觅信聚合

我的关注

这市场谁在赚钱啊，亏了整整一个月了 https://t.co/ZnX3MqibFd

<@&1519026403734061156>

<@&1519026403734061156>#黄金##XAU# 进一步验证3942这个上涨是调整浪，不是见底大反弹更不是新的上涨。

<@&1519026403734061156>

<@&1519026403734061156>#SK海力士##000660# 见顶以后调整非常凶猛。上周给出了回撤区间，今天到了该区间了。这里即将见底反弹，形成空头陷阱。图二分享位置。专栏 · SK海力士

世界可以毁灭了，累了，就这样吧。 https://t.co/KmVehw6DII

<@&1519026403734061156>

<@&1519026403734061156>#原油##CL!# 6月28号给出了上涨目标74.73实现了，7月9号给出了回撤区间也实现了。图二是今天7月13日给出的上涨目标。专栏 · 原油

2026-07-13 00:45

年将其光子集成电路（PIC）容量扩大 30 倍，从每月 500 张晶圆扩大到 25,000 张

- Hanmi 半导体进入 CoWoS 封装设备市场

- $NVDA 和 NTT 将于 7 月 24 日举办会议讨论 CPO 策略

- 今年 NAND 闪存价格将上涨 5 倍，明年市场规模将达到 4890 亿美元，由 RAG/推理触发。三星和 SK 海力士进行紧急厂房投资，NAND 设备供应商大幅增长

- 燃气轮机瓶颈，40% 的供应缺口伴随着惊人的 5 年交货周期（大型厂房需要它们进行大规模生产）

- AI 探针卡组装已经遇到严重的瓶颈，显然需要像 Innovation Service 的机器等解决方案

- 南亚科 79.5% 的毛利率，内存大幅增长。4 倍的资本支出用于容量和先进封装

- 中芯星的 STAR 市场 IPO 将于 7 月 16 日举行，这应该会给内存供应链中的玩家带来很多关注

- KYEC 在美国投资 14 亿美元用于 $TSM 亚利桑那州的产出测试设施。像 $AMKR 这样的许多玩家应该会在 2028 年大幅增长

- $INTC 首席执行官上个月警告说，氦气可能会阻碍 AI 芯片的制造成本和交货时间

- 3D NAND 词线从 375 层节点开始从钨转向钼

- SambaNova 为 AI 推理获得了 JPM 的支持，并以 110 亿美元的估值筹集了 10 亿美元资金

- 预计 2027 年 HBM 价格将翻倍，因为 $NVDA 的 Rubin 平台推动了需求

- $MU 提供 5 亿美元的资金支持 GlobalWafers 的美国制造能力

- $META 的 'Iris' 将于 9 月通过 $AVGO 和 TSMC 进入大规模生产，Meta 旨在将计算能力增加到 2027 年的 14GW

- 大立光学已经获得了其首个 CPO FAU 订单，计划于明年中旬开始大规模生产（这可能是 Foci 和其他公司的迹象）

- 三星和 SK 海力士已经推迟了混合键合封装技术的实施，用于 HBM4

- 内存成本（DRAM/NAND）已经占据了 400 美元以下智能手机的 60% 的 BOM，从而导致了严重的销量收缩

- SK 海力士成功通过纳斯达克 ADR 筹集了 265 亿美元

- $TSLA 发布了采购指南，要求供应商在 9 月之前每周生产 1000 台，并在年底前将生产量增加到 2000-2500 台，以满足其第三代 Optimus 机器人的需求。显然，Alliance Technology 和 A-Link 可能处于这个供应链中，用于谐波减速器和视觉镜头

每天都会浏览这些内容，但不想成为新闻记者，所以只是整理了我觉得有趣的东西。

2026-07-13 00:45

原文：
Just a semi recap TLDR:

- $GLW glass bridge per Morgan Stanley has potential, but hard to displace FAU (like FOCI) in short term

- $SPCX Starlink Gen 3 is scaling to 100,000 units (10x prev gen) creating possible capacity constraints for suppliers of switches to CCL.

- PCB supply shortages are projected to persist until 2028, and component shortages/price hikes are already forcing ODMs like Inventec to issue conservative H2 shipment

- DeepSeek and Zhipu are developing custom ASICs to bypass $NVDA (kinda expected by now).

- Anthropic has achieved a $30B ARR and is projected to hit >$1B in Q3 profit. Turns out these frontier labs are more profitable than people think.

- US admin pressured $AAPL to source from $INTC, in exchange for tariff exception. Kinda pressuring alignment away from TSM.

- $TSM plans a 30x expansion of its Photonic Integrated Circuit (PIC) capacity by 2028, growing from 500 to 25,000 wafers per month

- Hanmi Semiconductor is entering the CoWoS packaging equipment market

- $NVDA and NTT hosting a conference July 24th to discuss CPO strategies.

- 5x general NAND flash price hike this year, market size $489B by next year triggered by RAG/inference. Samsung and SK Hynix make emergency fab investments, NAND equipment suppliers go brrrr.

- Gas turbine bottleneck, 40% supply gap accompanied by staggering 5-year delivery cycles (mega-fabs need them for mass production).

- AI probe card assembly has hit a severe bottleneck apparently. Which is attempted to by solved with things like Innovation Service's machines.

- Nanya, 79.5% gross margins reported, memory go brr. 4x capex for capacity/advanced packaging.

- CXMT's STAR Market IPO on July 16, so should bring a lot of attention to memory players in that supply chain.

- KYEC $1.4B investment into US for $TSM Arizona output test facilities. Lot of these players like $AMKR and

2026-07-13 00:45

others should go brr in 2028.

- $INTC CEO warned last month that helium could hinder the manufacturing costs and delivery times of AI chips

- 3D NAND word lines are shifting from Tungsten to Molybdenum starting at the 375-layer node

- SambaNova secured JPM for AI inference and raised $1B at an $11B valuation.

- HBM prices projections to double in 2027 as $NVDA Rubin platform drives demand.

- $MU provides $500 million in funding to support GlobalWafers' US manufacturing capacity

- $META 'Iris' will enter mass production in September via $AVGO and TSMC and Meta aims to double computing power to 14GW by 2027.

- Largan Precision has secured its first CPO FAU order, mass production slated for middle of next year (kinda indication around Foci and others).

- Samsung and SK Hynix have delayed the implementation of hybrid bonding packaging technology for HBM4 apparently

- Memory costs (DRAM/NAND) have reached 60% of the BOM for sub-$400 smartphones, causing a severe volume contraction.

- SK Hynix successfully rasied $26.5 billion through a Nasdaq ADR

- $TSLA issued procurement guidelines requiring suppliers to reach weekly production of 1,000 units by September and double to 2,000-2,500 by year end for its 3rd gen Optimus robot. Apparently Alliance Technology and A-Link may be in this supply chain harmonic reducers and vision lenses?

Kinda go through all this stuff every day, but don't wanna be a news reporter so just consolidated stuff I found interesting.

翻译：
以下是半年总结的简要版：

- 摩根士丹利认为 $GLW 的玻璃桥梁具有潜力，但短期内难以取代 FAU（如 FOCI）

- $SPCX 的 Starlink Gen 3 将扩大到 10 万台（比前一代扩大 10 倍），可能会为 CCL 的开关供应商带来容量限制

- 预计 PCB 供应短缺将持续到 2028 年，组件短缺和价格上涨已经迫使像 Inventec 这样的 ODM 发布保守的第二季度出货量

- DeepSeek 和 Zhipu 正在开发自定义 ASIC 以绕过 $NVDA（早就料到了）

- Anthropic 已实现 300 亿美元的年收入，并预计在第三季度实现超过 10 亿美元的利润。原来这些前沿实验室比人们想象的更有利可图

- 美国政府迫使 $AAPL 从 $INTC 采购，以换取关税豁免。有点迫使与 TSM 脱离关系

- $TSM 计划到 2028

Just a semi recap TLDR: - $GLW glass bridge per Morgan Stanley has potential, but hard to displace FAU (like FOCI) in short term - $SPCX Starlink Gen ...

Just a semi recap TLDR:- $GLW glass bridge per Morgan Stanley has potential, but hard to displace FAU (like FOCI) in short term- $SPCX Starlink Gen 3 is scaling to 100,000 units (10x prev gen) creating possible capacity constraints for suppliers of switches to CCL.- PCB supply shortages are projected to persist until 2028, and component shortages/price hikes are already forcing ODMs like Inventec to issue conservative H2 shipment- DeepSeek and Zhipu are developing custom ASICs to bypass $NVDA (kinda expected by now). - Anthropic has achieved a $30B ARR and is projected to hit >$1B in Q3 profit. Turns out these frontier labs are more profitable than people think. - US admin pressured $AAPL to source from $INTC, in exchange for tariff exception. Kinda pressuring alignment away from TSM. - $TSM plans a 30x expansion of its Photonic Integrated Circuit (PIC) capacity by 2028, growing from 500 to 25,000 wafers per month- Hanmi Semiconductor is entering the CoWoS packaging equipment market- $NVDA and NTT hosting a conference July 24th to discuss CPO strategies. - 5x general NAND flash price hike this year, market size $489B by next year triggered by RAG/inference. Samsung and SK Hynix make emergency fab investments, NAND equipment suppliers go brrrr. - Gas turbine bottleneck, 40% supply gap accompanied by staggering 5-year delivery cycles (mega-fabs need them for mass production).- AI probe card assembly has hit a severe bottleneck apparently. Which is attempted to by solved with things like Innovation Service's machines. - Nanya, 79.5% gross margins reported, memory go brr. 4x capex for capacity/advanced packaging. - CXMT's STAR Market IPO on July 16, so should bring a lot of attention to memory players in that supply chain. - KYEC $1.4B investment into US for $TSM Arizona output test facilities. Lot of these players like $AMKR and others should go brr in 2028.- $INTC CEO warned last month that helium could hinder the manufacturing costs and delivery times of AI chips- 3D NAND word lines are shifting from Tungsten to Molybdenum starting at the 375-layer node- SambaNova secured JPM for AI inference and raised $1B at an $11B valuation.- HBM prices projections to double in 2027 as $NVDA Rubin platform drives demand.- $MU provides $500 million in funding to support GlobalWafers' US manufacturing capacity - $META 'Iris' will enter mass production in September via $AVGO and TSMC and Meta aims to double computing power to 14GW by 2027.- Largan Precision has secured its first CPO FAU order, mass production slated for middle of next year (kinda indication around Foci and others). - Samsung and SK Hynix have delayed the implementation of hybrid bonding packaging technology for HBM4 apparently - Memory costs (DRAM/NAND) have reached 60% of the BOM for sub-$400 smartphones, causing a severe volume contraction.- SK Hynix successfully rasied $26.5 billion through a Nasdaq ADR- $TSLA issued procurement guidelines requiring suppliers to reach weekly production of 1,000 units by September and double to 2,000-2,500 by year end for its 3rd gen Optimus robot. Apparently Alliance Technology and A-Link may be in this supply chain harmonic reducers and vision lenses?Kinda go through all this stuff every day, but don't wanna be a news reporter so just consolidated stuff I found interesting.

2026-07-12 18:28

原文：
Looks like the Strait of Hormuz is open for the 50th time this year per Trump.

I’m getting tired looking at macro. https://t.co/GuYgkwC2Q8

翻译：
看起来霍尔木兹海峡今年已经是第50次在特朗普任期内开放了。

我已经厌倦了看宏观经济数据了。 https://t.co/GuYgkwC2Q8

Looks like the Strait of Hormuz is open for the 50th time this year per Trump. I’m getting tired looking at macro.

Looks like the Strait of Hormuz is open for the 50th time this year per Trump.I’m getting tired looking at macro.

應該是上周提過的半導體降溫後有利於大盤創高應該就近期看有沒有消息刺激上去

2026-07-12 11:25

原文：
Apparently there’s a new popular trend to bearpost trillion dollar companies.

Like $NVDA or $TSM.

By finding isolating technical details, then blowing it out of proportions by claiming their whole program is delayed.

-> if the company doesn’t respond, stock prices/sentiment are damaged.

-> if the company issues a response, they can’t go into detail since supply chains are confidential… but people still doubt them

-> Poster gain more impressions, then does it again with more views the next time.

Don’t quite think this is very healthy, with more folks copying that playbook. Especially around incentive structures.

翻译：
显然，有一种新的流行趋势，就是针对市值万亿美元的公司进行负面帖子。

比如 $NVDA 或 $TSM。

通过找到某些技术细节，然后夸大其词，声称他们的整个项目都被延迟了。

-> 如果公司不回应，股价/市场情绪就会受到损害。

-> 如果公司发出回应，他们不能透露细节，因为供应链是保密的…… 但人们仍然会怀疑他们

-> 发帖人获得更多的浏览量，然后下次再这样做，获得更多的浏览量。

我不认为这是一种健康的趋势，尤其是在激励结构方面，更多的人会复制这种做法。

Apparently there’s a new popular trend to bearpost trillion dollar companies. Like $NVDA or $TSM. By finding isolating technical details, then blowin...

Apparently there’s a new popular trend to bearpost trillion dollar companies.Like $NVDA or $TSM.By finding isolating technical details, then blowing it out of proportions by claiming their whole program is delayed.-> if the company doesn’t respond, stock prices/sentiment are damaged. -> if the company issues a response, they can’t go into detail since supply chains are confidential… but people still doubt them-> Poster gain more impressions, then does it again with more views the next time.Don’t quite think this is very healthy, with more folks copying that playbook. Especially around incentive structures.

<@&1519026403734061156>

<@&1519026403734061156>量子计算机，再不进行抗量子升级的话，会很难看。有一些在行动，指数还没有行动的意思。

2026-07-12 11:01

原文：
Was just doing more research into Nippon Chemical since my post on disrupting the optical supply chains with:

NCI red phosphorus (#1 in the world w/ Rasa) -> $AXTI + InP substrate -> $IQE / $TSEM | $LITE / $COHR etc.

There's the MLCC chain with high purity barium carbonate / barium titanate:
-> TDK (they have a JV for ceramics)
-> MLCC (current bottleneck)

and they have relationships with Murata, high confidence for MLCC too. Apparently they own ~.05% of Murata? Which I found hilarious, since it's now worth ~$55M. And 280,000 shares of Adeka.

Then there's phosphine gas:
-> Goes into Entegris
-> Kioxia/ $MU / Sk Hynix / $TSM type fabs.

this also likely ends up in $IQE, Landmark, VPEC, $COHR as well so dual exposure there.

Looking into phosphoric acid, which likely is involved in NAND/Kioxia processes + ADEKA partnership.

Basically I found this company back at $180m or so, which is pretty interesting that it could easily break so many different supply chains?

Feels like one of the most important companies in the world lol.

But all these different downstream bottlenecks leads to close to little material revenue/operating profit growth.
Just due to Japanese cultural reasons of not wanting to hike prices.

Still doing a lot of research though, companies like $SIVE or NCI are much more interesting to me, just to see how far the rabbit hole goes.

翻译：
刚刚对日本化学公司进行了更多的研究，因为我之前发表了关于破坏光学供应链的帖子：

日本化学公司的红磷（全球排名第一，与Rasa合作）-> $AXTI + InP基板 -> $IQE / $TSEM | $LITE / $COHR 等。

还有高纯度的碳酸钡/钛酸钡的MLCC链：
-> TDK（他们在陶瓷领域有合资企业）
-> MLCC（当前的瓶颈）

他们与村田制作所有关系，对MLCC也很有信心。显然，他们拥有村田制作所大约0.05%的股份，我觉得很有趣，因为现在这已经价值约5500万美元了。他们还拥有28万股Adeka的股份。

然后还有磷化氢气体：
-> 流向Entegris
-> Kioxia/ $MU / Sk Hynix / $TSM类型的晶圆厂。

这也可能最终流向 $IQE, Landmark, VPEC, $COHR等公司，因此有双重曝光。

正在研究磷酸，这可能参与NAND/Kioxia的工艺以及与ADEKA的合作。

基本上，我在公司市值大约1.8亿美元时发现了这家公司，这很有趣，因为它可以轻松打破这么多不同的供应链？

感觉像世界上最重要的公司一样。

2026-07-12 11:01

但所有这些不同的下游瓶颈导致了很少的材料收入/营业利润增长。
只是由于日本文化原因，不想提高价格。

仍然在进行大量的研究，像 $SIVE 或 NCI 这样的公司对我来说更有趣，只是想看看兔子洞有多深。

[Subscribers Only] Was just doing more research into Nippon Chemical …

2026-07-12 10:13

原文：
So there's a IBK Research report on Boston Dynamics value chains from last month.

Just a summary:

IBK maps these companies to Atlas as suppliers:
- Hwashin (010690) / body, arms, legs
- LG Energy (373220) / battery
- Hyundai Autoever (307950) / integration
- Hyundai Mobis (012330) / actuators

As for humanoid volume ramps:

They're modeling for, 11.29K in 2028, 20k 2029, 30k in 2030... 40k in 2031, and 50k in 2032.

Not quite sure why IBK and other institutions are a fan of linearly modeling S-curve volume ramps...

Like adding +10K per year, don't quite think it's volume ramp is going to work like that... if I had to guess it would look more like:

- 15-20k 2028
- 40k-70k 2029
- 90k-140k for 2030

Since Boston Dynamics is projecting 30k capacity by 2028 (I'm sure they'd aim to get more online by 2029-2030), as China collectively is already doing 100k EOY in 2026.

In terms of competitive landscape they name:

- $TSLA, Figure, Apptronik, $CCXI (Agility), as US players. Then Boston Dynamics (Korea parent owned now)
- Unitree, Fourier, AGibot, UBtech, $XPEV as the Chinese leaders.
- Neura, Pal Robotics, Wandercraft, Oversonic, as the EU leaders.

They also did quite a lot of valuation modeling around Hyundai Mobis/Hyundai Autoever/Glovis.

Regarding the BD economic ownership from 27.9% from Hyundai Motor, 11.3% from Mobis, and 11.3% from Glovis. So at least institution are valuing humanoid segments inside companies lot more now.

Then from report assumptions:
- 31 actuators per Atlas
- $1K per actuator in 2028
- $134K Atlas ASP,

Which implies actuator cost of final selling price is roughly 23–28%.

Probably the more interesting statement was IBK stated that actuator capacity is biggest signal for volume ramp. eg. every 310,000 units of actuator capacity supports 10,000 more robots.

So tracking actuator outputs, yields, ASP is a cleaner read on

2026-07-12 10:13

2028-2030 ramp.

I didn't have much personal takeaways, but hopefully others find it interesting, maybe around Hwashin as a core supplier or around actuator capacity as an indicator.

翻译：
上个月，IBK 研究机构发布了一份关于波士顿动力（Boston Dynamics）价值链的报告。

简单总结如下：

IBK 将以下公司映射到 Atlas 供应商：
- Hwashin (010690) / 机身、手臂、腿部
- LG Energy (373220) / 电池
- Hyundai Autoever (307950) / 集成
- Hyundai Mobis (012330) / 执行器

关于人形机器人产量的增长：

他们预测2028年为11.29K，2029年为20K，2030年为30K... 2031年为40K，2032年为50K。

不太清楚为什么 IBK 和其他机构喜欢线性地建模 S 曲线产量增长...

就像每年增加 +10K，这样增长产量不太可能实现... 如果我要猜测，产量增长可能更像：

- 2028 年：15-20K
- 2029 年：40K-70K
- 2030 年：90K-140K

由于波士顿动力预计到 2028 年将达到 30K 的产能（我相信他们会努力在 2029-2030 年在线上增加更多产能），而中国已经在 2026 年年底实现了 100K 的产量。

在竞争格局方面，他们提到了：

- $TSLA、Figure、Apptronik、$CCXI（Agility）作为美国玩家。然后是波士顿动力（现在由韩国母公司所有）
- Unitree、Fourier、AGibot、UBtech、$XPEV 作为中国领军企业。
- Neura、Pal Robotics、Wandercraft、Oversonic 作为欧盟领军企业。

他们还对现代モビス（Hyundai Mobis）、现代オートエバー（Hyundai Autoever）和Glovis 进行了大量的估值建模。

关于波士顿动力的经济所有权，现代汽车拥有 27.9%，モビス拥有 11.3%，Glovis 拥有 11.3%。因此，至少机构现在更看重公司内部的人形机器人业务。

然后，从报告假设中可以看出：
- 每个 Atlas 机器人有 31 个执行器
- 2028 年每个执行器的成本为 1K 美元
- Atlas 机器人的平均售价为 134K 美元

这意味着执行器的成本占最终售价的 23-28%。

可能更有趣的陈述是 IBK 表示，执行器的产能是产量增长的最大信号。例如，每 310,000 个执行器的产能可以支持 10,000 台机器人的生产。

因此，跟踪执行器的产量、产量率和平均售价是了解 2028-2030 年产量增长的更清晰的指标。

我没有太多的个人感想，但希望其他人会觉得它有趣，也许是关于 Hwashin 作为核心供应商或关于执行器产能作为指标。

🔄

Loading...