SMH
Bull/Bear sentiment by day
SMH Mentioned Tweets
S
@ShanghaoJin
userId: 858124064476479500
3/28/2026, 1:11:38 AM
View on X >算力需求只会指数级上升,再无回头路了
智力通缩也只会越来越严重。靠知识、经验的“好学生”一旦失业,就再也没机会找到工作了
$SMHBull
L
@labubu_trader
userId: 54811865
3/25/2026, 4:32:46 PM
View on X >RT @jukan05: The fact that memory stocks are crashing because of Google’s Turboquant is a pretty good indicator of how many clueless people…
$SMHBear
L
@labubu_trader
userId: 54811865
3/25/2026, 4:12:52 PM
View on X >RT @labubu_trader: I started building AI inference infra since almost 10 years ago, and my take was memory is the biggest bottleneck in ter…
$SMHNeutral
L
@labubu_trader
userId: 54811865
3/25/2026, 4:04:42 PM
View on X >I'd treat this as the DeepSeek moment for memory names(SK Hynix, Samsung, Micron)
I checked the original paper: https://t.co/8LzkLgP5np, it's not surprising to me that new quantization method will get better and better. And since KV Cache is the bottleneck for lots of agentic use cases(long context window), there will be more and more optimization trying to compress the size and save the memory bandwidth.
However, reduce the KV cache by 10X that doesn't mean the economic demand to HBM/DDR will shrink in the future(Jevons paradox)
I already bought the dip when $MU had the 2nd leg intraday to test the LOD but didn't break down further.
$SMHBull
B
@Balder13946731
userId: 1814168337540436000
3/24/2026, 3:45:14 AM
View on X >伯恩斯坦预计马斯克的TeraFab的投资总额需要5-13万亿美元。
同时有人预计,哪怕以现在最快的速度购买 $ASML 的可以制造2nm 的极紫外光光刻机,也需要到2028年才能装配完毕。
然后流片测试和首批次生产至少要到2030年。
而马斯克计划2030年每个月生产100万片晶圆,超过目前甚至2030年所有已知所有芯片厂扩产计划的产能总和。
这饼太大,恐怕一般人吃不下去啊😧
马斯克在2020年特斯拉电池日放话到2030年实现3TWh的电池产能,而如今约为0.08 TWh
$SMHNeutral
L
@labubu_trader
userId: 54811865
3/23/2026, 5:25:16 PM
View on X >I think we are entering into a super choppy period overwhelmed by the headlines similar to March/April last year.
From my experience of last year, chasing long or short in this environment will be super painful as no clear trends.
That's why I chose this "buy when price level hits and hold" strategy and use positioning to control the risk.
At the opening today I already cut the postion from around 35-40% to roughly 25% as planed.
But I iterated the portfolio a little bit compared to the original post:
Now I'm holding MU/SNDK/LITE/TSEM shares and AMD/INTC leap calls(2027 March 0.75 delta) instead of TQQQ.
Also holding CVX/VG 3months calls for hedging purpose.
$SMHBull
B
@Balder13946731
userId: 1814168337540436000
3/23/2026, 12:33:04 AM
View on X >Signals are mixed.
感觉就像回到了周五收盘的原点。原油、VIX和国债,没有很强烈的空头信号。
但是下跌指示中比较强烈的信号是动能板块表现疲软,例如存储、芯片制造和加密货币等都表现疲弱。
$SMHBear
L
@labubu_trader
userId: 54811865
3/21/2026, 3:04:36 AM
View on X >Heard from some folks from China at GTC: CXMT will expand capacity by 100% and YMTC by 50% in 18 months, as required by the Chinese government for strategic supply chain security.
CC @jukan05
$SMHNeutral
B
@Balder13946731
userId: 1814168337540436000
3/17/2026, 5:31:13 PM
View on X >和传闻不同, $MU 也将供应Rubin的HBM4而不是由三星海力司士独家供应。
$SMHBull
B
@Balder13946731
userId: 1814168337540436000
3/16/2026, 3:04:49 AM
View on X >川普需要这次与中国的会面,大豆的出口对本次中期选举影响巨大。
赶紧把GPU的出口放出来。
$SMHBull
L
@labubu_trader
userId: 54811865
3/15/2026, 1:50:57 PM
View on X >RT @zephyr_z9: I hope it's a new x86
Would be so cool
EMIB partnership is expected and boring
$SMHNeutral
S
@ShanghaoJin
userId: 858124064476479500
3/14/2026, 10:45:17 AM
View on X >光趋势性从scale-out,进而scale-up更靠近计算核心。是叙事极强的AI基建结构性大转向
Hock Tan说的不是“光”不行,而是“铜”能再战几年
相信光本没问题,唯一毛病是预先定价~~
“贵”才是原罪
如果接下来有不那么贵的机会,别犹豫
记住我的2个规则:
1. 参与算力提升
2. 迭代收入增速超过成本增速
$SMHBull
B
@Balder13946731
userId: 1814168337540436000
3/14/2026, 1:33:49 AM
View on X >商务部撤销了原本要额外增加的AI芯片出口规定。
$SMHBull
L
@labubu_trader
userId: 54811865
3/12/2026, 12:38:40 PM
View on X >RT @dnystedt: War in Middle East impact on Taiwan Semiconductors
No shortages of Natural Gas or Helium for Now
Natural Gas:
Natural gas…
$SMHNeutral
L
@labubu_trader
userId: 54811865
3/11/2026, 7:01:44 AM
View on X >RT @TacticzH: Incredibly proud of this one!
I've heard and read a lot about photonics lately, but I finally decided to dive in myself.
It…
$SMHNeutral
B
@Balder13946731
userId: 1814168337540436000
3/10/2026, 4:13:17 AM
View on X >长江存储虽然目前地位不如三星海力信,但是在专利和创新上却远远超越了二者,这可能在未来帮助中国存储产业取得主导地位。
中国击败韩国取得主导地位的产业,另一个例子就是造船业。造船业的今天可能是存储的明天。
$SMHNeutral
S
@ShanghaoJin
userId: 858124064476479500
3/10/2026, 2:58:38 AM
View on X >给你们一个抄底、选票的思路
AI 是工业革命级别的。不是每个票价格跌到位就会反弹的,证伪“负叙事”是很难的
在技术革命下证明自己只有收入增速加快,收入增速还要超过Capex增速。这种在叙事的,跌了就别太计较估值了
最容易完成这艰巨任务的是,通过供算力中心,需要产品迭代收入增加的半导体
$SMHBull
B
@Balder13946731
userId: 1814168337540436000
3/9/2026, 10:43:01 PM
View on X >OpenClaw创始人惊讶,中国居然出现了OpenClaw硬件商店..
(不过疑似仅仅只是AI) https://t.co/gqJkMZzDfa
$SMHNeutral
L
@labubu_trader
userId: 54811865
3/5/2026, 9:32:26 PM
View on X >Yes it was indeed a negative catalyst for optical names. 😭
Had big loss in optical names today and SMH puts didn’t cover much.
$SMHBear
S
@ShanghaoJin
userId: 858124064476479500
3/5/2026, 10:40:24 AM
View on X >价值在贡献算力:
算法>硬件
rack协同> AP > Die
别N5 N3 N2了,以后这些都不会缺的,N5过剩就会你们预料快得多
黑帮老大搓圆仔汤,分的是AP线程,AP叙事都不好杀
配件属于风口来回换
ps 自己胡言乱语的小笔记
$SMHBear
S
@ShanghaoJin
userId: 858124064476479500
3/3/2026, 3:23:14 PM
View on X >RT @ShanghaoJin: @frankyluan 腰斩优先买光模块、其次存储
不腰斩买intc和nv
$SMHBull
B
@Balder13946731
userId: 1814168337540436000
3/3/2026, 5:48:43 AM
View on X >河南人造钻石产业已从装饰品转化为数据中心不可或缺的散热核心部件,未来可能是一个和存储芯片一样赚钱的行业,但同时也是极度高耗能的产业。
$SMHNeutral
L
@labubu_trader
userId: 54811865
2/25/2026, 2:25:07 AM
View on X >Fin was my coworker and a strong SoC architect. Here is the English translation of his view on using SRAM in LLM training and inference, as the twitter built in translation seems to be not working:
Last year I spoke with MatX’s chief architect. I told him that the weakness of the SRAM approach is that it’s not scalable. Especially given that HBM speeds are increasing exponentially with each generation, it’s hard for SRAM’s speed advantage to remain sustainable over the long term. I asked him whether MatX’s architecture is scalable.
He thought for a moment and then confidently told me that it is scalable.
Later, I looked into it and found out that they place the weights in SRAM and the KV cache in HBM.
After thinking it through, it actually makes sense in light of the future agent trend: multi-turn conversations mean long contexts, and the KV cache and context size will explode, while batching is not that high. If you put that part in SRAM like Groq does, it wouldn’t be sufficient; placing it in HBM is a more reasonable choice.
Using SRAM to hold the weights is essentially trading cost for speed, but compared to Groq or Cerebras, it’s more scalable—especially for agent-oriented workflows
$SMHNeutral
L
@labubu_trader
userId: 54811865
2/25/2026, 1:16:14 AM
View on X >RT @insane_analyst: Memory stocks should not be viewed on P/E basis.
Literally go to 2023 $MU earnings calls where Sanjay was publicly beg…
$SMHNeutral
B
@Balder13946731
userId: 1814168337540436000
2/24/2026, 4:34:04 AM
View on X >Citrini 认为大盘涨到8000点,美国经济就会陷入巨大衰退,由于AI过度生产以及分配不均。
目前大量经济指标都指向AI可能带来滞涨,但我觉得这种结构性问题需要一个引爆点。否则目前的依然有缓冲空间。
那就是Data Center带来的巨量基础设施投资依然支撑着美国的消费和就业。只有以下事情全部兑现,Citrini的预言才会实现:
1. 由于中国大量廉价模型的冲击,美国的大模型始终无法达到足以覆盖数据中心建设成本的收入(2030年达到2万亿)
2. 由于算力过剩,美国云计算公司开始减记折旧损失,并且引爆了循环融资的泡沫,导致云计算公司、GPU制造商、AI初创公司价值全面暴跌。
3. 由于数据中心建设带动的就业潮消失,体力劳动以及脑力劳动的双重失业潮叠加,经济陷入消费不足的通缩萧条。
$SMHNeutral
B
@Balder13946731
userId: 1814168337540436000
2/23/2026, 12:51:17 AM
View on X >好的我总结一下下周我要干什么:
我们要做多金属
侧击半导体
大盘可能持续上涨,继续抄底云果软
韩国股市也可以继续搞
$SMHBull
B
@Balder13946731
userId: 1814168337540436000
2/22/2026, 5:00:59 PM
View on X >雷声大雨点小,我们要看内存期货的价格。
$SMHNeutral
B
@Balder13946731
userId: 1814168337540436000
2/22/2026, 4:10:57 AM
View on X >这是一个很严重的问题
H100老显卡的价格目前来说,其实是在13000美元左右,大约是全新的一半。虽然没有这位说的那么夸张,但是很多因素,每年硬件的折旧依然有50%
这件事情在 $NVDA 全面铺开B系列的销售,以及电能价格进一步上涨时就会全面在市场发酵
因为届时仍然让这些老旧芯片多运行1秒,都是在亏钱
这是一个2月前的故事,却不是鬼故事,而是三大云厂商迟早要说的真故事
$SMHBear
B
@Balder13946731
userId: 1814168337540436000
2/21/2026, 8:08:14 PM
View on X >台湾的台积电、韩国的三星和海力士会摧毁美国互联网时代以来积累的财富。
$SMHBear
L
@labubu_trader
userId: 54811865
2/19/2026, 3:45:32 AM
View on X >I think $MSFT will likely be the first one to start the “Capex Recession” and bring the plague to $SMH. Why?
1. MSFT was the most reluctant to join the capex arms race and had multiple quarters of Azure growth misses due to “supply constraints.”
2. In podcasts, Satya has expressed confusion about why the market focuses so much on Azure growth instead of their progress in AI applications.
3. He also said MSFT should focus on the high-margin software business rather than the low-margin GPU rental business.
4. Satya is not the founder, so he doesn’t have the guts to ignore angry shareholders and steer the company however he wants. He can’t bear the failure of screwing up a $3T company, which would ruin his career and life.
5. Even in the AI cloud business, MSFT has no cost advantage over AWS and GCP.
6. $MSFT stock took the most damage recently alongside $ORCL
In summary, I think $MSFT will very likely be the first to quit this arms race and embrace AI modeling and software business.
When? I think once $MSFT dips below $340 (the level when the overall bull market started in 10/2023) and stays below it, the next earnings report will be when he announces this.
$SMHBear
B
@Balder13946731
userId: 1814168337540436000
2/18/2026, 5:52:26 PM
View on X >倒反天罡。不过确实目前的趋势是,对于数据中心,GPU似乎没有HBM重要。有HBM你就可以开始赚钱了,毕竟开源模型那么多。
$SMHNeutral
S
@ShanghaoJin
userId: 858124064476479500
2/18/2026, 12:06:30 AM
View on X >This is exactly the situation right now:
GPU has no real substitutes
$SMHBull
B
@Balder13946731
userId: 1814168337540436000
2/17/2026, 6:30:07 AM
View on X >有诸多技术上的巧合可能会导致中国开辟出一条前所未有的发展道路。
那就是通过出口智能来出口能源。
其实如果放在10年前,这可能是无稽之谈,因为美国盛产化石能源,中国还要进口。然而随着技术的进步,尤其是太阳能和风能的发展,目前美国每一度电的生产成本是0.1美元,而中国由于规模化的生产以及西部的开发,能源成本为0.03美元1度。重要的是,到2030年,美国将没有足够的电能支撑预计算力中心能源消耗的一半,而中国将额外增产三倍于全球预计AI需求的3倍电力。
这些过剩的能源就会变成为一串串的Token毫无关税的流向全世界,现在中国只需要无脑疯狂扩产发电设施,就有可能垄断全球智能市场。
当然芯片依然是这一构想的最重要一块拼图。
$SMHBull
B
@Balder13946731
userId: 1814168337540436000
2/17/2026, 5:02:35 AM
View on X >有人觉得随着技术和硬件的进步,人工智能的耗能会越来越少,我觉得不是。人类的大脑是人体最耗能的器官,大约2%的体重,消耗着25%的能量。智能天然要消耗很多能源。
产生每个Token的耗能在减少,但是产生更高的智能却需要消耗着更多Token。而目前,Token消耗的速度远远高于GPU的效能提升。
$SMHNeutral
L
@labubu_trader
userId: 54811865
2/16/2026, 4:59:29 PM
View on X >My current holding in the trading accounts:
Long: XLU/XLI/XLE, CVX/XOM/RIG, CAT/DE/ADM, Gold, Semi caps, ORCL/IGV(closed igv/pltr puts when igv hit $80 last week), LLY, ULTA, EWU/EWG/EWJ/EWZ
Short: QQQ/ARKK, COIN/MSTR
$SMHBull
B
@Balder13946731
userId: 1814168337540436000
2/15/2026, 3:52:06 PM
View on X >Elon Musk总结了3个未来美国的AI可能被中国完爆的几个原因:
1. 美国电力不够,美国的发电量到2030年都会和现在持平。
2. 芯片制程和能力提升的性价比正在遭遇物理极限和瓶颈,在3 nm以后,芯片制程已经没有真实提升,每一点名义上的提升都需要巨大成本。
3. 美国缺少足够的AI工程和科学研究人才。
$SMHBear
S
@ShanghaoJin
userId: 858124064476479500
2/13/2026, 2:52:29 AM
View on X >一行升百业落,有“算力”统治
算力中制成只保护下限,封装才决定上线
TSMC不管N2 N3都不缺die,瓶颈在cowos分货。其实AI fabless复杂需求各个不同,驱向前道封装。高度定制化前道封装形成的是TSM与NV双向壁垒
亦不能忽视系统级封装壁垒。NV在封装到tray/rack、甚至cluster踩的坑全成为追赶者的moat
$SMHNeutral
S
@ShanghaoJin
userId: 858124064476479500
2/11/2026, 7:03:16 AM
View on X >这个跟我们团队看法很一致:存储现货高点谨慎投存储资产
不太敢追高因为虽然数据中心HBM一定缺货,但是随着电子消费端缓解,“痛感”消失。过程中,市场叙事会无厘头地逆转。这就好像现在买不到货,所以无厘头地抢股票买一样~
另外MU只是慢,小作文有点过了
另外之另外,存储周期性强要看PS而不是PE
$SMHNeutral
B
@Balder13946731
userId: 1814168337540436000
2/10/2026, 5:14:53 PM
View on X >全球存储产业陷入一片疑云。可以确定的是CMXT并不会制造HBM。所以不会制造HBM的产商可能会面临压力。
$SMHNeutral
L
@labubu_trader
userId: 54811865
2/10/2026, 1:46:44 AM
View on X >Holding many semicap names given they are less crowded than storage/optical names I guess.
$SMHBull
L
@labubu_trader
userId: 54811865
2/9/2026, 5:45:14 AM
View on X >This sounds way more reasonable than SemiAnalysis’s complete rule-out of $MU in the HBM4 supply.
$SMHBull
B
@Balder13946731
userId: 1814168337540436000
2/9/2026, 3:35:03 AM
View on X >我觉得 $TSM 可能会进一步上涨,原因就在于高市早苗提出的协助防御台湾的观点,这可能被认为可以加强对中国的威慑和保护台积电在日本的投资。
但是日本这个经济体依然问题重重,亚洲各种货币都在升值,唯独日元疯狂贬值,贬值就算了,贸易逆差还在扩大,这是境内产业能力完全衰退,在东亚怪物圈站不住脚的表现。
$SMHNeutral
S
@ShanghaoJin
userId: 858124064476479500
2/7/2026, 2:45:01 AM
View on X >其实为什么capex被市场惩罚都不会有人退缩
模型端是春秋五霸战国七雄
背后是“圈地运动”,在玩大富翁。现在AI早期,买到GPU等于买到地,停止投入等于出局
一个取代7亿白领20%,几万亿的市场,现在丢这点钱当门票都值
$SMHBull
L
@labubu_trader
userId: 54811865
2/6/2026, 3:26:43 PM
View on X >Big win today.
Semi's narrative got strongly reinforced in the MAG7 ER. It's acting like a safe haven now, despite big sell pressure this week.
I'd expect this sector to get even more crowded than before until NVDA ER. I will definitely take profits aggressively before it to avoid the risk.
$SMHBull
B
@Balder13946731
userId: 1814168337540436000
2/5/2026, 10:52:31 PM
View on X >苹果的商业模式彻底赢麻:
当微软和谷歌在为每块 H100 显卡打破头、每年烧掉 1000 亿美元搞基建时,苹果确实不仅没有加入这场疯狂的“绞肉机”,反而坐在岸边等着收过路费。
苹果之所以能做到“看起来一毛钱不花”(实际上是花小钱办大事),是因为它掌握了 AI 产业链上最稀缺的资源:用户入口。
以下是苹果这套“空手套白狼”战略的底层逻辑:
1. “不仅不付钱,还要你倒贴”的霸王条款
你可能听说了,苹果在集成 ChatGPT 时,据报道并没有付给 OpenAI 一分钱现金。
• 逻辑是这样的: 苹果对 OpenAI 说:“我拥有全球 22 亿活跃设备(iPhone, iPad, Mac)。我把你的 ChatGPT 放到 Siri 里,这是世界上最顶级的流量入口。这一带来的品牌曝光和潜在订阅用户,价值几十亿。所以,我不收你‘上架费’就不错了,你还想让我付钱?”
• 结果: OpenAI 承担了昂贵的云端推理成本(每一次问答都要烧钱),而苹果免费获得了最先进的 AI 能力。
• 未来: 苹果还会引入 Google Gemini、Claude (Anthropic) 等模型。一旦这些大模型变成**“大宗商品”**(即大家的智商都差不多),苹果就可以让它们竞价上岗,甚至向它们收取“过路费”(就像 App Store 抽成 30% 那样)。
2. 把“算力成本”转嫁给你(消费者)
这是苹果最核心的技术护城河——端侧计算 (On-Device AI)。
• 微软/谷歌的模式: 你问一个问题 -> 数据传到云端 -> 巨大的服务器运转 -> 消耗公司的电费和显卡 -> 传回结果。每一句话都在烧公司的钱。
• 苹果的模式: 苹果在 iPhone 里塞进了强大的 NPU (神经网络引擎)。当你在手机上用“苹果智能”处理邮件、修图、总结短信时,主要消耗的是你手机的电量和你买手机时付的硬件钱。
• 精髓: 苹果巧妙地把本来属于公司的OpEx(运营成本),变成了用户购买手机时的硬件成本。用户自己买了算力(iPhone),还要自己付电费来运行 AI。苹果的边际成本接近于零。
3. “隐私”是最好的省钱借口
苹果一直高举“隐私”大旗,这在 AI 时代成了绝佳的战略武器。
• 苹果宣称:为了保护你的隐私,大多数 AI 任务应该在手机本地完成,不要上传云端。
• 潜台词: “既然在本地完成了,我就不需要建那么多昂贵的数据中心了。”
• 只有遇到本地搞不定的复杂问题(比如写长篇论文),它才会调用“私有云”或 ChatGPT。这种**“分级处理”**策略,极大降低了对后台算力的需求。
4. 坐享其成的“后发优势”
在科技圈,First Mover(先行者)往往是先烈的代名词,而苹果一直是Smart Follower(聪明的追随者)。
• 试错成本: 微软和谷歌花费数百亿去探索什么才是好的 AI 产品,去教育市场,去踩坑(比如 AI 产生幻觉、说错话)。
• 苹果的摘果子: 苹果静静地看着。等技术路线成熟了,用户习惯养成了,成本降下来了,它再把最成熟的技术整合进 iOS,包装成一个流畅、好用的功能(而不是一个极客的聊天机器人)。
• 它不需要做“最强”的模型,它只需要做“最好用”的体验。
5. 商业模式的降维打击
• Google/Meta 的死穴: 它们主要靠广告赚钱。AI 给出的答案太直接,用户就不点链接、不看广告了。所以 AI 对它们现有的商业模式是破坏性的。
• 苹果的顺风车: 苹果靠卖硬件赚钱。AI 越强,你需要换更强的 iPhone 16/17 来运行它。
• AI 不是苹果的产品,AI 是促进 iPhone 销量的燃料。
• 所以,苹果不需要靠 AI 本身赚钱,只要你为了 AI 去换新手机,苹果就已经赢麻了。
总结
苹果之所以能“赢”,是因为它不玩**“谁的模型参数大”的游戏,而是玩“谁拥有用户”**的游戏。
当微软和谷歌在泥潭里为了算力打得头破血流、利润率被 CapEx 拖垮时,苹果站在高地上,手里拿着通往用户的钥匙,对它们说:
“你们谁打赢了,就上来给我打工吧。”
$SMHNeutral
S
@ShanghaoJin
userId: 858124064476479500
2/5/2026, 9:46:18 PM
View on X >1. 指数跌的时候大家会无视叙事
2. Goog未来负叙事将是用TPU比NV在大集群训练劣势
当B系列卡的模型出来的时候,发现Genimi也就这样
接着会讨论TPU performance/w的问题,不适合军备竞赛
这时候+ Goog - OAI trade可能平仓
我不是Fud,自己大量极低成本的Goog储蓄从来不卖。持有无所谓这些周期的
$SMHNeutral
L
@labubu_trader
userId: 54811865
2/5/2026, 5:29:21 PM
View on X >Gladly, I sold all of my semi puts yesterday.
Semiconductors are the strongest sector in the market today after Google's explosive capex increase announced in yesterday's earnings report.
I'm starting to add back my semi positions and only added the ones still performing strongly today. Will add more if NQ slides to 24000: https://t.co/Y96cv5RiVk
I don't see any fundamental or macro weakness for semis. Will hold until NVDA's earnings to cut.
NFA.
$SMHBull
B
@Balder13946731
userId: 1814168337540436000
2/5/2026, 4:04:24 PM
View on X >而且今年的Capex是可以预计到一些事情的。
那就是半导体公司、Memory、芯片制造公司,业绩一定会继续保持强势。
所以目前的低价是机会。这和去年一样。去年Deepseek引起大崩盘,但是这并不能阻止这些大公司拼了命的砸钱,让这些半导体公司赚得盆满钵满。
$SMHBull
S
@ShanghaoJin
userId: 858124064476479500
2/4/2026, 7:14:53 PM
View on X >不乌鸦嘴了~
两种交易仓位你们考虑觉得会不会平仓
1. long Semi short software
2. long goog short OAI
然后,关注vix,因为systemic position (vol control + cta)其实挺满的
“抗打”的下限高,看到其票稀里哗啦破位,有时候也挺爽的
btw 我是死多头,等goog“拯救”大家
$SMHBull
S
@ShanghaoJin
userId: 858124064476479500
2/4/2026, 8:53:18 AM
View on X >股市叙事太割裂了
软件砸地板上,却还质疑AI真实性
存储、光模块涨上天,还不相信NV需求
既不相信基础算力供不应求、又疯炒配件
等来轮调整,让那些按照28年盈利估值的跌个70%+,市场就健康了
$SMHBear
B
@Balder13946731
userId: 1814168337540436000
2/3/2026, 7:17:41 AM
View on X >The physics of space data centers simply doesn't add up.
Thermodynamics: Vacuum is a thermos. Cooling 1GW of compute in space requires ~1 sq km of radiators.
Bandwidth: AI infra relies on TB/s interconnects. You can't replace NVLink with RF/Laser links for distributed training.
Radiation: Modern 3nm/5nm GPUs fry instantly in space. Rad-hard chips are decades behind in compute density.
Musk isn't a wizard. Reusable rockets and Starlink obey physics; current Space DC concepts do not.
#SpaceX #xAI
$SMHNeutral
S
@ShanghaoJin
userId: 858124064476479500
2/2/2026, 4:04:28 AM
View on X >20bil ATM发股票!
算力短缺共识如此确定之际,其实最大风险反而是几个CapEx巨头被CDS、财务问题挤压,被迫减少CapEX
这样单个公司会让出GPU份额永远掉队,行业也会被短期质疑下跌
Orcl要all in“看看究竟谁是真傻逼”。反正有卡份额不买必死无疑,冲不过去就早死早超生呗 https://t.co/JwyAJEhuTz
$SMHBull
S
@ShanghaoJin
userId: 858124064476479500
2/1/2026, 3:50:11 AM
View on X >币圈愁云惨淡的哭喊啥?弄得好像跌了多少的样子
去看看我跑掉的INTC和卖飞的AAOI,再看看接下去几周的存储、金银。这都哪跟哪儿啊?
其实华尔街不怕你跌,怕你不“性感”了。TSLA年初跌起来那么猛,没见那么多人去Fud呀
$SMHNeutral
B
@Balder13946731
userId: 1814168337540436000
2/1/2026, 2:58:36 AM
View on X >Can China disrupt RAM industry?
$SMHNeutral
S
@ShanghaoJin
userId: 858124064476479500
1/31/2026, 12:51:45 PM
View on X >他们担心OAI没拿到NV钱
其实比左脚踩右脚,华尔街更担心的是你们片都不想骗我了
不用太担心,这些都是真的,token需求、算力芯片供应长期严重紧缺
$SMHBull
B
@Balder13946731
userId: 1814168337540436000
1/28/2026, 7:28:25 AM
View on X >明天开始只做能源和半导体,Bet the Capex!
$SMHBull
S
@ShanghaoJin
userId: 858124064476479500
1/28/2026, 3:45:35 AM
View on X >涨吧涨吧,大家都不把消费电子市场当回事
内存卖不到,CPU也涨价,大荒年大家一起死
厉害国的EV车被两头挤压,产能过剩、margin归零
为什么“厉害”总能把把压住“厉害”到最“厉害”的行业?
$SMHBear
S
@ShanghaoJin
userId: 858124064476479500
1/26/2026, 12:50:41 PM
View on X >没错,消费电子是荒年~~囤货吞的手机、车都买不到,严重影响出货
$SMHBear
M
@mat78704
userId: 1805786844543819800
1/25/2026, 6:49:18 AM
View on X >Why do I foresee that $smh will fly really high into Jan 30? Am I hallucinating?
$SMHBull
L
@labubu_trader
userId: 54811865
1/25/2026, 12:21:30 AM
View on X >Master Fin’s insights about neoclouds are remarkable.
As the memory/gpu supply get tighter and more expensive. Token consumption by AI agent is skyrocketing. Also given Mr T’s affordable electricity policy. Neoclouds’ abundant computation and electricity capacity will be more and more valuable and they will be more and more scarce this year.
$SMHBull
S
@ShanghaoJin
userId: 858124064476479500
1/23/2026, 11:54:39 PM
View on X >还觉得我卖INTC去买大票MSFT、NV太土了吗?
投资别总想着冲,要胆子小,逃顶了要“后怕”。我现在就有点后怕
晶圆厂提产能、提良率之后才能提margin。不能overnight捧天上,半导体行业十年吃💩才能一朝吃肉
不过泼完冷水再奶下INTC。“长坡厚雪”—5年后你们会看到半导体不同的格局,他将是重要一环
$SMHNeutral
B
@Balder13946731
userId: 1814168337540436000
1/23/2026, 2:15:48 AM
View on X >It makes perfect sense. $MU also trades at a huge premium compared to Samsung and SK Hynix. This is a clear case of 'American Exceptionalism' in the markets—US-listed assets simply command higher valuation multiples.
$SMHNeutral
M
@mat78704
userId: 1805786844543819800
1/22/2026, 11:57:57 PM
View on X >明日可以去找semi里面找哪些tickers不跌。这些可能就是要涨的。可以watch起来。但我自己明天大概率不会做多。
$SMHBull
L
@labubu_trader
userId: 54811865
1/22/2026, 11:50:52 PM
View on X >What concerned me about $INTC ER is semi is currently consensus long and super crowded. We might see tomorrow:
1. some cascading selloff in other semi names as funds sell the basket
2. Long semi short software unwind
If this occurs tomorrow, I'll use this opportunity to BTD
$SMHBull
L
@labubu_trader
userId: 54811865
1/22/2026, 3:19:54 AM
View on X >TACO as planed, the 2nd target was never hit so I had to close the all shorts around the 1st target today. Full long already with bunch of industrial medal, memory/storage, semi equipment, optical, cpu names, rty/iwm. Considering adding a few software back such as MSFT and DDOG.
$SMHBull
S
@ShanghaoJin
userId: 858124064476479500
1/22/2026, 12:30:21 AM
View on X >ARM还没到他,但是好日子会来的
btw,特别最近涨的多的,包括存储,季报最好小心点,upside surprise难度都挺大的
市场小作文满天飞,买方预期都打满了
$SMHBear
B
@Balder13946731
userId: 1814168337540436000
1/22/2026, 12:15:21 AM
View on X >Breakout Alert: Is $AMD the Next $MU? 🚀
For months, both $AMD and $MU were coiled in the $200-$250 consolidation zone. Then, $MU exploded, fueled by insatiable AI memory demand, shattering resistance and launching into a powerful new trend. Now, with the CPU & AI revolution hitting hyperdrive, $AMD is primed to ignite. The chart shows a massive potential 80%+ upside as it follows Micron's breakout path. Don't miss the next big move in semi's! 🔥📈
$SMHBull
B
@Balder13946731
userId: 1814168337540436000
1/20/2026, 4:55:16 AM
View on X >半导体制造和存储 is unbeatable
$SMHBull
B
@Balder13946731
userId: 1814168337540436000
1/19/2026, 7:33:10 AM
View on X >由于恶劣天气影响,阿勒泰硅矿石减产,导致工业硅大厂可能在近期进行停产。可能会影响全球半导体的生产。
$SMHBear
B
@Balder13946731
userId: 1814168337540436000
1/17/2026, 1:37:31 AM
View on X >2026 Alpha Thesis: Hardware Maturation + Supply Shock. 🚀⚡️
The "Promise Phase" is over. The "Deployment Phase" is here.
1️⃣ Rocket ( $RKLB, $FLY , $ASTS , $RDW): SpaceX IPO rumors = sector re-rate. Neutron launch is the value unlock.
2️⃣ CPU ( $AMD, $INTC): Cloud is crowded; the "Edge AI" PC supercycle is the volume trade.
3️⃣ DRAM ( $MU $SNDK $AMAT $ASML $TSM $LRCX): HBM is a supply black hole. Standard memory shortage = pricing power.
4️⃣ Energy ( $VST, $CEG, $BE): No power = No AI. Nuclear baseload is the new gold.
The trade: Long scarcity & physical infra. 📈
#StockMarket #Investing #AI #Space
$SMHBull
B
@Balder13946731
userId: 1814168337540436000
1/16/2026, 10:00:39 PM
View on X >我认为现在开始必须要专注于半导体和能源板块。
It will be a party!
$SMHBull
M
@mat78704
userId: 1805786844543819800
1/16/2026, 4:58:26 AM
View on X >这个波段将于明日完美收官。Jan 2带着粉子们抄了spy 680 darkpool level的底。然后在上周三,周五带着大家止盈。周日再次提示周一可以买入。今天周四在下跌前提示不要追高半导体。我个人账户这个波段收入不错,所以我不贪,call全部平仓,等待下一个机会的到来。
$SMHNeutral
M
@mat78704
userId: 1805786844543819800
1/15/2026, 8:09:18 PM
View on X >刚才跌之前没减仓的,明早建议把semi,大科技,科技动能股,数字货币miners都扔了。
$SMHBear
B
@Balder13946731
userId: 1814168337540436000
1/15/2026, 3:34:08 AM
View on X >有人认为Elon是一个白人至上主义者和社会达尔文主义者。
如果Optimus真的改变了人类的生产力,这样的Elon真的会让大部分人躺平吗? https://t.co/wFbqEq2Opw
$SMHNeutral
B
@Balder13946731
userId: 1814168337540436000
1/14/2026, 8:28:59 PM
View on X >不是很懂,这是对芯片取25%的出口税?可是这可能虚弱美国的芯片竞争力。
$SMHBear