半年开始大规模生产，拥有超大规模计算供应商，如 Lumilens（“前三名超大规模计算初始客户”）
7. TFLN + $SIVE 的 CW 激光器与 Lightium 合作
8. 可能与 $MRVL Celestial 和 CPO 玩家如 Lightelligence/Lightmatter 建立直接关系
9. 多个新的未披露关系，为可插拔产品跟随 Jabil 的季度记录

随着新的 Trendforce 报告显示 $AMD 和其他超大规模计算公司正在尝试采购长期协议（LTAs）用于连续波激光源，这将成为独立连续波源的直接催化剂。

当超大规模计算供应商从 Jabil 到 O-Net 被激励去大规模生产尽可能多的产品时：这对于 Sivers 的收入相对于当前估值来说是非常重要的，这看起来就像是一个等待游戏。

即使在过去的一周里：

- $SIVE 为量产而进行了超额认购的机构融资... 这是很微妙的，因为 Sivers 是 fab-lite，所以他们不会在内部进行代工投资来扩大规模。可能是为了与 Win Semi 和其他合作伙伴（他们也提到了其他合作伙伴）合作，以扩大激光器生产和代工分配。

所以，这可能是收入增长的重要信号。

- Sivers 也提到了纳斯达克上市的完成，目标是在接下来的几个季度内完成（可能是 2026 年下半年或 2027 年第一季度）。

这将为并购努力提供资金，因为在当地瑞典市场的融资环境中，这是不可行的。

至于成为下一个 $LITE：

并购使他们的激光器更加有价值，因此下游知识产权收购 -> 合同制造，如 $FN 和其他公司，以制造完整的 1.6T 可插拔产品或光学引擎。这是他们实现这一目标的方式，因为激光器阵列的平均销售价格（ASP）扩大，人们正在建模，这不会带来 600 亿美元以上的估值。

在此期间，将会有很多桥接架构，如 NPO/可插拔产品等，以及关于某些架构延迟的噪音。

但市场误解了激光公司，如 $LITE、$SIVE、$AAOI 等，在不同架构中被使用，相比之下，如果你只看某些被动光学组件。

所以，当市场看到“CPO 延迟头条”时，算法会卖出受益于其他架构的激光公司。

被包含在可插拔 1.6T 的 CPO 扩展中（Sivers 被包含在内），有助于弥补收入等待缺口，直到 2027 年下半年规模扩张的拐点。

我个人持有长期观点，因为我以前从未见过一个价值 ~14 亿美元的公司与这么多超大规模计算公司映射。

简而言之：
- 等待不同架构的量产拉动效应
-> 1.6T LRO/CPO 扩展于 2026 年下半年开始，进入高容量拉动 2027 年
-> 2027 年下半年 CPO 扩张量产拉动
- 等待纳斯达克上市，可能在 2026 年下半年或 2027 年第一季度，用于并购努力，除非 Sivers 在此期间找到更具创意的股权融资方式。

2026-07-06 08:26

原文：
I personally think $SIVE can be the next $LITE.

In the past few months alone, we've seen:

1. Partnerships with O-Net pushing ELS into mass production
2. $JBL 1.6T LRO mass production signals with "relatively dramatic moats" for pluggables using Sivers.
3. $GFS SCALE reference level laser for hyperscalers with pluggable, NPO, CPO.
-> -> where $AMD and others went to GFS for CPO.
4. Ayar, which joined $NVDA NVLink for CPO
-> -> which removed Lumentum/Macom from their website and likely made Sivers their primary laser supplier.
-> -> AlChip likely Trainium win from Amazon price placement (Ayar's customer)
-> -> GUC rack level design in with Ayar.
-> -> Raised $500m for mass production by AMD, Alchip, Mediatek, and NVIDIA
5. ~ $AEVA starting HVM H2 2026.
6. $POET starting HVM H2 2026 with hyperscaler suppliers like Lumilens ("top 3 hyperscaler initial customer")
7. TFLN + $SIVE CW Lasers with Lightium
8. Likely direct relationships with $MRVL Celestial and CPO players like Lightelligence/Lightmatter.
9. Multiple new undisclosed relationships for pluggables following Jabil in their quarterly transcripts

With new Trendforce reports that $AMD and other hyperscalers are trying to source LTAs for CW laser sources, serving as a direct catalyst for independent CW sources.

So when hyperscaler suppliers from Jabil to O-Net are incentivized to mass produce as many as they can: That's very material for revenue for Sivers relative to current valuations, and it looks like just a waiting game.

Even in the past week:

- $SIVE raised an oversubscribed institutional round for volume ramp... This is very nuanced since Sivers is fab-lite so it's not going to in-house foundry capex to scale. Likely toward Win Semi and others (they mentioned other partners too), for laser scaling + foundry allocations.

So this is likely signaling material for revenue ramp is coming.

- Sivers

2026-07-06 08:24

NVLink 进行 CPO
-> -> 从他们的网站上删除了 Lumentum/Macom，可能使 Sivers 成为他们的主要激光器供应商
-> -> AlChip 可能从 Amazon 的价格定位中获得 Trainium 的胜利（Ayar 的客户）
-> -> GUC 在 Ayar 中进行机架级设计
-> -> 通过 AMD、Alchip、MediaTek 和 NVIDIA 的投资，获得了 5 亿美元的资金用于大规模生产
5. ~ $AEVA 预计在 2026 年下半年开始大规模生产
6. $POET 预计在 2026 年下半年开始大规模生产，拥有像 Lumilens 这样的超大规模计算机供应商（“前三名超大规模计算机初始客户”）
7. TFLN + $SIVE 与 Lightium 合作开发 CW 激光器
8. 可能与 $MRVL Celestial 和 CPO 玩家如 Lightelligence/Lightmatter 建立直接关系
9. 在 Jabil 的季度财报中，多个新未披露的关系用于可插拔产品

根据 Trendforce 的最新报告，$AMD 和其他超大规模计算机公司正在尝试寻找长期协议（LTAs）用于连续波激光源，这将成为独立连续波源的直接催化剂。

当超大规模计算机供应商从 Jabil 到 O-Net 被激励去生产尽可能多的产品时：这对于 Sivers 的收入来说是非常重要的，相对于当前的估值来说，这看起来只是一个等待游戏。

即使在过去的一周里：

- $SIVE 为量产而进行了超额认购的机构融资... 这是很微妙的，因为 Sivers 是 fab-light，所以它不会在内部进行代工投资来扩大规模。可能是为了与 Win Semi 和其他公司（他们也提到了其他合作伙伴）合作，进行激光器扩大和代工分配。

所以，这可能是收入增长的重要信号。

- Sivers 也提到了在接下来的几个季度内完成纳斯达克上市的计划（我估计时间框架可能是 2026 年下半年或 2027 年第一季度）。

这将为并购努力提供资金，因为在当地瑞典市场的融资环境中，这是不可行的。

至于成为下一个 $LITE：

并购使他们的激光器更加有价值，因此下游知识产权收购 -> 合同制造，如 $FN 和其他公司，生产完整的 1.6T 可插拔产品或光学引擎。这是他们实现这一目标的方式，因为激光器阵列的平均销售价格（ASP）扩大，人们正在根据此进行建模，这不会带来 600 亿美元以上的估值。

在此期间，将会有很多桥接架构，如 NPO/可插拔产品等，以及关于某些架构延迟的噪音。

但市场误解了像 $LITE、$SIVE、$AAOI 等激光器公司被用于不同的架构，而不是仅仅查看某些被动光学组件。

所以，当市场看到“CPO延迟头条”时，算法会抛售受益于其他架构的激光器公司。

被纳入可插拔 1.6T 增长到 CPO 规模（Sivers 被纳入其中），有助于弥补收入增长的差距，直到 2027 年下半年规模扩大。

我个人持有长期观点，因为我以前从未见过一个价值 ~14 亿美元的公司与这么多超大规模计算机公司映射。

总结：
- 等待不同架构的量产拉动效应
-> 1.6T LRO/CPO 规模扩大，2026 年下半年开始，2027 年进入高量产
-> 2027 年下半年 CPO 规模扩大量产
- 等待纳斯达克上市，可能在 2026 年下半年或 2027 年第一季度，除非 Sivers 在此期间找到更具创意的股权融资方式。

2026-07-06 08:24

also mentioned NASDAQ listing completion targeted in the next few quarters (probably H2 2026 or Q1 2027 is my est. timeframe).

This would fund M&A efforts, since it's impossible with the fundraising environments in local Swedish markets.

As for becoming the next $LITE:

M&A makes their lasers more valuable, so downstream IP acqusition -> into contract manufacturing like $FN, and others to make the full 1.6T pluggable or optical engines. Is how they get there, since laser array ASP scaling that people are modeling off of, wouldn't command a $60B+ valuations.

There's going to be a lot of bridge architectures like NPO/pluggables, etc and noise around certain architectural delays in the meantime.

But markets misunderstand laser companies like $LITE, $SIVE, $AAOI and others are used across different architectures compared to if you just look at certain passive optical components.

So markets see "CPO delay headlines" algos sell off laser companies that benefit from other architectures.

Being included in the pluggable 1.6T ramp to CPO scale out (Which Sivers is included in), helps bridge revenue waiting gaps until scale up inflection point H2 2027.

I'm personally holding long term, since I haven't seen a ~$1.4B company mapping to this many hyperscalers before.

TLDR:
- Waiting on volume ramps from different architectures to play out across their hyperscaler supplier mapping
-> 1.6T LRO/CPO scale out late H2 2026 start into high volume ramp 2027
-> H2 2027 CPO scale up volume ramp
- Waiting on NASDAQ listing likely H2 2026/Q1 2027 for M&A efforts to fully take off, unless Sivers get more creative with equity financing in the meantime.

翻译：
我个人认为 $SIVE 可以成为下一个 $LITE。

在过去的几个月里，我们已经看到：

1. 与 O-Net 合作，将 ELS 推向大规模生产
2. $JBL 1.6T LRO 大规模生产信号，使用 Sivers 的可插拔产品具有“相对戏剧性的护城河”
3. $GFS SCALE 参考级激光器，用于超大规模计算机的可插拔、NPO、CPO
-> -> 其中 $AMD 和其他公司已转向 GFS 进行 CPO
4. Ayar 加入 $NVDA

2026-07-06 08:24

原文：
I personally think $SIVE can be the next $LITE.

In the past few months alone, we've seen:

1. Partnerships with O-Net pushing ELS into mass production
2. $JBL 1.6T LRO mass production signals with "relatively dramatic moats" for pluggables using Sivers.
3. $GFS SCALE reference level laser for hyperscalers with pluggable, NPO, CPO.
-> -> where $AMD and others went to GFS for CPO.
4. Ayar, which joined $NVDA NVLink for CPO
-> -> which removed Lumentum/Macom from their website and likely made Sivers their primary laser supplier.
-> -> AlChip likely Trainium win from Amazon price placement (Ayar's customer)
-> -> GUC rack level design in with Ayar.
-> -> Raised $500m for mass production by AMD, Alchip, Mediatek, and NVIDIA
5. ~ $AEVA starting HVM H2 2026.
6. $POET starting HVM H2 2026 with hyperscaler suppliers like Lumilens ("top 3 hyperscaler initial customer")
7. TFLN + $SIVE CW Lasers with Lightium
8. Likely direct relationships with $MRVL Celestial and CPO players like Lightelligence/Lightmatter.
9. Multiple new undisclosed relationships for pluggables following Jabil in their quarterly transcripts

With new Trendforce reports that $AMD and other hyperscalers are trying to source LTAs for CW laser sources, serving as a direct catalyst for independent CW sources.

So when hyperscaler suppliers from Jabil to O-Net are incentivized to mass produce as many as they can: That's very material for revenue for Sivers relative to current valuations, and it looks like just a waiting game.

Even in the past week:

- $SIVE raised an oversubscribed institutional round for volume ramp... This is very nuanced since Sivers is fab-light so it's not going to in-house foundry capex to scale. Likely toward Win Semi and others (they mentioned other partners too), for laser scaling + foundry allocations.

So this is likely signaling material for revenue ramp is coming.

- Sivers

2026-07-06 05:49

原文：
Nomura’s report cites $AXTI and $IQE as the leading players.

Then regarding ASP updates:

- 2in InP substrate price hike 42-76%
- 3in InP substrates price hike 78%
- 2in EML Epiwafer price hike 50-75%
- 3in CW Epiwafer price hike ~40%+
- Prices lower domestically in China as expected.

Hyperscalers bottlenecking upstream photonics confirmation validation… on all fronts from pricing, demand, to supply shortages.

My personal expectation was that price hikes would keep going up like $SNDK given ramping optical demand.

We’ll see if this turns out right given another 2 years of the demand curve.

翻译：
野村证券的报告将 $AXTI 和 $IQE 列为领先者。

关于平均销售价格（ASP）更新：

- 2英寸磷化铟（InP）基板价格上涨 42-76%
- 3英寸磷化铟基板价格上涨 78%
- 2英寸 EML 外延片价格上涨 50-75%
- 3英寸连续波（CW）外延片价格上涨约 40%+
- 中国国内价格如预期般下降。

超大规模企业对上游光子学的瓶颈确认验证……从价格、需求到供应短缺的所有方面。

我个人预期价格上涨会像 $SNDK 一样持续上涨，因为光学需求正在增加。

我们将看到这是否在另外 2 年的需求曲线中得到验证。

2026-07-05 20:33

原文：
Came across an interesting report from SVRC Research called "State of Robotics 2026", published in April.

Which listed:

1. Figure AI
2. Agility Robotics $CCXI
3. Apptronik
4. $TSLA
5. Boston Dynamics
6. Physical Intelligence
7. 1X Technologies
8. $AMZN Robotics
9. Covariant
10. Skild AI

As the National Champions of the United States robotics program.

"The United States leads the world in where robotics is heading: Fundation models, OpenAI-style scaling laws applied to action, autonomous vehicles.

While losing the race on where robotics is shipping today."

Then it frames:

1. Rare Earths Exposure: from Neodymium for motors to samarium-cobalt for high-temp applications as a critical vulnerability.

2. Actuator dependency. Series elastic actuators, quasi-direct-drive motors, and precision reducers overwhelmingly sourced from Japan, Germany, and China

As one of the main vulnerabilities alongside Manufacturing velocity/data collection cost/regulations. Then their take was:

"With at least six well-funded US humanoid companies competing for a market still in early formation, we expect at least two significant consolidation events (acquisition or merger) in 2027".

With Logistics / E-commerce (like $AMZN / $FDX) and Automotive from $GM to $FORD as being the immediate top use cases for deployment.

I think it's just interesting to see a lot of my points I've been talking about reiterated by research firms. Regardless, I do think it's going to be a major frontier race between the US and China.

Agility Robotics (which I own), Tesla, Figure, and Apptronik as leaders representing the USA. Competing against Unitree, AGIbot, Ubtech, and others in China.

翻译：
我偶然发现了一份来自SVRC Research的有趣报告，名为“2026机器人状态”，于四月份发布。

报告列出了以下公司：

1. Figure AI
2. Agility Robotics $CCXI
3. Apptronik
4. $TSLA
5. Boston Dynamics
6. Physical Intelligence
7. 1X Technologies
8. $AMZN

2026-07-05 20:33

Robotics
9. Covariant
10. Skild AI

作为美国机器人项目的国家冠军。

“美国在机器人领域的发展方向上领先于世界：基础模型、OpenAI风格的扩展规律应用于行动、自动驾驶车辆。

但在机器人当前的应用方面却落后于其他国家。”

然后报告指出：

1. 稀土元素的依赖：从用于电机的钕到用于高温应用的钐钴，稀土元素的供应成为一个关键的弱点。

2. 执行器的依赖性。系列弹性执行器、准直接驱动电机和精密减速器主要来自日本、德国和中国

这是主要弱点之一，除了制造速度、数据收集成本和法规等因素。报告的结论是：

“由于至少有六家资金充足的美国人形机器人公司正在竞争一个仍处于初期形成阶段的市场，我们预计2027年将会有至少两次重大整合事件（收购或合并）”。

物流/电子商务（如$AMZN / $FDX）和汽车行业从$GM到$FORD将是部署的首要用例。

我认为看到研究公司重申我一直在讨论的许多观点很有趣。无论如何，我确实认为这将是一场美国和中国之间的重大竞争。

Agility Robotics（我持有）、特斯拉、Figure和Apptronik是代表美国的领先者。他们将与中国的Unitree、AGIbot、Ubtech等公司竞争。

2026-07-05 15:37

原文：
Just some consolidated updates on memory:

- $MU leads new 1.5T Yen investment in Hiroshima ~$9.3B. (bullish read through for Disco, Advantest, Resonac, Towa) since capex is localized.

- Morgan Stanley pointed out NAND will continue to be in short supply into 2027 so $SNDK / Kioxia type players are happy alongisde $SIMO and upstream.

- MS remains especially positive on Macronix/Winbond

- UBS expects the average price of DDR contracts in the Q3 2026 to increase 32% | 18% Q4, vs. 17% and 12% est.

- UBS expects NAND flash to be raised 30% from prev quarter.

- Samsung reportedly plans 20% DRAM hike Q3. TrendForce recently forecast DRAM contract prices to rise 13 to 18 percent in the third quarter from the previous quarter, so this hike beats expectations.

Something to note:

Lot people see 20%... and don't think it's a lot compared to the 70-80% from previous quarters.

But if you hike something by 100%, then hike something by 100%, then hike something by 30%, it's a lot more than people estimate since it's compounded. Similar to tracking inflation.

I've already made projections going into 2028 from the start of the year on my memory names...

I'm just sitting back and watching things play out through all the "memory optimization" and "they can't keep price hiking like this!" noise.

翻译：
以下是一些关于内存的最新消息：

- $MU 领投了1.5万亿日元（约93亿美元）的投资于广岛，这对Disco、Advantest、Resonac和Towa来说是一个积极的信号，因为资本支出是本地化的。

- 摩根士丹利指出，NAND将继续在2027年面临短缺，因此$SNDK / Kioxia等公司以及$SIMO和上游公司将受益。

- 摩根士丹利对Macronix/Winbond尤其持积极态度。

- 瑞银预计2026年第三季度DDR合同的平均价格将增加32%，第四季度将增加18%，高于之前的预估17%和12%。

- 瑞银预计NAND闪存的价格将比上个季度增加30%。

- 据报道，三星计划在第三季度将DRAM价格提高20%。趋势力研究所最近预测，第三季度DRAM合同价格将比上个季度增加13%至18%，因此这次涨价超出了预期。

需要注意的是：

很多人看到20%的涨价，觉得不算很多，尤其是与之前的70-80%的涨价相比。

但是，如果你先将价格提高100%，然后再提高100%，然后再提高30%，那么实际的涨价远远超过人们的预期，因为这是复利的效果，类似于追踪通货膨胀。

我已经从今年开始，对我的内存股票做出了2028年的预测。

我只是坐回去，观察事情的发展，听着所有关于“内存优化”和“他们不能一直这样涨价！”的喧嚣。

2026-07-05 13:39

原文：
Looks like there's a high power cylindrical cell / BBU cell shortage (aka: bottleneck):

- Samsung SDI supplies the cells to Simplo, which assembles them into BBUs for $META to $AMZN.
- As demand from data centers has increased, production of Samsung SDI's cylindrical cells has also risen sharply.
- Industry sources said Panasonic, Samsung are experiencing supply shortages for BBU cells.

So main beneficiaries imo look like:

- Samsung SDI (KRX: 006400) / Panasonic Energy as cleaneset winners
- Murata (6981) / this keeps showing up everywhere with MLCC and others lol
- LG Energy (KRX: 373220) / incoming

There's not really much direct US players? But adjacent read through like $VRT, $ETN or $BWA / $ENS. Not quite in my domain emphasis cup of tea though.

I wouldn't conflate this as all shortages having a massive TAM, but there could be opportunities...

翻译：
看起来高功率圆柱形电池/BBU电池（也就是瓶颈）短缺：

- 三星SDI向Simplo供应电池，后者将其组装成BBU用于$META和$AMZN。
- 随着数据中心需求的增加，三星SDI的圆柱形电池生产也大幅增加。
- 业内人士表示，松下和三星正在经历BBU电池供应短缺。

因此，我认为主要受益者是：

- 三星SDI（KRX：006400）/松下能源，看起来是最大的赢家
- Murata（6981）/这家公司在MLCC和其他领域不断出现，很有趣
- LG能源（KRX：373220）/即将到来

美国玩家并不多，但可以关注$VRT、$ETN或$BWA / $ENS等相关公司。虽然这不是我的专长，但还是值得注意的。

我不会把所有短缺都看作是一个巨大的市场机会，但还是有一些潜在机会... 🤔

2026-07-05 11:58

auto: eg. “70% of process and technology overlaps with cars”.

And then have AI auto as another major growth vector per. $TSM chairmain (also for power semis too).

So feels like everything comes together for them auto players dragged down by downturns, so I would guess maybe 1 1/2 years, in 2028, things start to pick up for Western auto involved in humanoids.

Since non-China supply chains (like Agibot/unitree) are first to market and already volume ramping compared to western counterparts.

There's other "purer exposure" component suppliers like $VPG and possibly $TSLA / Figure was my guess, but $150/robot that they revealed for mass production is a bit small considering you reach 50K mass production and that's around $7.5M revenue.

So upstream components for auto players for robotics doesn't move any needle unless you get to massive humanoid scale...

which is why I've personally shifted to highest concentration for downstream players like Agility while valuations are relatively small.

I still think we'll see an Anthropic/OpenAI type leader in the humanoid/robotics space soon before all the other physical ai suppliers in the sector get brought up beyond hype.

But just informal personal thoughts I try and keep in mind moving forward with legacy auto players in a downturn + new growth vectors 2Y out

翻译：
以下是关于波士顿动力学、人形机器人和汽车行业的高层次思考：

据传他们的供应商包括：
三星电机 / 英特尔
LG Innotek 为 Atlas 提供视觉感知组件
三星电子 / SK 海力士 / 美光 - 内存
英伟达 - Jetson Thor 计算
谷歌 DeepMind
LG 能源解决方案 - 据报道是波士顿动力学 Atlas 电池供应商
CATL - 电池
SPG 公司 / 에스피지 (058610.KQ) - 减速器 / 谐波驱动风格机器人齿轮
SL 公司 / 에스엘 (005850.KRX) - 与波士顿动力学签订腿部模块供应合同
SOS 实验室 / 에스오에스랩 (464080.KQ) - 预计将成为 Atlas LiDAR 供应商
Sungwoo Hitech - 据报道可能是 Atlas 外壳供应商
现代摩比斯 - Atlas 执行器；机器人手 / 夹持器
ROBOTIS / 로보티즈 (108490.KQ) - 机器人手 / 夹持器 / 机器人执行器 / 伺服模块
HL Mando / HL만도 (204320.KRX) - 执行器 / 汽车运动控制部件
ABB (ABBN.SWX) - 执行器供应商 / 生态系统
Samhyun / 삼현 (437730.KQ) - 智能执行器 /

2026-07-05 11:58

电机控制汽车零部件
Seoyon E-Hwa / 서연이화 (200880.KRX) - 汽车内外饰零部件；潜在机器人外壳供应商

这些信息来自最近在 Naver 上流传的文章。

除了谷歌、英伟达和美光等西方公司外，还有通过 LG Innotek 的 $AEVA。

$OUST 是 Velodyne 与波士顿动力学 Spot 的 LiDAR 协议遗产。Onsemi 是 SOS 实验室 LiDAR SPAD 数组合作伙伴。

供应链很大，所以可能还有其他公司。

我实际上并不看好波士顿动力学的商业化，即使我对他们的技术很看好。

所以，个人而言，如果他们不能像 Unitree 那样大规模生产 30,000 台机器，那么在上游供应链上进行长期投资就没有意义。

一般来说，汽车公司对我来说很有趣，因为它们有广泛的人形机器人曝光。

Schaeffler 的名字被提及，他们曾与大约 45 家人形机器人制造商合作，可能占大多数，但还没有确认的证据。机器人和汽车之间有太多的重叠：例如，“70% 的工艺和技术与汽车重叠”。

然后还有 AI 汽车作为另一个主要的增长矢量，根据 $TSM 主席的说法（对于电力半导体也是如此）。

所以，感觉一切都汇聚在一起，汽车公司被拖累了，所以我猜测可能在 1 年半后，到 2028 年，西方汽车公司参与人形机器人的事情开始好转。

由于非中国供应链（如 Agibot/Unitree）已经率先进入市场并且正在扩大生产规模，而西方对应公司尚未如此。

还有其他“纯粹的曝光”组件供应商，如 $VPG 和可能的 $TSLA / Figure 是我的猜测，但他们揭露的每台机器 150 美元的价格对于大规模生产来说有点小，当你达到 50,000 台的大规模生产时，收入约为 750 万美元。

因此，汽车公司的上游组件对于机器人来说没有任何意义，除非你达到巨大的规模...

这就是为什么我个人转向了对下游玩家如 Agility 的关注，而他们的估值相对较小。

我仍然认为我们很快就会看到一个像 Anthropic/OpenAI 那样的领导者出现在人形机器人/机器人领域，在所有其他物理 AI 供应商在该领域被炒作之前。

但这些只是我个人随时牢记的非正式想法，考虑到传统汽车公司在衰退中以及 2 年后新的增长矢量。

2026-07-05 11:58

原文：
Just high level Boston Dynamics + humanoid + automotive player directional thoughts:

As for their "speculated suppliers":

Samsung Electro-Mechanics / $INTC
LG Innotek vision sensing components into Atlas
Samsung Electronics / Sk Hynix / $MU - Memory
$NVDA - Jetson Thor compute
Google DeepMind
LG Energy Solution - reported BD Atlas battery supplier
$CATL - Battery
SPG Co / 에스피지 (058610.KQ) - Reducers / harmonic-drive-style robot gearing
SL Corp / 에스엘 (005850.KRX) - leg-module supply contract with Boston Dynamics
SOS Lab / 에스오에스랩 (464080.KQ) - Atlas LiDAR supplier expected
Sungwoo Hitech - Reported likely Atlas exterior-cover supplier
Hyundai Mobis - Atlas actuators; robot hand/gripper
ROBOTIS / 로보티즈 (108490.KQ) - Robot hand/gripper/robot actuators/servo modules
HL Mando / HL만도 (204320.KRX) - Actuators / automotive motion-control parts
ABB (ABBN.SWX) - actuator supplier/ecosystem
Samhyun / 삼현 (437730.KQ) - Smart actuators / motor-control auto parts
Seoyon E-Hwa / 서연이화 (200880.KRX)~₩0.31T Auto interior/exterior parts; potential robot exterior

- from some naver article that's getting passed around recently.

As for more Western names aside from the $GOOGl $NVDA / $MU types you have $AEVA through LG innotek.

$OUST was Velodyne legacy LiDAR agreement with BD Spot. Onsemi was SOS Lab LiDAR SPAD-array partner.

And probably a few others since supply chain is large.
_

I'm actually not bullish on Boston Dynamics commercialization, even though I am for them technologically.

So personally speaking, if they're not mass producing 30K like unitree soon, going long on the upstream supply chain doesn't really make sense.

Generally speaking, auto players are very interesting to me for broad humanoid exposure.

Names Schaeffler, which was working with ~45 humanoid makers, is probably in majority? But no confirmed proof yet. There's way too much robotics overlap with

2026-07-05 09:46

原文：
Leaked Anthropic docs show plans to secure 1.4GW capacity from Australia, amounting to ~$21.6B.

Recently $IREN, $SHAZ, and other Neoclouds have been building sovereign DCs in Australia.

Guess like a lot of things are stating to make more sense connection wise?

(disclosure: no open positions in any of the above).

翻译：
泄露的Anthropic文件显示，计划从澳大利亚获得1.4GW的容量，约合216亿美元。

最近，$IREN、$SHAZ和其他Neoclouds公司在澳大利亚建造了主权数据中心。

看起来很多事情开始从联系的角度变得有意义了？

（披露：我在上述任何公司中都没有开放的仓位）。

🔄

Loading...