我的关注

原文：
So I'm looking through $MRVL SEC filings for fun over the weekend.

For constraints they're citing:

"advanced wafer fabrication, advanced packaging, and large body substrates."

which results in increased lead times, inability to meet demand, and increased costs...

-> So $TSM is Marvell's sole source foundry for all of our advanced process-node wafers according to their filings (no surprise there).

the less obvious one is Large Body Substrates to look at:

ABF substrates are already known, but lage body substrates might be a more acute bottleneck inside ABF.

There's..

Daeduck Electronics (353200)
Kinsus (3189) - "large-area high-layer-count ABF substrates" already up +496.07% YTD tho
LG Innotek (011070) - Large AI substrates with "Developing large AI substrates with a NA Big Tech customer, mass production beginning H2 26"
$ATS +370.12%

for the "lesser known" names outside of Ibiden, Nan Ya PCB, and Unimicron (but many have already rerated).

In May they worded it as supply challenges have occurred / may occur and large-body substrates were in a continuing tight environment.

Now it's "We are currently in a supply constrained environment."

Maybe there's more emerging players that I missed that I'll go hunting for.

But just disclosure on what acute bottleneck $MRVL is facing.

翻译：
所以我周末闲着翻看 $MRVL 的 SEC 文件。

他们提到的限制因素是：

“先进晶圆制造、先进封装和大尺寸基板。”

导致交期延长、无法满足需求、成本上升……

→ 根据文件显示，$TSM 是 Marvell 所有先进制程晶圆的唯一代工（这并不意外）。

较不明显的因素是大尺寸基板，需要关注：

ABF 基板已经为人所知，但大尺寸基板可能是 ABF 内更紧迫的瓶颈。

有……

Daeduck Electronics (353200)
Kinsus (3189) - “大面积高层数 ABF 基板”，今年迄今已上涨 +496.07%
LG Innotek (011070) - 大型 AI 基板，文中称“与北美大型科技客户合作开发大型 AI 基板，量产预计 26 年下半年开始”
$ATS +370.12%

这些是 Ibiden、Nan Ya PCB、Unimicron 之外的“较不为人知”公司（但许多已被重新评级）。

5 月时他们的表述是供应挑战已出现/可能出现，且大尺寸基板仍处于持续紧张的环境。

现在的说法是“我们目前正处于供应受限的环境”。

也许还有我遗漏的其他新兴玩家，我会去寻找。

这只是披露 $MRVL 正面临的急性瓶颈。

原文：
Okay just to simplify this:

There is a guy who eats a lot called $NVDA. Nvidia only wants to buy high-priced potatos off the farmers.

A potato farmer $MU, that was selling carrots... now shifted farmland to potatos.

Nvidia caused a great potato famine cause they can eat a lot. So everyone else who farmed carrots like SK Hynix and Samsung moved their farmland capacity to grow potato.

But... people still need to eat. And carrots are a healthy part of their everyday diet..

Now there's no more carrots aside from a player like "ESMT", the legacy carrot farmer.

So now, because everyone buys carrots off ESMT, there's a shortage and the price goes up.

So ESMT needs more carrot seeds (from PSMC). But PSMC is running low on carrot seeds to grow the carrots.

So PSMC charges ESMT more for the seeds.

But the price ESMT sells the carrots at are much higher than what PSMC hikes the carrod seed price.

now... ESMT is making enough from selling carrots that it traded 1.9x P/E off July's earnings.

Can ESMT keep selling these carrots through 2027?

All the other farmers thinks so since it's both hard + low incentive to migrate their valuable potato farms back to carrot farms.

翻译：
好吧，简化一下：

有个吃得很多的家伙叫 $NVDA。Nvidia 只想从农民手里买高价土豆。

一个原本卖胡萝卜的土豆农场主 $MU…现在把耕地转向种土豆。

Nvidia 导致了严重的土豆荒，因为它吃得太多。于是其他种胡萝卜的公司，如 SK Hynix 和 Samsung，也把耕地容量转向种土豆。

但是……人们仍然需要吃东西，胡萝卜是日常饮食中健康的一部分。

现在除了像 “ESMT” 这样的老牌胡萝卜种植者，已经没有其他胡萝卜了。

于是因为大家都向 ESMT 购买胡萝卜，供不应求，价格上涨。

ESMT 需要更多胡萝卜种子（来自 PSMC）。但 PSMC 的胡萝卜种子库存已不足。

于是 PSMC 提高了种子价格向 ESMT 收费。

然而 ESMT 销售胡萝卜的价格远高于 PSMC 提价后的种子成本。

现在……ESMT 从卖胡萝卜中赚得足够，已实现 1.9 倍 P/E，超过七月的盈利。

ESMT 能在 2027 年之前继续卖这些胡萝卜吗？

其他所有农场主都认为可以，因为把有价值的土豆农场迁回胡萝卜农场既困难又缺乏动力。

wafer hike effects.

Maybe playing duoQ with Rank 1 Challenger "ESMT" in LoL classic... Is more fun than playing with hardstuck "Micron" in League?

翻译：
Yep，ESMT（3006）在多个传统存储细分市场都有敞口。

其产品组合包括：SLC/SPI NAND、NOR Flash、DDR2/DDR3、MCP/eMCP，所以这些老牌存储业务都覆盖到了。

如果你想看 ESMT 未来会受益的更大行业涨价情况：

1. SLC NAND：2026 年下半年约 +120–170%
2. DDR2：2026 年第三季度约 +35–40%
3. DDR4：2026 年第三季度 +30–40%（同时 DDR3 也在趋紧）
4. NOR：2026 年下半年高密度产品约 +60–65% 以上
5. MCP/eMCP：按我自己的估算，ESMT 可能环比约 +15–25%

官方拆分如下：

DDR2 + DDR3：约占营收 45%
DDR4/其他 DRAM：约占营收 10%
eMCP/MCP：约占 30%
模拟 + 其他：约占 15%

无论如何，关键还是持续时间/涨幅：

-> “主要全球 SLC NAND 供应商目前都没有计划在短期内新增产能。”
-> TrendForce 明确预计，库存消耗会让 SLC NAND 在第四季度进一步趋紧。
-> Winbond 撤出 DDR2，会让库存更紧张（而且他们的配额已经排到 2029-2030 年）
-> ESMT 管理层表示，市场的主要问题是供给，不是需求；公司晶圆代工配额只能覆盖客户订单的大约 60–70%
-> Nanya：“存储短缺可能持续到 2027 年底，且供需在 2028 年仍会偏紧”
-> Nanya 还明确表示，产能受限可能导致 DDR5、LPDDR5、DDR4、LPDDR4 和 DDR3 全面短缺。
-> 随着成熟制程资源被调去别处，ESMT 还把 eMCP 列为会受订单转移利好的产品之一
______

如果要我再说一遍我的看法：

我认为，尽管已经把 PSMC 45% 的晶圆涨价因素算进去了，但 ESMT 约 1.9 倍的 7 月年化运行市盈率仍然可能继续压缩。

行业评论支持这种强劲态势能持续到 2027 年。

更锦上添花的是，如果我们开始看到长期协议（LTA）以及 2029-2030 年的可见度（但目前还没计入）。

我觉得市场要么根本不知道这件事，要么是把持续时间定价错了，要么就是高估了 PSMC 涨晶圆价格的影响。

也许在 LoL 经典版里，和排名第一的 Challenger“ESMT”双排，比和卡在原地上不去的“Micron”一起玩更有意思？

原文：
Yep, ESMT (3006) has exposure to multiple legacy memory segments.

Has a portfolio of: SLC/SPI NAND, NOR Flash, DDR2/DDR3, MCP/eMCP, so all the fun legacy stuff.

If you want to look at broader industry price hikes coming up that ESMT has exposure to:

1. SLC NAND: ~120-170% for H2 2026.
2. DDR2: ~35-40% for Q3 2026.
3. DDR4: +30–40% Q3 (with DDR3 tightening)
4. NOR +60–65%+ for H2 (high density)
5. MCP/eMCP: maybe ~+15–25% QoQ for ESMT from my own estimate

For official breakdown:

DDR2 + DDR3: ~45% of revenue
DDR4/other DRAM ~ 10% of revenue
eMCP/MCP ≈30%
Analog + other ≈15%

Regardless, duration/hikes is the key:

-> "none of the major global SLC NAND suppliers plan to add new production capacity in the near term."
-> TrendForce specifically expects depleted inventories to make SLC NAND even tighter in Q4.
-> Winbond withdrawing DDR2, would make inventory even tighter (and they're doing allocations into 2029-2030)
-> ESMT management said the market's main problem was supply, not demand; the company's foundry allocations covered only roughly 60–70% of customer orders
-> Nanya: "memory shortages could last through the end of 2027, with supply and demand still tight in 2028"
->Nanya explicitly said capacity constraints could cause shortages across DDR5, LPDDR5, DDR4, LPDDR4 and DDR3.
-> ESMT included eMCP among the products receiving order-transfer benefits as mature node resources were being pulled elsewhere
______

If I had to state my thoughts again:

I think ESMT's ~1.9x July annualized run-rate P/E could compress further, despite factoring in PSMC's 45% wafer hike due to further legacy memory price hikes.

Industry commentary supports durability throughout 2027.

Cherry on top is if we start seeing LTAs and visibility into 2029-2030 (but not factored in).

I think markets either don't know about this, are mispricing duration, or overestimating PSMC's

原文：
$MU: goes from $133B -> $1T+ in a year.

$RDDT after a 2-month correction: "Garbage stock" https://t.co/HlIXwJ7ZdB

翻译：
$MU: 从 $133B -> $1T+，一年内。

$RDDT 在两个月回调后：“垃圾股” https://t.co/HlIXwJ7ZdB

原文：
Missed this mention earlier but IC design insiders reveal that OSAT providers have already warned clients:

"traditional packaging capacity could face a deficit exceeding 20% by 2027" (eg. $AMKR, ethereum:0x041ff0e49f6f774e7dc7bd10ee4a14c00b1d80b2, Powertech as potential beneficaries)

With: "wire bonding equipment" emerging as the primary bottleneck (ASMPT, $KLIC)

-> influx of AI server shipments has triggered a spike in demand for mature chips
-> existing mature packaging capacity at OSATs is being rapidly absorbed
-> securing early access to traditional packaging capacity is now becoming a challenge

Cascading bottleneck patterns are occurring across memory (eg. DDR2, DDR3) and now packaging capacity I guess.

翻译：
之前错过了这条提及，但 IC 设计内部人士透露，OSAT 供应商已经警告客户：

“传统封装产能到 2027 年可能出现超过 20% 的缺口”（例如 $AMKR、ethereum:0x041ff0e49f6f774e7dc7bd10ee4a14c00b1d80b2，Powertech 可能受益）

并指出：“线键合设备”正成为主要瓶颈（ASMPT，$KLIC）

-> AI 服务器出货激增导致对成熟芯片的需求飙升
-> OSAT 现有的成熟封装产能正被快速消化
-> 抢先获取传统封装产能已成挑战

记忆体（如 DDR2、DDR3）以及现在的封装产能正出现层层瓶颈。

这些分配能带来多少产能收入（美国分析师的 AOI 通话中有一半围绕这些）
- $GFS NPO/CPO/可插拔规模的影响以及下游终端客户机会
- Ayar / CPO 选手 2028 年的 ramp（曾募集 $500m+）或其他如 ELS 生产与 O‑net、TAM 机会
- 那 6 家新可插拔选手的规模，而不是问“为什么不能披露它们”
- 澄清 CFO 关于上市时间表和并购 TAM 扩张机会的法律说明

这样投资者才能了解 2027‑2028 年的未来增长。

然而，因引导性问题的提法/回答方式，整个通话都围绕当前财务（

原文：
Yes. I genuinely don't have a clue why $SIVE still focuses on local Swedish markets for communication/private fundraising.

Over the US.

The questions you get from Swedish heavy audiences include:

- Leading questions like how would you "stop the bleeding"
- "Why can't you disclosure your private customers"
- Why focus on transceivers (it's obvious)
- Implications for non-photonics orders

So 1/3rd of the time, the CEO needs to be defending leading/accusatory questions around what "opportunity pipeline means"

US analysts would have pressed:

- Economic implications from 2 foundry allocations during a bottleneck + ASP hike/operating leverage
- Around much capacity revenue would those allocations lead to (half of AOI calls are around those from US analysts)
- $GFS NPO/CPO/pluggable scale implications and downstream end customers opportunities
- Ayar / CPO player 2028 ramp (that happened to raise $500m+) or other products like ELS production with O-net and TAM opportunities there.
- How big those 6 new pluggable players were, rather than asking "why can't you disclosure them"
- Clarifying CFO legalposts around listing timeline and M&A TAM expansion opportunities

So investors can get a sense of future growth for 2027-2028.

Instead, the way leading questions are framed/answers, the entire calls were focused around current financials (which US investors don't care about), non-photonic implications, or skepticism of wording.

Rather than exciting US investors about economic implications from their time in a CW laser bottleneck.

The more time $SIVE spends in Swedish markets, the more they will be valued as such.

翻译：
是的。我真的不明白为什么 $SIVE 仍然把沟通/私募重点放在瑞典本地市场。

而不是美国。

来自瑞典重度受众的问题包括：

- 引导性问题，例如你会如何“止血”
- “为什么不能披露你的私有客户”
- 为什么聚焦收发器（显而易见）
- 对非光子订单的影响

所以大约三分之一的时间，CEO 需要为“机会管线”意味着什么的引导/指责性问题进行辩护。

美国分析师会关注：

- 在瓶颈期间两家代工分配的经济影响 + ASP 上涨/运营杠杆
-

最重要的是，谁在乎跟客户竞争？如果客户说：“如果你们做可插拔，我们就把激光业务交给别家做”

那又是谁？$LITE / $COHR / $AAOI 也都把激光产能转去内部使用了。你们很多亚洲玩家也已经分配好了。

现在就该尽可能利用这个产能瓶颈，因为几乎已经没有合格的选择了。

而且我敢肯定，所有合作伙伴也都在试图向上游垂直整合到激光层面，并参与竞争。

如果在 Q2 那个时间窗口里就推进新的 OE/可插拔布局，情况会完全不一样。

_

总结：

$SIVE 应该瞄准下一个 $LITE，像硅谷公司那样迅速扩张。

把那 7000 万美元资本用来在 H2 加速推进，至少到那时完成就绪。并预计 NASDAQ 上市在 2027 年 H1 完成。

不是在那个时间点还停留在“评估”。

同时利用资本和股权，借助你的估值向下游扩展到光引擎/ELS/光收发器，并收购一家 Celestial/Ayar/Cloud Lite 这样的初创公司。

至于电话会里的措辞，市场关心的是代工产能分配 + 6 家可插拔玩家带来的经济价值，而不是其他业务板块。

$SIVE 迄今一直非常保守，但他们需要加速，并且要面向有前瞻性的美国/国际股东进行沟通。

不要只迎合那些更关心 Q2/TTM 收入的瑞典本地受众。

否则他们会被当成亚洲零部件供应商，最终也会被按欧盟公司的估值来定价。

already allocated.

Abuse the current bottleneck as much as possible because there's almost no qualified choices left.

And I'm certain all the partners are trying to vertically integrate upward toward the laser level too and compete.

If there was a new OE/pluggable accusation attempted during the Q2 time frame, things would have been a lot different.

_

TLDR:

$SIVE should aim to be the next $LITE and blitzscale like a Silicon Valley company.

Use that $70m capital to move faster H2, and at the bare minimum finish readiness then. And expect NASDAQ listing to be finished H1 2027.

Not "evaluation" at that timeframe.

And use the capital + equity to expand downward into optical engines/ELS/optical transceivers using your equity valuation, and buy a Celestial/Ayar/Cloud Lite type startup.

As for the wording of the call, markets care about economic value of the foundry allocation + 6 pluggable players, not other business segments.

$SIVE has been extremely conservative to date, but
they need to speed up and communicate to forward looking US/Int shareholders.

Not geared toward local Swedish audiences who care about Q2/TTM revenue.

Otherwise they'll be treated like an Asian component supplier and eventually be valued like an EU one.

翻译：
如果我要挑 $SIVE 这次财报电话会的毛病：

1. 管理层对最重要的两个披露不够重视。

-> 6 个可插拔业务合作
-> 铸造产能（在行业产能瓶颈期）

这才是电话会最关键的部分，本应是重点，用来体现他们的经济规模。但我们只听到了 2-3 句。

而且本不该还得靠 X 上一个二次元头像来解释这些内容的含义。

2. CFO 需要重新培训，或者换一个新的。

作为股东，听到那些律师式的回答真的很烦。连我都被这句话气到了：

“我们会评估时机，并决定在那个时候条件是否合适再推进。”

到现在就应该默认他们会申请双重上市，以摆脱充满敌意的瑞典市场。

股东出资不是为了听这种律师腔的回答。

再说一次，这可能是 CFO 能给出的最差回答之一了，必须给出明确承诺，并且时间表要更快。

3. 7000 万美元本该用于并购和双重上市。

-> 说实话，这个阶段把资本投去做混合制造，纯粹是在浪费资本；我听到他们在这个时间点提这个，真的很失望。

这件事最终是需要做的，但前提是有 2 家代工供应商。

-> 你现在有大量新增资本，还有很高的市值。

应该用它去做类似 $AVGO 的收购，把 Cloud Light 这种面向可插拔业务的 IP，或者光引擎相关资产拿下来。

Lumilens 2 年内从 0 做到 55 亿美元，现在还拿到了超大规模云厂商合作。

Sivers 应该尽快跳出激光这个瓶颈，不要只做零部件供应商。

原文：
If I had to be critical of $SIVE earnings:

1. Management underweighted the two disclosures that matter most.

-> 6 pluggable engagements
-> Foundry capacity (during an industry bottleneck)

That's the most important part of the call and should have been the focus to provide economic scale of them. But we got 2-3 sentences about it.

And it shouldn't take an Anime avatar on X to talk most about the implications of those.

2. CFO needs retraining or bring on a new one.

The lawyer like answers were infuriating to listen to as a shareholder. Even I got pissed off with this statement:

"we will evaluate the timing and decide whether the conditions are right to move forward at that particular time."

It should be a 100% given by now they file for dual listing to escape hostile Swedish markets.

Shareholders didn't fund a future growth type company to hear lawyer-like answers.

Again, this was probably one of the worst answers the CFO could have gave and there needs to be firm commitment along with faster timelines.

3. $70M should have gone toward M&A and dual listing.

-> Brutally honestly speaking, it's a waste of capital to focus on hybrid manufacturing at this stage and I was disappointed to hear this during this timeframe.

It's eventually needed but with 2 foundry suppliers

-> You have a ton of new capital and large marketcap.

Use it to pull $AVGO style acquisitions of Cloud Light style IP for pluggables or optical engines.

Lumilens went from 0 -> $5.5B in 2 years and now with hyperscaler engagements.

Sivers should expand out of the laser chokepoint as fast as possible and not stay just a component vendor.

And most of all, who cares about competing with customers? If a customer says: "if you do pluggables, we'll go with other players for lasers"

who?? $LITE / $COHR / $AAOI reroutes their lasers to internal usage. Lot of your Asian players

原文：
Something else to look at aside from DDR2/DDR3

Are your legacy SLC NAND fabless companies with wafer allocations given the 120-170% contract pricing revisions.

Etron/ESMT are in this same bottleneck too (with varying exposure).

Then there’s others like Jeju Semiconductor (080220) (need to look at inventory/revenue mix)

Where Q2 annualized is 4.5x p/e ish. But if you’re hiking by150-170%, numbers should compress a lot depending on exposure.

翻译：
除了 DDR2/DDR3 之外，还可以关注其他方面

你的传统 SLC NAND 无晶圆代工公司在获得 120‑170% 合同价格修订后，是否仍有晶圆配额。

Etron/ESMT 也处于同样的瓶颈（暴露程度不同）。

还有像 Jeju Semiconductor（080220）（需要查看库存/收入结构）

其中 Q2 年化约为 4.5 倍 p/e。但如果涨幅在 150‑170%，根据暴露程度，数字应该会大幅压缩。