我的关注

failed my own cash conversion test (so I placed them in the inventory beneficiaries).

I'm sure I might have missed some nuance here, but for the sake of time/modeling, wanted a $SNDK style name that was actively converting into profitability.

Since a lot of them just stocked up or profited off on inventory (which also should translate to earnings) but... despite their low P/E actually lost money H1 or with Transcend issued convertible bonds/dilution to fund it.

- Transcend is probably the one I'm going to look at a bit further, just did lite DD so far since gross margin hit 77%.

- PSMC fwd p/e is a bit higher compared to the rest but they're hiking wafer costs ~45%, so maybe you get ~6.7–9.5× forward P/E, buttt maybe they keep hiking.

TLDR:

-> Liked Etron, ESMT since there's a lot of earnings discovery recently, and would get priced in sooner or later.
-> Nanya/Winbond are probably the most durable of them all.

I do think DDR2/DDR3 trade is more uncharted terrtiory versus your HBM/NAND Samsung/Sk Hynix/Micron, etc.

And we're starting to see another $SNDK type cycle (but maybe less durable, not sure).

Just a notes/thoughts here.

翻译：
我其实把一笔可观的资金配置到了旧制程 DRAM 交易上（今年 Q1 也做了旧制程 NAND/NOR）。

而且我把旧制程 DRAM 变成了一个中期交易。因为我开始看到明显的提价和供需失衡表述。

下面只是我对 7 月运行率的一些个人笔记：

ESMT (3006) - $2.53B 市值，1.94x P/E
Etron (5351) - $1.19B，4.15x P/E
Winbond (2344) - $25.3B，~6.2x P/E
Nanya (2408) - $55.3B，~5.5x P/E
PSMC (6770) - $9.4B，~19.6x P/E
TeamGroup (4967) - $728M，~2.4x P/E
Transcend (2451) - $4.06B，~3.3x P/E
Innodisk (5289) - $4.57B，~3.1x P/E
Apacer (8271) - $937M，~3.2x P/E
ADATA (3260) - $4.40B，~2.3x P/E
Silicon Power (4973)，$318M，~4.1x P/E

- ESMT（fabless）净利润增长异常强劲，是台湾的晶圆供应商，成本可以顺价转嫁，DDR2/DDR3 结构性供给不足。我认为它的盈利持续性会延续到 2027 年，这也是为什么 1.94 倍 P/E 非常有吸引力。

- Etron（fabless）我之前提过，但它的月度盈利在持续爆发式增长，而且他们和 Nanya 有一个合资项目，会在 2027 年启动，所以它大概是我第二喜欢的标的。

- Winbond

可能是我上面最喜欢的，因为他们确实拥有晶圆厂，并且能覆盖下游产品，而不只是像 PSMC 那样只做晶圆。我认为目前市场对它的需求可见度已经延伸到 2029-2030 年，最初讨论出来的结果是这样。它现在可能已经不算完全被忽视的名字了，但我确实认为在提价和较低的远期 P/E 下会被重估。也许 Winbond 对应 2027 年盈利 3.5 倍左右……这其实也非常强，但 ESMT 现在 2026 年盈利 1.9 倍就更离谱了。

Nanya 也类似，它拥有 DRAM 晶圆厂，但 Nanya 已经被研究得很充分了。

- TeamGroup 预计 2027 年上半年会出现更严重的存储短缺，2027 年之后继续收紧，而且他们库存非常高。但这更像是强库存驱动的利润。

- TeamGroup、ADATA（库存极重）、Apacer、Innodisk 和 Silicon Power 的库存都在快速增长，但没通过我自己的现金转换测试，所以我把它们放进了“库存受益者”这一类。

我确定自己可能漏掉了一些细节，但为了节省时间/建模，我想找的是一个类似 $SNDK 那种、正在积极转化为盈利的标的。

因为很多公司只是囤了库存，或者靠库存获利（这也应该会反映到业绩里），但……尽管它们 P/E 很低，H1 还是亏钱了，或者像 Transcend 那样发了可转债/稀释股本来融资。

- Transcend 可能是我接下来会进一步研究的对象，我目前只做了浅度尽调，因为它的毛利率已经到 77% 了。

- PSMC 的远期 P/E 比其他几家高一些，但他们把晶圆价格上调了大约 45%，所以也许对应的是约 6.7–9.5 倍的远期 P/E，不过……他们也可能继续提价。

总结：

-> 我喜欢 Etron、ESMT，因为最近盈利发现很多，迟早会体现在估值里。
-> Nanya/Winbond 可能是里面最稳的。

我觉得 DDR2/DDR3 这条线比你提到的 HBM/NAND、Samsung/Sk Hynix/Micron 等更像是未充分开发的领域。

而且我们开始看到又一轮类似 $SNDK 的周期（但可能没那么持久，不确定）。

先做一些笔记/想法。

原文：
I actually allocated a sizeable amount to the legacy DRAM trade (did legacy NAND/NOR Q1 this year).

And made legacy DRAM a medium term trade. Since I'm starting to see substantial price hikes and demand imbalance statements.

Just some of my personal notes on July runrates:

ESMT (3006) - $2.53B MC, 1.94x P/E
Etron (5351) - $1.19B, 4.15x P/E
Winbond (2344) - $25.3B, ~6.2x P/E
Nanya (2408) - $55.3B, ~5.5x P/E
PSMC (6770) - $9.4B, ~19.6x P/E
TeamGroup (4967) - $728M, ~2.4x P/E
Transcend (2451) - $4.06B, ~3.3x P/E
Innodisk (5289) - $4.57B, ~3.1x P/E
Apacer (8271) - $937M, ~3.2x P/E
ADATA (3260) - $4.40B, ~2.3x PE
Silicon Power (4973), $318M, ~4.1x P/E

- ESMT (fabless) extraordinary net income growth, wafer supplier in Taiwan, pass through cost, DDR2/DDR3 structurally undersupplied, I see earnings durability throughout 2027, which is why 1.94X P/E was extmreley interesting

- Etron (fabless) I covered before, but monthly profitability is exploding over time, they have a JV with Nanya which kicks off in 2027, and probably my second favorite name

- Winbond, is probably my favorites up there since they actually own fabs and captures downstream products, rather than just wafers like PSMC. Demand visibility I think extends to 2029-2030 now form initial discussions, not exacty an undiscovered name anymore, but I do think it gets re-rated given price hikes/low forward p/e. Maybe 3.5x 2027 earnings for Winbond.. which is also really stellar, but 1.9x current 2026 earnings from ESMT was just insane.

Would say something similar to Nanya that owns DRAM fabs, but Nayna is already well covered.

- Teamgroup expected worse memory shortage around H1 2027, tightening beyond 2027, and they have massive inventory. But it's very inventory heavy type profit.

- TeamGroup, ADATA (extremely inventory heavy), Apacer, Innodisk and Silicon Power had rapidly growing inventory but

原文：
I love how so many people requested $SIVE that Saxo Japan went out of their way to list it!

Just goes to show how much demand there is for this laser company.

Special shoutout to Saxo. And of course, excited for Sivers earnings + planned Nasdaq listing coming up.

翻译：
我很喜欢这么多人都在要 $SIVE，以至于 Saxo Japan 还特地把它上架了！

这正好说明了这家激光公司的需求有多大。

特别感谢 Saxo。当然，我也很期待 Sivers 接下来的财报，以及计划中的纳斯达克上市。

万美元
4 月：1.453 亿美元
5 月：1.403 亿美元
6 月：1.522 亿美元
7 月：2.128 亿美元

1–7 月整体：
约 8.79 亿美元营收
约 3.60 亿美元净利润

但问题往往在于可持续性：

肯定会有人说“这只是 7 月，看 TTM 才对”，但每个季度旧制程 DRAM 都在继续涨价，而且后面还有更多涨价（所以看 TTM 是不对的），不过每月 1 亿美元以上净利润也许只是异常值，等看其他月份财报再说。

在我看过的 Teamgroup / Apacer / ADATA / Innodisk / Etron 这类公司里，ESMT 似乎是赢家（而且不是单纯在倒库存），它确实具备一定定价权。

在晶圆分配和毛利显著提升的背景下，我觉得它是这几家公司里最强的一个，因为现金会持续流入资产负债表。

而且盈利和短缺是可持续的吗？

- “DDR4 / LPDDR4 的结构性供给缺口到 2028 年及以后仍然明显”——Winbond 2026 年 5 月投资者演示材料（这也支持进一步的连锁传导）

Winbond 过去也把 DDR4/DDR3 描述为存在持续的结构性供给缺口，且“短期内没有解决方案”

- TrendForce 说，无法或不愿锁定 DDR4 的 OEM 正在重新设计/降配到 DDR3，这又会迫使 DDR3 用户转向 DDR2

- 尽管 DDR2 现在很重要，但覆盖和参与者都不多……这点其实挺有意思的。

TLDR：

- 一家市值 25 亿美元的公司，资产 9.43 亿美元，净现金约 3.95 亿美元
- 7 月每月净利润 1.1 亿美元，年化 7 月 P/E 仅 1.9x
- 即便在熊市情景下，若盈利不可持续、月净利润回落到 5,500 万美元，P/E 也只有 3.8x
- Winbond 退出 DDR2 市场有利于业绩
- 和其他存储股一样，熊市担忧都在于盈利能否持续，但行业预计 2027 年供给还会进一步收紧。

*披露：我有持仓，非投资建议。

希望大家觉得有意思，我会继续跟踪 DDR4 / DDR3 / DDR2 的趋势，看后续怎么演化。

原文：
Thoughts on this 1.9x P/E memory company: (annualized off recent ER)

Spent a lot of time going through the memory stack for new $SNDK style names because cascading structural shortages from HBM/DDR5 into DDR4/DDR3/DDR2.

ESMT at $2.5B MC caught my interest.

Here's a note from my high-level thoughts/research on this name:

July's (1 month) earnings were:
- $212.8M revenue / $109.5M (net income, +- few M depending on FX)

Annualized that's: $1.314B net income/year off a $2.5B MC, so 1.9x July runrate P/E.

- Cash on hand: $455.7M (~$395.9M net cash)
- $943.5M assets
- $249.6M inventory, $146.3M receivables
- $596.0M book equity

They ~$359.6M net income in the first seven months of 2026, and it's accelerated to ~$109.5M a month net income.

What's even more interesting is Winbond is gradually withdrawing from DDR2 (June Trendforce report), which should give increased leverage to ESMT H2 (with price forecasts of DDR2 being 35-40% Q3)

What's not disclosed is wafer allocations from PSMC and possible risk is margin decrease due to upstream wafer hikes.

(You can think of an analogy like Kioxia: PSMC, ESMT: Sandisk, but without ownership)

PSMC did raise DRAM wafer-start prices about 45% (started in July, maybe shows up Nov/Dec) and ESMT said H2 margins could slip due to wafers having a higher cost.

So wanted to look at the bear case, and do some speculative napkin math wafer calculations based on their "2026 wafer capacity is roughly unchanged from 2025 and only satisfies ~60–70% of customer demand" claims

Napkin math:

ESMT did ~NT$13.5B revenue in 2024 at near 0 profitability, so monthly COGS was ~$30M, of which wafers might be $20M.

A 45% hike on that base gets you somewhere around ~$9M/month headwind.

For a harsher stress test, using Q2's ~31.3% COGS/revenue ratio as a proxy for July, July's $212.8M rev would imply roughly $66.6M of monthly COGS.

If

100% of that were exposed to the +45% DRAM wafer repricing: would imply roughly a $30M/month pretax headwind

-> net income would fall to $85M/month, or about 2.45× annualized P/E (which is still pretty impressive)

Fun thing is July's net income report was only because TWSE forced a supplemental disclosure after abnormal trading.

So, with Winbond winding down their segments, and DRAM est. going up Q/Q, signals point to ESMT maintaining similar profitability levels. There's DDR3 and others as main exposure, but DDR4 hikes trickles down the stack.

Revenue growth looks like:

Feb 2026: $65.6M
March: $95.1M
April: $145.3M
May: $140.3M
June: $152.2M
July: $212.8M

Jan–July as a whole:
~$879M revenue
~$360M net income

But the issue most of the times is durability:

There's going to be folks saying "oh it's just July, look at TTM" but each quarter legacy DRAM keeps getting hiked, and there's further hikes ahead (so TTM is the wrong way to go), but maybe $100m+ net income a month is an anomaly, we'll see what other earnings are.

Out of the Teamgroup/Apacer/ADATA/Innodisk/Etron type names I looked at that actually can maintain pricing power as well (and not just flipping inventory).

Felt like ESMT was the winner of the bunch (where cash just keeps flowing into balance sheet) given wafer allocations and an substantial increase in gross profit.

And durable are earnings + the shortage?

- “DDR4 / LPDDR4 structural supply gap remains evident through 2028 and onward" Winbond's May 2026 investor presentation (which supports further cascading)

Winbond had characterized DDR4/DDR3 as having ongoing structural supply gaps with "no near-term resolution"

- TrendForce says OEMs unable or unwilling to secure DDR4 are redesigning/down-specing into DDR3, which then forces DDR3 users toward DDR2

- There's not much coverage or players in DDR2 despite its importance now... which is

appreciated.

TLDR:

- $2.5B company with $943M in assets, ~$395M net cash
- $110M net income a month in July, 1.9x annualized July P/E
- Even in a bear case where earnings are unsustainable, reverting monthly net income back to $55m, would give it a 3.8x P/E.
- Winbond withdrawing from DDR2 markets helps earnings
- Same bear case like every other memory if earnings can persist, but industry sees further tightened supply in 2027.

*disclosure, have positions, NFA, not an investment recommendation.

Hope others find this interesting, I'll be tracking DDR4/DDR3/DDR2 trends to see how this plays out.

翻译：
对这家 1.9x P/E 的存储公司有什么看法：（按最近一次财报年化）

我花了很多时间梳理存储链条，想找新的 $SNDK 这类标的，因为 HBM/DDR5 向下传导，正引发 DDR4/DDR3/DDR2 的连锁结构性短缺。

ESMT，市值 25 亿美元，引起了我的兴趣。

下面是我对这个标的一些高层想法/研究笔记：

7 月（1 个月）的业绩为：
- 营收 2.128 亿美元 / 净利润 1.095 亿美元（因汇率有几百万美元浮动）

按年化算，相当于 12.14 亿美元净利润，对应 25 亿美元市值，也就是 7 月运行率 P/E 只有 1.9x。

- 现金：4.557 亿美元（净现金约 3.959 亿美元）
- 资产：9.435 亿美元
- 存货：2.496 亿美元，应收账款：1.463 亿美元
- 账面股东权益：5.96 亿美元

2026 年前七个月他们大约赚了 3.596 亿美元净利润，而且已经加速到每月约 1.095 亿美元净利润。

更有意思的是，Winbond 正在逐步退出 DDR2（6 月 TrendForce 报告），这应该会进一步增强 ESMT 下半年的议价能力（DDR2 价格预测在 Q3 还有 35-40% 上涨空间）

未披露的是来自 PSMC 的晶圆分配情况，潜在风险是上游晶圆涨价导致毛利率下滑。

（可以把它理解成一个类比：PSMC 对应 Kioxia，ESMT 对应 Sandisk，但没有股权关系）

PSMC 已经把 DRAM 晶圆起始价上调了约 45%（7 月开始，可能在 11 月/12 月体现出来），而 ESMT 也提到，由于晶圆成本更高，下半年毛利率可能会回落。

所以我也想看一下熊市情景，并基于他们“2026 年晶圆产能大致与 2025 年持平，且只能满足约 60–70% 客户需求”的说法，做一些很粗略的晶圆测算。

粗算如下：

ESMT 2024 年营收大约为 135 亿新台币，盈利几乎为零，因此月度 COGS 约 3,000 万美元，其中晶圆可能占 2,000 万美元。

在这个基础上上调 45%，大约会带来每月 900 万美元左右的压力。

更严格一点测试，如果用 Q2 约 31.3% 的 COGS/营收比作为 7 月的参考，7 月 2.128 亿美元营收意味着月度 COGS 大约 6,660 万美元。

如果这部分 100% 都暴露在 +45% 的 DRAM 晶圆重定价下：意味着每月税前压力大约 3,000 万美元

-> 净利润会降到每月 8,500 万美元，年化 P/E 约 2.45x（依然相当夸张）

有意思的是，7 月净利润披露是因为 TWSE 在异常交易后强制要求补充公告。

所以，随着 Winbond 收缩相关业务，加上 DRAM 估值环比上升，信号都指向 ESMT 还能维持相近的盈利水平。它主要暴露在 DDR3 等品类上，但 DDR4 的涨价也会向下传导到整个链条。

营收增长看起来像这样：

2026 年 2 月：6,560 万美元
3 月：9,510