20250609

399 posts

20250609 banner
20250609

20250609

@explore250

開始在這邊發瘋

Katılım Şubat 2020
791 Takip Edilen46 Takipçiler
20250609 retweetledi
墓碑科技
墓碑科技@mubeitech·
人年轻的时候都是左派。 直到他们准备买套房,或者自己做点小生意。 这是澳大利亚女演员 Holly Valance 最近在采访里说的一句大实话。 她说: "每个人刚开始都是左派,然后总在某个时刻醒悟过来。 等他们开始赚钱、工作、试着经营生意、或者买房的时候。 你就会突然发现,那些左倾的漂亮想法,到底有多烂。" 这话为什么能让无数人共鸣? 因为左翼的叙事,太容易让人产生道德优越感了。 免费福利、给富人加税、消灭不平等。 二十岁的时候如果不信这些,你可能会觉得自己没有同情心。 但现实是,情怀是免费的,生活却是有价格标签的。 当你真正进入社会,拿到第一张工资单,看着被政府生生抽走的一大块税金,你开始肉疼。 当你省吃俭用想买套房,却发现高福利背后是印钞机带来的高通胀,房价被推到了天际。 当你试着开个小店,发现层出不穷的政府监管和劳动法限制,压得你连气都喘不过来。 这时候,你终于醒了。 你发现那些天天把"大爱"挂在嘴边的政客,其实是在用你辛苦工作的血汗税金,去供养一套低效的官僚系统和不愿意工作的人。 一个人从左派变成保守派,根本不需要去读什么高深的政治经济学。 去税务局报一次税,或者去银行申请一次贷款,生活自然会给你上一课。
中文
63
531
2K
106.3K
20250609 retweetledi
Mark Minervini
Mark Minervini@markminervini·
Everyone seems to have all the answers, but few have the money and success. Many years ago, I woke up one morning; I looked into the mirror and I ask myself a very important question: if you're so smart, why aren't you rich? That was the day I started shutting my mouth and listening carefully to those who had more success than me. Because the truth was - at that point - I wasn't that smart yet. But I had to get my ego out of the equation. Then my life changed profoundly.
English
101
271
2.9K
164.3K
20250609 retweetledi
Electra
Electra@Electrarythm·
There is a certain type of person everywhere now, especially online. He consumes endless information every day: philosophy, psychology, productivity, spirituality, neuroscience, business, self-improvement, history. He knows a little about everything and deeply experiences almost nothing. His entire identity becomes built around understanding instead of living. He watches videos about confidence instead of speaking confidently. Reads about discipline instead of becoming disciplined. Studies relationships instead of learning how to love. Consumes motivational content instead of taking action. He feels intelligent because he is constantly mentally stimulated. But stimulation is not transformation. Most of the time, knowledge becomes emotional protection. Reality is unpredictable. Reality humiliates. Reality exposes weakness. Books and ideas do not. Inside information, he can continue imagining himself as intelligent, deep, insightful, different from ordinary people. So he remains trapped in preparation. He constantly feels as if he is "becoming" someone, while his real life remains strangely untouched. He develops sophisticated language for problems he never confronts directly. He can explain human behavior beautifully while being unable to handle ordinary discomfort, rejection, uncertainty, loneliness, or risk. He slowly turns life into observation instead of participation. The internet rewards this personality heavily. He receives validation for sounding aware rather than becoming capable. Eventually, he begins confusing self-analysis with growth and information with wisdom. But beneath the intelligence usually exists the same thing: fear. Fear of failure. Fear of embarrassment. Fear of reality answering back. Because action destroys fantasy. The moment he truly acts, he can no longer hide inside potential.
Electra tweet media
Tej Dosa@ComedicBizman

The fastest way to turn into a NPC is to fill every moment of stillness with audio books, podcasts, CEO interviews, tweets, threads, and YouTube videos. The fastest way to turn into the Main Character is to spend more time in stillness and give yourself 4 hours to create.

English
548
4.1K
29.6K
3.9M
20250609 retweetledi
Mark Minervini
Mark Minervini@markminervini·
One of the most important things you can do as a stock trader is to be patient with yourself and allow yourself time to grow. Not all flowers bloom at the same time, and not every journey unfolds on the same schedule. It may take you longer than someone else to develop the skills, discipline, and understanding needed to succeed, but that doesn't mean you're any less capable. Avoid measuring your progress against others. Everyone learns at a different pace and reaches important milestones at different points in life. Trading is a personal journey, and your timeline is your own. I know this firsthand because I was a very slow starter. It took me six years before I even became profitable. There were many times when I could have concluded that I simply didn't have what it takes. But persistence, patience, and a commitment to continual improvement made all the difference. Most of all, I knew that those who succeeded were just men like me, and if they did it, then so could I. So give yourself space and grace. Give yourself time. Stay committed to the process and focus on getting a little better each day. Above all, be patient with yourself—the big rewards come to those who refuse to quit before their time arrives. If I could do it, so can you. minervini.com
English
26
65
708
48.5K
20250609 retweetledi
Herman Jin
Herman Jin@ShanghaoJin·
凡所有相,皆是虚妄。若见诸相非相,则见如来 作为一个AI死多头,我知道在所有人狂欢的时候,但凡发出一点悲观的言论都必被讥讽、嘲笑 但是,我很想写此文,其实是复盘与反思:不但是为自己这两年半写X、做交易做的;也是也是对这个疯狂市场做的 很幸运,作为闲暇的爱好,写X结识了很多新朋友。感谢互相欣赏与启发,在过去2年里,我从24年4月初清仓BTC换成AVGO NVDA,大选前AAOI PLTR TSLA INTC,关税战后INTC GOOG,25年底又加仓INTC AAOI,到今年春节逃跑、再满仓AMD NOK(最近又操作了ORCL LITE)。虽然,有买彩票SUP归零,MRVL MDB也有赚有亏,但是我一直坚信AI对token的需求会让半导体every dog has its time,获得了不差得收益。期间不少小伙伴们也都赚到不少于10倍收益 然而,此刻要清醒的意识到,收益并不是我们选股牛,眼光独到,而是市场牛!我复盘了我自己今年3月最后一周,周末前全仓买入。赚钱并不是因为当时买对了AMD,而是当时call 对了市场资金结构 所以,在这个时点上更需要静心凝神,好好反思下市场接下去会如何 喝醉酒的人是不会知道自己喝醉的;沉醉美梦中的人也不会意识到黄粱一梦,这过程可以很久,我们也可以继续疯,但是必须要清醒地在更高阶意识中,认识到随时有被敲醒的风险 One of the four beasts saying, COME AND SEE – Revelation 6:1 不同于华尔街很多人,我是“坚信”AI的 30年来第一次,AI模型用并行计算替代串联计算,把半导体的进步直接变成了生产力(之前科技进步只是作为平台,让用科技的软件、以及用软件的人创造价值) AI的结果是token需求指数级升高vs 而半导体供应只能线性增加。加上整个半导体老登们在过去几年黄金时间,没有因 “信” 而扩产,今年需求真正开始爆发后全产业链缺货。造就了这轮大牛市 不过,大牛市的命脉是市场price in了: 模型收入加速度,不远得未来可以justify Hyperscalers 的Capex 要知道任何动摇命脉预期的都会让市场陷入恐慌 1. 铜墙铁壁:horseman之一就是微观、公司层面真的无懈可击 从看模型架构、从架构看需求,微观看半导体每一个赛道,每一个公司都近乎完美。每一个,都越看的细致,越了解深入,越是有自信---- 这就是典型低PE泡沫 Rule of Thumb:当看了几十、上百家公司,每家需求、盈利都真实solid,你就必须质疑这个市场有没有问题 低PE泡沫一样是一个泡沫,只是他不同于2000年。高PE是个预期泡沫,市场会不停设定更高的收益预期,一旦核心公司预期破灭,整体预期立即调整,多米诺骨牌一样一片片倒下。这就好像去年的IGV,市场忽然不给EV/EBITDA估值了 但是低PE泡沫没有这个问题,盈利增长会给你足够信心在稍微下跌之后不断加仓冲进去。毕竟谁能拒绝个位数PE的存储,每年涨价涨盈利,毛利率80%。这个过程中,小跌小买,大跌大买。只要不被切中“命门”市场会非常坚挺。持续性正反馈,好像美梦一样让人不愿醒来,也不相信是做梦(买过p2p的知道我说的什么感觉吧) 不对称反馈:明显的体验是在过去1年中,市场对于宏观负面风险的反馈明显弱于正面利好。此时,并非没有“野蛮人”,而是公司盈利能力的城池固若金汤。不管伊朗危机、油价、通胀、利率、Fed统统能拒之门外。反之,一旦命门被破,所有野蛮人都会杀回来,过去没跌的够的一次性补完 2. 去监管与宏观流动性宽裕: 整个金融体系自从2008年金融海啸之后经历了一轮长期的监管与去杠杆,为了弥补系统性流动性损失美联储对应展开了一轮持续的QE。银行当时平均杠杆x30~x40,后来降低到了x15~20。去年美联储降低了eSLR要求,实质性让银行理论可扩4-5万亿美元资产负债表 其实,4月后银行间流动性泛滥,也加剧了风险资产狂欢,低PE下杠杆入市 流动性是一切之王,流动性泛滥征服一切。如果说疫情期间120b/月是细水长流,现在就是开闸放水。对应过往平衡的QE没有转向QT情况下,直接放任银行去监管,制造大量流动性 但是流动性又如多巴胺,不是快乐分子,而是期待分子 放任之后,现在通胀黏着,Fed其实没有短期降息余地。甚至在危机一旦发生之后,都很难去QE,只有降息手段的Fed面对可能的危机是一个纸老虎。而短期流动性的另一面是,目前半导体、特别是存储中的杠杆已经太高了 3. 群雄并起:He went forth conquering, and to conquer 可能因为TSMC带头不相信AI,半导体行业整体在过去2年没有应对AI需求大规模扩产 后果是,当整个产业面临硅基洪流需求,忽然傻眼。过去半导体产业链主要是为了消费电子设计、定价的。而现在,不但订单忽然拉满,而且需要不停迭代架构 NVDA就从GPU fabless,进步到rack方案解决,再往token factory迈进。每个过程需要的技术、供货商都不同。所有供货商都没有经过如此量产设计。这就好像把一台拖拉机强行拉上高铁,让他以200mile速度奔驰,每个零件都会嗡嗡作响。半导体每个细分行业都缺货,缺产能,提高不够产能 与之伴生的一个问题是:原来卖方市场只有TSMC一个“守门员”控货源、控margin、控价格。原仔汤搓的产业有肉吃,需求方对价格产能有预期,良性发展。今天忽然五亿探长“雷诺”控制不住了,每个“小混混“都能来涨价、收保护费。真正的结果不是半导体都牛叉,而是失控:失控会影响每GW需要投入的成本核算,影响到命脉的预期。会给capex与模型收入会师蒙上阴影 4. 钢丝起舞 AI模型收入预期太满 回顾一下,市场坚信定价Anthropic + OpenAI + Gemini的收入会快速增加让Capex合理。这条命脉的信仰是不能有一点动摇的。哪怕略微的动摇都会导致二级市场剧烈波动 那么,过去GPT 5.0也有模型不及预期,市场质疑scaling law为什么没有让半导体大跌? 首先当时半导体没有像今天这样全面开花,也没有如此高的杠杆;其次,当时hyperscaler的现金流冗余较多,只要他们表示相信,拿钱来capex支持是很容易度过难关的。毕竟一旦他们有钱、继续capex,可见2年NVDA收益确定性高,没有人能顶风做空NVDA的 后来也发生过GPT被Gemini追平,并且导致Orcl持有其大量backlog订单而被做空。不过大总统的好兄弟老Larry敢对wall street说F you,用股权融资也要继续加注。很快, Opus横空出世让人意识到AGI时代已经到了 然而,今年capex已经到了770b,明年要1tril,华尔街是不会看Hyperscaler总营收多少的(道理就跟当时ORCL一样:OAI Anthropic可都是拿了他们投资款的),他们必须看到Anthropic 与OAI的增速持续,这个才是维系整个链条自转下去的核心 与此同时,这个游戏中大厂一直担任AI的监护人角色。但是他们的自由现金流已经变负了。“父母老了,力尽了“,后面的路要靠自己。而且市场满满预期下,对这条路一根筋的执着,不会给“主线”一丝一毫的容错 (你们可以回想NV rack液冷出问题,交换机CPO良率问题,这些小巫见大巫、假以时日一定能解决的问题,市场尚且没有任何容错) 命门在Anthropic:大过,大者过也。栋桡,本末弱也 供应延迟、乃至技术卡壳的问题是腠理之疾;任何模型端的风吹草动才是心腹大患 现在御三家模型的统一问题:算力不够,所以失智 我在自己大量使用模型编程、交易程序部署,多个AWS服务互动过程中发现,Opus4.8实际能力已经远远不如中国模型kimi了。虽然GPT还勉强可堪一用,但是也在逐步降智。市场在用固有软件思维考虑AI, 系统开发成本高、使用成本很低。试用好用将来也就可以稳定保证系统质量。但是AI是一个工厂,模型输出需要成本,饭店吃饭的人多了,厨房做不过来,质量自然变差了 第二个市场固有思维:指数的token需求是没有波动的,也不因质量改变。事实上,虽然我因为模型变蠢多用了很多token。但是我质疑token质量不断下降,token消耗增加的怪异剪刀差可以持久。现在很多公司确实对token使用有KPI考核,很多惯性订阅使用者也会加速使用,但如果算力瓶颈无法解决,我非常担心Anthropic增长曲线会变缓。而且算力问题非一蹴而就,也不是一个新模型能解决的 但是当市场开始意识到这个问题的时候,投资人会质问: 1. 跑不过蒸馏你的,模型花那么多钱训练干嘛? 2. 模型是不是商品化的model commoditize ? 当承载整个AI时代的模型“梦”也将受到质疑,整个盈收逻辑将来能否自我循环跑通将被质疑,所有今天拿了PE估值的半导体都同样会被质疑--- 你是不是周期性的?? 有人会说,失智既然因为算力不够就应该加大投入半导体解决问题 没错!但是钱呢? 如果刨掉OAI 与Anthropic拿到的融资款反哺云厂,他们能有多少净现金流可以继续支持这个投入?这个账对华尔街来说非常好算。当然不排除他们像当时orcl,哪怕融股权也要继续投入,熬过这关。到那时,可能市场的反应就会远远超过去年,假设AMZN Meta如果面临CDS 飙升、负现金流,他能否像老larry一样孤注一掷是不确定的 另外有人会说,OAI收入增加能不能弥补投资人担心。我认为预期如此之满的状况下,因为“比烂”而被替代是很难让人安心的(两军对战关羽被阵斩了,你说还有个张飞?没用的) 同时,我们也要意识到。这个Low PE的泡沫又是非常坚韧的,在没有直击到命门伤到命脉预期之前,几乎无敌。不过,一旦点中死穴就会轰然倒塌 我就业以来也曾看过好几次次纸碎金迷,潮起潮落,没有一次全民疯狂会ends well; 也没有一次梦在其中的人,会想明白这个赛道“怎么输”。最近一次币圈小伙伴熟悉的Defi summer。Uniswap很短时间狂卷几十bil USD,当时觉得取代部分传统金融只是时间问题、多少问题,根本不会想、也想不明白“怎么会输”。然而,如果喝醉酒的人想明白怎么会输,他就不会输了 不管Anthropic模型怎么样,面临什么问题,都不会动摇AI是人类第三次工业革命。AI过程不会受任何危机影响,碾着狂欢者的尸体也会继续前行,直到他替代几亿脑力劳动者 最后评论Low PE泡沫:不要说做空,哪怕踏空,痛苦指数都非常高。回头看看2007年做空subprime的哪些基金,有几个活到了最后狂欢? 牛市踏空都会变成笑柄的环境,我接下去会需要很多Aura力量与专注的力量,去一边看空一边做多。请原谅会减少在推特上消耗自己修为论道了,赠言: 可以喝酒,但是不能喝醉 可以讲故事,但是不能真信 可以跳上party桌子跳舞,但是双眼必须紧紧盯着DJ,他放着音乐跑了,你也得跟着跑
中文
458
574
3K
1.5M
20250609 retweetledi
Jukan
Jukan@jukan05·
TSMC CEO’s remarks to TSMC employees regarding bonuses Q: Why are engineers taking a pay cut (smaller bonus) while VPs and above are getting stock — effectively a raise? A: I'll have to go back and look into the relationship between the two. Q: The articles of incorporation stipulate that profit-sharing must be no less than 1%. If the company keeps cutting it down toward 1%, can the articles be amended to protect employee rights? A: No. Q: Last time, the compensation restructuring was "cut the bonus, raise base salary." This time you're cutting the bonus again — will base salaries go up? A: Future raises will be larger for lower job grades and smaller for higher ones. Q: Is there a ceiling on compensation? A: There is no ceiling. Q: (The questioner's logic was unclear. The gist was questioning whether overexpansion of fabs and headcount is diluting the bonus pool.) Q: The bonus isn't increasing with EPS, but you mentioned earlier that the employee birth rate is 5x Taiwan's average. That doesn't add up. Shouldn't the company follow the government's lead and have the company raise the kids? A: Not possible. Q: Rather than putting that money into green energy investments, why not pay it out to employees, or at least restrict it to charitable donations? A: Not possible. Q: Even in downturns, the money doesn't get returned to employees. Wouldn't it be better to spend it on improving company benefits? A: The money doesn't get redistributed that way. Q: Since you're developing AI, can we use fewer people and pay out more in bonuses? A: I'm working on it. Q: Can we spend more on software and hardware equipment? A: Yes. Q: Will ETE (equipment-related roles) get raises? A: I'll go back and check. They should be increasing. Q: You said you can't keep sustaining 1.3x growth — is that because you're seeing some operational red flags down the road? A: There are no red flags. (Management interjects) Complains that other companies' stocks are pumping for no good reason, and that TSMC's stock is severely undervalued. 95% of the world's AI chips are made at TSMC — please keep up the good work, everyone. Q: Elon Musk is offering 2x salary to poach our people. Aren't you worried? A: First, I'm not worried. Second, do you really believe Musk? Do you believe you can smoke cigars like he does? Q: Anything on the financials going forward? A: We're considering continuing to increase the dividend. Q: Looking at the Samsung strike, TSMC's profit-sharing ratio has already dropped to 10%. Can you guarantee that at least 10% will continue to be distributed to employees going forward? A: No. The DRAM industry "earns in one year and lives off it for ten" — it's a different industry, not comparable. Q: Do you support employees forming a union? A: If a union is formed, management has to spend a lot of energy dealing with it, so I don't encourage it. You're already getting this much in bonuses — why would you want a union? Q: What if business turns south in the future? A: That's not a scenario we're assuming. But if the global economy really did collapse, management would take care of its people.
English
40
33
438
257.5K
20250609 retweetledi
Serenity
Serenity@aleabitoreddit·
And now… $MU finally hits a $1 Trillion marketcap. I did say this looks like the next $NVDA given how memory demand looks structural with AI. This stock probably made a lot of millionaires going from $80 to $887.
Serenity tweet media
Serenity@aleabitoreddit

$MU looks like the next Nvidia. When Nvidia was $400B (now $4.5T+), markets thought GPUs were a short-term cycle. Same with memory today. AI has broken that cycle. With the same "Made in America" and White House backing like $INTC: Don't overthink things with Micron.

English
185
89
2.1K
725.9K
20250609 retweetledi
Yael _Formosa 鹿島之風🧋🇺🇸🇮🇱🇯🇵
『澳洲女演員 Holly Valance 表示,每個人「一開始都會偏左」。 但當你開始「經營生意或買房子」時,就會「醒過來」。 「然後你就會發現,他們那些想法有多爛。」 很難反駁啊,Holly。』
CyberTruck Owner@CyberTruckFan9

@Ryandally08 Not wrong Churchill once said: "Anyone who was not a liberal at 20 years of age had no heart, while anyone who was still a liberal at 40 had no head."

中文
3
22
238
16.2K
20250609 retweetledi
Woody
Woody@woody168888·
投資的本質是很簡單乾脆的,不要被幻象給迷失,不要執迷某個標的,我們需要的是當下最好的故事,最多人同時買入,走勢最好看的標的,這才有韭菜賺錢的空間。 而不是整天瞎猜主力割不割我,這邊持倉.資金費率有問題,整天PVP思維,或者是明明主流在AI卻選擇了傳產股票,在漲漲跌跌中損失了本金。 認命吧,我們韭菜只能在漲漲漲漲漲大勢中去贏錢、贏大錢,其他沒有趨勢的都不具有發財的條件。
中文
12
18
268
11K
20250609 retweetledi
Serenity
Serenity@aleabitoreddit·
“Leading” Glass Substrate players that were name dropped if you’re curious: • $LPK — TGV Equipment • $GLW — Glass Materials • $ASGLY (5201 T)— Glass Materials • $NIDGY (5214 T) — Glass Materials • $LRCX — Etching Systems • $DSCSY (6146 T)— Dicing Equipment • $SMHSF — Bonding Systems • $ONTO — Inspection Tools • $KLAC — Inspection Tools Fun to see the stuff I’ve called out early in the year like LPK at ~$150m MC get mentioned as a critical player by Trendforce and others.
Serenity tweet media
English
101
197
2K
366.6K
20250609 retweetledi
Mark Minervini
Mark Minervini@markminervini·
No one cares about your money or your future like you do. Learn how to invest; control your financial destiny. It starts with sound criteria and proper risk management. Have a great day.🤑
Mark Minervini@markminervini

There have only been a couple of times in my life when I entrusted others to manage my money—mostly early on, when I was inexperienced looking for guidance and looking for someone to get me rich. In hindsight, I’m actually grateful for those lessons. They forced me to take full control of my financial future. What I learned from those mistakes ultimately made me thousands of times what I lost. But not long ago, I made a similar mistake. I placed trust in @loukerner to be both a friend and competent in a space I was exploring. While crypto performed exceptionally well (tripling in price), the results from the SPAC he pitched were quite the opposite to the tune of an 85% loss. Fortunately, the investment was a tiny test position simply because we were "friends." Unfortunately, not everyone I know was as cautious. 99.999% of my money is managed personally—and that’s exactly how I like it: in the best hands I know… my own. My suggestion to you is: learn how to invest and self-direct your capital. Why? Because no one cares about your money or your future like you do.

English
24
32
710
81.4K
20250609 retweetledi
LIWEI_TW Capital
LIWEI_TW Capital@LIWEI_TWCapital·
How many Taiwanese semiconductor companies are now genuine AI infra bottlenecks? Are there actually near-term alternatives to Taiwan’s semi-ecosystem? Or are we also giving ourselves a bit too much credit for the AI buildout narrative?
Leonard@Leoskie_L

我自己整理的台股/美股 半導體瓶頸 有需要就存入書籤!! L1 — 材料層 高功率、窄線寬雷射 3105 穩懋 (WIN)​:InP 基板,衛星通訊/激光雷達核心,中國管制後需求放大 4-6 倍 COHR Coherent​:雷射元件,光通訊/資料中心 LITE Lumentum​:光通訊雷射,3D 感測 玻璃核心基板 GLW 康寧​:TGV 玻璃通孔技術,先進封裝核心材料,全球 60%+ 市佔 ABF 基板 3105 穩懋​:台灣 ABF 載板供應鏈 3037 欣興 Unimicron​:ABF 基板,AI 晶片封裝 3189 景碩 Kinsus​:ABF 基板 先進 PCB 4958 臻鼎 Zhen Ding​:高密度互連 PCB Compeq /敬鵬​:車用/AI PCB ——— L2 — 電力層 800V 電力傳輸 / 電源架與匯流排 2308 台達電 Delta​:寬溫電源(-40°C~85°C)、SST 固態變壓器、電源機架 1519 華城​:電力變壓器、軍工基地電網 VRT Vertiv​:數據中心電源+散熱整合 ETN Eaton​:電力管理,800V 架構 MLCC Yageo 國巨​:全球前三,AI 伺服器需求 電感器 奇力新 Chilisin​:AI 電源模組 2308 台達電​:電感器/電源模組 電網變壓器 1519 華城​:AI 數據中心電網擴容 2308 台達電​:固態變壓器,效率 98.5% 燃氣渦輪機 GE Vernova (GEV)​:數據中心備用發電 Siemens Energy​:燃氣渦輪機 ——— L3 — 散熱層 熱管理與液冷系統 3017 奇鋐​:邊緣設備散熱,功率密度 1.5 倍 3324 雙鴻​:散熱 Cooler Master 訊凱​:液冷散熱模組 VRT Vertiv​:液冷整體方案 ——— L4 — 記憶體層 HBM 與先進記憶體 MU Micron​:HBM3E 綁定 NVIDIA,1γ 節點,DDR4 毛利率 67.9% 2408 南亞科​:DDR4 供給缺口,投信大力買超 2344 華邦電​:NOR Flash/DRAM,利基型記憶體 6770 力積 PSMC​:DRAM 設計 SK Hynix​:HBM 全球第一 ——— L5 — 連接層 矽光子學 ALAB Astera Labs​:CXL 互連晶片,記憶體池化 LITE Lumentum​:光通訊元件,矽光子 COHR Coherent​:光收發模組 Intel​:矽光子收發器 高速 SerDes CRDO Credo​:高速 SerDes MRVL Marvell​:高速互連晶片 ——— L6 — 封裝層 先進封裝 2330 台積電 TSM​:CoWoS 壟斷 3443 創意電子 GUC​:AI 芯片 ASIC 設計 6643 M31​:軍工芯片 PHY/高速接口 IP 混合鍵合 2330 台積電​:SoIC 混合鍵合技術 ASML​:混合鍵合設備 BESI​:混合鍵合設備 ——— L7 — 管理層 BMC 晶片 5274 信驊 Aspeed​:BMC 壟斷,伺服器 ——— L8 — 雲端 CapEx 雲端基礎設施 ORCL Oracle​:IL6 認證唯一, AMZN AWS​:雲端市佔 31% MSFT Azure​:雲端市佔 22% GOOGL Google Cloud​:雲端市佔 11% ——— 額外補充核心 AI 大腦(對應半導體核心) NVDA​:AI GPU 絕對核心,CUDA 生態 AMD / AVGO​:第二算力 + 網絡 ASIC ARM / INTC / GFS / UMC​:架構與成熟製程

English
2
1
16
6.9K
20250609 retweetledi
Jukan
Jukan@jukan05·
I read Goldman Sachs’ AI report, and I was genuinely impressed. The core insight is as follows: Agentic AI could turn AI from a capex-heavy cost burden into a business where usage growth drives margin expansion. As token costs fall, more complex agents become economically viable. These agents then consume far more tokens through longer context windows, repeated reasoning loops, validation, tool use, and always-on background monitoring. This increase in token usage improves infrastructure utilization, strengthens unit economics, and gives hyperscalers and model providers more room to reinvest in model quality, distribution, and capacity. In other words, the bull case for AI capex is not simply that usage will grow. It is that this usage growth can increasingly flow through at attractive incremental margins. Goldman Sachs argues that this margin inflection is beginning to appear from 2026 onward.
Jukan tweet media
Jukan@jukan05

We have only just entered the early innings.

English
106
450
2.9K
508.1K
20250609 retweetledi
Kay Capital
Kay Capital@keyahayek·
如果你的男朋友这两天没有打开 Claude Code / Codex, 没有满嘴「半导体超级周期」、DRAM、NAND、HBM、PCB 永远缺货,没有突然研究韩国国运,什么 SK 海力士是 AI 铲子股、三星明年盈利世界第一、斗山能源是韩国版 GEV,没有深夜感慨这是国运、是东亚制造业复兴。 说明他没有任何投资理财观念,看不懂产业趋势,错过了时代贝塔。 但也说明他不会被泡沫诱惑,没有 FOMO,不会顶部 All in ,不会把被套也不卖说成长期主义。这种男人,情绪稳定,账户也稳定。 可以考虑结婚了。
中文
291
111
1.6K
338.1K
20250609 retweetledi
林晚晚的猫
林晚晚的猫@linwanwan823·
这是我今年见过最牛的「美股新股神」。 24岁的德国人,被OpenAI开除, 转头开基金,2亿变50+亿。 -先说这个哥们。Leopold Aschenbrenner。 我去年半夜跟朋友喝酒, 有个马来人按头推荐让我去看一篇 《Situational Awareness》, 说AI正在朝这篇写的方向发展。 后来我才知道, 这文章当时在硅谷VC圈里炸锅, 川普的女儿伊万卡也转了。 就是Leopold Aschenbrenner离职后写的。 -去年4月,这哥们据说因为泄密被OpenAI开除。 出来之后没多久,直接开了个对冲基金, 基金名字就叫那篇文章, Situational Awareness Fund。 LP是Stripe两兄弟和前GitHub CEO Nat Friedman。 去年Q4披露持仓2.5亿,今年再披露直接就55亿。 一年涨了20多倍。 -我们再来看他持仓。 第一仓 Bloom Energy。给数据中心卖燃料电池发电的。去年4月16块一股,今天279块。一年17倍。 我去年也盲跟着买了, 买了到今天一直没动,赚了5倍左右。 第二仓 CoreWeave看涨期权, AI算力卖家,英伟达亲儿子,IPO半年股票翻了3倍多。 第三仓 Intel看涨期权。 去年所有人都说Intel完了的时候他进去重仓买CALL,结果Intel一年涨356%。 所以晚晚我现在有个新思路: 紧跟OpenAI离职人群。 看他们在做什么,基本上就能看到AI投资和下一个大方向的路标。
林晚晚的猫 tweet media
中文
105
373
1.8K
153.9K
20250609 retweetledi
LEI
LEI@TheMarketMemo·
躲過下跌,則可能錯過上漲。 然而錯過上漲的代價是巨大的!
中文
17
59
429
95.5K
20250609 retweetledi
Macro_Lin
Macro_Lin@LinQingV·
把先进封装设备这条线从头到尾分析了一遍,从工艺原理到设备供应商再到各家存储厂的路线选择,信息量比较大,整理一条长推把脉络串起来。 从封装工艺看,目前主流的把die连在一起的方式有三种。Mass Reflow,就是传统回流焊,把所有bump一次性过炉焊上去,速度快成本低,但pitch不能太细。TCB(热压缩键合),逐颗die加热加压,精度更高,能做更细的pitch,但速度慢。Hybrid Bonding,铜pad直接键合,连bump都取消了,pitch可以压到10μm以下,是终极方案但成本也最高。 这三种工艺对应的场景完全不同。台积电的CoWoS-S,就是英伟达GPU用的那个2.5D封装,把逻辑die和HBM贴到硅中介层上,用的是mass reflow。台积电的SoIC,3D垂直堆叠方案,用的是Hybrid Bonding,AMD的X3D就是这个路线。Intel的Foveros Direct 3D也是HB路线,bond pitch做到9μm,下一代要缩到3μm。 HBM这边更有意思,三家存储厂走了完全不同的路线。SK海力士从HBM2E开始用MR-MUF,一次性回流加模塑底填充,效率高,对翘曲控制好,12层堆叠良率做到了业界最高。三星和美光走的是TC-NCF,逐层热压缩加非导电薄膜。两条路线各有优劣,但SK海力士靠MR-MUF的良率优势建立了HBM市场的绝对领先。 到了HBM4这一代(2026年量产),路线分化更加剧烈。SK海力士的态度是能用MR-MUF就继续用,同时储备HB技术作为备选。三星反过来,要在HBM4上直接跳到Hybrid Bonding,它叫HCB,但目前良率据说只有10%左右。美光继续TC-NCF做12层HBM4。至于HB什么时候真正成为HBM的标配,大概率要等HBM4E甚至HBM5,因为JEDEC把HBM4高度标准从720μm放宽到775μm,给bump方案多续了一代命。 这里面最值得关注的变化在设备端。TCB设备的供应链正在剧烈重组。SK海力士那边,Hanmi原来是独家供应商,现在ASMPT快速切入,拿下了大约一半的份额,Hanmi和Hanwha之间还爆发了专利战,互相消耗。三星用自家SEMES的设备,体系内循环。美光最关键,之前用Shinkawa和Hanmi,但HBM4对精度要求更高,据渠道信息美光将全面转向BESI,而且因为地缘政治因素不太可能用ASMPT。 所以BESI的增长逻辑其实比市场叙事里说的更丰富。市场只盯着Hybrid Bonding的垄断故事,但TCB这条线BESI也在吃份额。HB端,台积电SoIC放量加上Intel Foveros Direct上线,是确定性最高的增量。TCB端,美光HBM4转单是正在发生的事情。两条线同时扩张,这才是花旗把估值从37倍扩到40倍的底层支撑。 但40倍P/E 2027E定价的还是完美执行。设备公司的订单弹性是双向的,台积电如果在2027年调整HB的capex节奏,或者HBM4的bump方案比预期多撑一代,BESI的订单簿都会受到影响。这也是潜在的风险点。 Hybrid Bonding的产业方向倒是没有争议,设备供应链的重组也在加速。
Macro_Lin tweet media
中文
17
101
362
50.7K
20250609 retweetledi
Mark Minervini
Mark Minervini@markminervini·
If you are living an angry life, feeling like a victim, and you think you are hurting or spitting others by being unkind or apathetic, you are only depriving yourself of the fullness of life and you are living a lonely existence. Forgiveness, gratitude and love will free you from the demons that haunt you and make you bitter. 😇🙏 Don't try it... live it and trust it.
English
50
121
1.4K
166.4K