Ad astra per aspera!

Taipei, Taiwan
Joined May 2018
Ryan Wang 🇹🇼 retweeted
Elon Musk
@elonmusk17h
Next year, I will give one of my SpaceX shares to every kid in America. This will be the first of many.
35,090
53,062
12,697
520,717
26,489,061
Ryan Wang 🇹🇼 retweeted
According to @insane_analyst, several people who attended Anthropic's investor day told him that the company was "vague and gave little information" along with nobody asking an intelligent question. Goes on to say he suspects a subset of departments at Jane Street may have blown up leading to the market selloff last week.
16
32
6
361
57,619
Ryan Wang 🇹🇼 retweeted
Replying to @eric_khun
As long as I’m CEO.
124
195
86
5,123
646,125
Ryan Wang 🇹🇼 retweeted
I’ve spent a few days thinking about the Tesla Roadster reveal and landed somewhere I haven’t seen much discussion. I’m neither a Tesla fanboy nor a Tesla hater. And I couldn’t care less whether what we saw technically qualifies as a flying car. Three takeaways: (1) The risk of that live demonstration going horribly wrong was real. Almost no CEO would willingly expose their company to that kind of public humiliation for a product with so little financial upside. Elon did. (2) That willingness to attempt the seemingly impossible is what makes Tesla special. It’s also arguably Elon’s greatest contribution to America. He’s helped bring back something we’ve lost. The American spirit of chasing big scary ideas that most people think are impossible. That mindset is spreading across a new generation of founders and its long term impact on America’s future could dwarf anything Tesla ever sells. I think this is deeply underappreciated. (3) Can we just acknowledge how insanely cool that moment was? We’ve reached a strange place culturally where even extraordinary feats of human ingenuity immediately become arguments about why they don’t count. Not everything needs to be a debate. Sometimes we should just be able to stand back together and marvel at what humans are capable of. I’ve been both bullish and critical of Tesla’s ambitions over the years. But watching Elon and his team push the boundaries of what’s possible reminds me why I’m glad we have him. Whether the Roadster ever makes Tesla a dollar or impacts $TSLA stock price is almost beside the point. We need more people willing to risk failure in pursuit of things the rest of us think are impossible. Because when they succeed we all win.
1,066
2,377
340
19,256
603,039
Ryan Wang 🇹🇼 retweeted
Honestly, I know I say this a lot, my team does all the hard work. They are the absolute best! So many you can call out, but I always tell them ‘I work for you, tell me what you need?’ Making great products starts with an idea, then committing everything to it, then GO!
51
119
8
2,565
80,214
Ryan Wang 🇹🇼 retweeted
Tesla Roadster: Lars Moravy deep dive with Henry Catchpole (Hagerty) — new details —— Powertrain specifics —— ✅ Dual rear motors deliver more power & torque than a full Model S Plaid ✅ No differential: each motor drives one rear wheel via its own 8.4:1 planetary gearset ✅ Individual inverter per motor ✅ Motors are smaller and more power-dense than the Plaid units —— Sucker system details —— ✅ Four fans of ~20 kW each (derived from Tesla’s heat-pump compressor motors) ✅ Drop-down independent skirts seal the underbody ✅ Generates >2G of cornering grip from 0–50 mph —— Thruster & tank specifics —— ✅ Nitrogen stored at up to 10,000 PSI ✅ Lars: “It just seemed wrong to put fuel into a car” ✅ Full 600L Rocket tank currently takes 6–7 hours to refill (target ~30 min) ✅ Rocket 0–60 of 0.55s is a rollout figure ✅ Thrust version reaches ~0.85s with over 5G —— Chassis & dynamics —— ✅ Track Mode lowers the car 40 mm and triples suspension stiffness ✅ Double-wishbone front / multi-link rear ✅ Front aero elements can move up to 80° ✅ Width: ~2 meters (78.7 in) —— Battery & performance notes —— ✅ Bespoke 4680 cells with aluminum cans (instead of steel) + double-sided cooling ✅ Target of 25–30 minutes of unrestricted track driving per charge ✅ Top speed: ~210 mph (Performance) rising to 230 mph on higher trims
4
27
4
230
12,711
Ryan Wang 🇹🇼 retweeted
Talked to Michael from Coreweave about debt six months ago: 18:18–23:43: Jason asks about CoreWeave’s debt; Michael answers (cleaned of filler words and stutters, otherwise as spoken) JASON: How does all that debt work? That is something you guys specialize in. Corporate debt… I’m in the venture business, people are like, “Why should I be in venture when corporate debt pays so well?” Corporate paper’s so huge. I’m curious how this fits in, and what interest rate people are paying on, you know, a billion dollars in infrastructure. What do they pay on that? MICHAEL: So CoreWeave has really been the innovator around a lot of the financing engines that have come to bear on this. We did the first GPU-based loans. I’m going to try to explain this in a way people can understand. What we do is we go out and we find a client. Let’s use Microsoft; you brought them up before. Microsoft comes to us and says, “We’d like to buy some compute from you.” And we say, “Okay, great. We’re going to sign a contract.” Once I have a contract in hand, I create something. It’s not a particularly creative name. It’s called the box. JASON: Yeah. MICHAEL: And what I do with the box is I take my contract with Microsoft and I put it in the box. I go to Jensen and buy the GPUs, I put them in the box. I take my data center contract, I put it in the box. And now the box governs cash flow. It has a waterfall of cash flow that comes into it and goes out of it. So the way it works is I build the compute, I deliver the compute to Microsoft, and they pay the box. They don’t pay me. It goes into the box, and the first thing it does is pay the data center. It pays the power bill. It pays the interest and the principal. And then whatever’s left flows back to us. So it is an incredibly well-structured, time-tested, pressure-tested vehicle to be able to borrow money against client paper and all of the other collateral around the deal. That’s why CoreWeave, a company many people had never heard of, was able to go out and raise $35 billion in 18 months to build infrastructure at scale. But what’s important to understand is the economics in this box are such that within 2.5 years of a 5-year deal, we have paid for everything. JASON: The principal’s been paid off. MICHAEL: The principal’s been paid off, the interest has been paid off. The return into the box is such that we’re able to generate returns to our company at the box level, which gives the most sophisticated lenders in the world, whether banks or private equity funds or whoever, confidence that they’re going to achieve the one rule of lending, which is “give me my money back.” JASON: Yes. Works better when that happens. So they look at this box and they’re like, “Wow, we’re really confident we’re going to get our money back.” And maybe they want 10 boxes. MICHAEL: That’s correct. JASON: And if any one box goes upside down, you can deal with it and it’s not as acute. MICHAEL: That’s correct. And they don’t cross-pollinate. They don’t cause contagion across the boxes. They’re all independent and discrete. That’s one. Number two: as you do this and show the lenders how this financing mechanism works, they continue to lend you money at progressively lower rates. When you think about our cost of capital over the last two years, we have dropped it by 600 basis points. JASON: Wow. MICHAEL: It is enormous. So you’re seeing a company driving its cost of capital down toward where the hyperscalers borrow, which will let us be competitive with them over time. And we have been extremely militant and diligent about feeding, watering and caring for those boxes, so we continue to have access to the capital markets in a way that lets us build and drive our business. JASON: That means you have to say no to maybe some people who want to be in the box. MICHAEL: Yeah. We look at some deals where they want to buy GPUs for a year, and I say that’s not a deal I can do, because it’s too short for me to amortize the expenses. So I won’t do that. JASON: And they can go to another provider who maybe wants to take that risk on, who has extra capacity. MICHAEL: Absolutely. But our business is really built around the risk management of getting to scale. In my mind, during this period of disequilibrium, when there aren’t enough GPUs in the world to provide compute for all the different use cases in AI, what matters for my company is to get enormously large. That drives down our cost of capital, and it means information flows in from all different parts of the market: large language models, high-speed trading, search, all of these things. They feed information back to us that tells us the next product we need to build, where they need help scaling, or what type of compute they need. All of that information flow is incredibly valuable to us. (Jason moves on to demand at 23:43.) $CRWV youtu.be/-JBhTBu9ZbA?t=1083
100
60
19
777
498,275
Ryan Wang 🇹🇼 retweeted
BG2 w/ @JensenHuang. Does AI Revenue Justify Capex, ‘27 AI Bottlenecks, Singularity, Open vs Closed & more. 🚀🇺🇸 @altcap @_clarktang @nvidia
42
133
25
1,055
345,961
Ryan Wang 🇹🇼 retweeted
#TSMC 3Q26 net profit surges 68.7% YoY to a record NT$763B, marking its 6th straight quarter of profit growth. Meanwhile, its 2nm ramp is gaining steam, with wafer revenue share jumping to 8% from just 3% in 2Q.💡More: pse.is/9r7zbz 🔗
2
7
3
33
7,752
Ryan Wang 🇹🇼 retweeted
Yields on one series of CoreWeave's bonds spiked to 15% this morning. Here's their full debt stack visualized. $CRWV
5
6
2
72
11,215
Ryan Wang 🇹🇼 retweeted
我不能允许中文圈的朋友错过这篇文章!读完大概8分钟,这可能是10月份最值得的8分钟。 我给大家翻译好了👇 如何变得更有雄心:写给与模型协作的人类使用指南 阅读时长:约 8 分钟 我在 Claude 团队工作。这是一段很有意思的经历:几乎每天,我都会发现有人借助 Claude 完成了某项壮举,而这些成就往往是我未曾考虑过、曾认为不可能做到,或者单纯因为缺乏好奇心和时间而未曾尝试的。就在这个月,它可能参与协助发现了一颗新行星;我看到了第一支自己真心喜欢的 AI 音乐视频;我们甚至还在讨论数学的新纪元。 模型的能力并非均衡发展。正如数学家陶哲轩(Terence Tao)所言,模型“在某些方向令人惊叹,而在另一些方向却显稚嫩”。有些人成功驾驭了这些惊人的能力,仿佛翱翔于天际——看着大家用我参与打造的产品攀登难以置信的高度,十分令人着迷。但还有许多人仍在努力适应他们的新羽翼,困惑于除了询问“法国的首都是哪里”或“帮我把这封邮件润色得别那么生气”之外,模型到底还能做什么。 我和其他从事 AI 工作的人经常呼吁:在使用 AI 时,我们应当更有雄心(ambitious)。这已经成为了一句集结号角式的口号。每当我们看到令人赞叹的应用时,就会转发并写道:“我们的野心还不够大。”但我深知将这种建议付诸实践有多么棘手,因此,这里有一份关于“如何变得更有雄心”的操作指南。 我认为“雄心”可以分为四种维度:速度(Speed)、高度(Altitude)、深度(Depth)以及广度(Range)。 速度(Speed): 用更少的时间完成同样的工作,这是 AI 经济价值的基本盘:如果你需要写代码,我用 Claude 能完成得更快。这一点已被广泛认知,且不需要在使用 AI 的方式上发挥太多创造力,因此本文略过不谈。 高度(Altitude): 提升抽象层级,关注最终结果而非单一具体任务。 深度(Depth): 像行业顶尖专家一样去审视和推演任务。不再停留于任务表层,而是深入一层,触及底层的原始数据、工具和根本问题。 广度(Range): 借鉴多个领域的经验与方法,而非仅仅局限于任务表面所需的那一个领域,去调用那些你平时通常不会用到的知识、工具和技术。 我主要使用 Claude(显而易见),但我的建议并不受限于特定模型。 高度(Altitude) 我们常随口将其称作“让模型在更高的抽象层级上运行”。当你给模型分配一项任务时,它执行的不过是你对“如何达成目标”所做的最优猜测。要在“高度”上更有雄心,不妨问问自己——接下来你希望发生什么?目标背后的终极目标是什么? 去年年初,我还会让 Claude 帮我排查并修复单个软件崩溃;而到了后来,我直接要求 Claude 确保我们的应用程序“零崩溃”。 让我们以一个日常(甚至可以说是琐碎)的办公任务为例。假设你在一家名为 Acme Inc 的虚构公司,需要为季度业务回顾(QBR)制作一套汇报幻灯片(PPT)。如果你输入一堆数据,Claude 很乐意为你生成几张幻灯片;如果你提供模板,它也可以顺畅填入内容。但这很难与“发现新行星”相提并论。来看看一些具体的步骤,让我们的思路逐层提升雄心: 基础步骤: 大多数业务回顾都想知道本季度发生了什么变化、原因是什么、以及应该采取什么对策。克制住仅在模板里更新数字的冲动——尽可能把你能获取到的所有原始数据喂给模型,让它理解你的业务逻辑:“告诉我本季度发生了什么变化,背后的原因是什么,以及我们应该如何应对。” 进阶步骤: 会议(至少高效的会议)最终都应产出明确的决策。向 Claude 提问:“基于你目前了解到的所有情况,梳理出眼前最紧迫的三个决策是什么,并针对潜在的发展方向展开分析,分别列出正反两面的论据。” 高阶步骤: 季度业务回顾诞生于这样一个时代:那时需要人类耗费大量精力筛选数据、厘清数据背后的含义,并判断公司是否需要做出业务调整。你可以这样要求 Claude:“业务评审的存在是为了让管理层及早发现问题,把资金和人力调度到正确的位置。请设计一种比‘季度汇报 PPT’更好的机制来实现这一点,并结合我们以往的历史数据证明该机制能更早发现问题;同时让这套新方案与当前机制并行运行一个季度。” “高度”策略在追求某种终极成果时效果最好,但在纯创意类任务中表现稍逊。判断何时适合在“高度”上发力的一个显著标志是:你是否需要为了手头正在做的事情去开会讨论。 当然,目前这一方法仍有局限(在 2026 年,“Claude,帮我去赚钱”对绝大多数公司来说单凭一句话还是行不通的),但我普遍发现:随着模型能力的增强,它们处理高层级抽象任务的表现也越来越出色。 “高度”也是构建持久工作流(将人类移出执行闭环、无需你反复输入一项又一项任务)的关键所在。要做好这一点,通常需要重构你的工作流程——为了让 Claude 能够自主确保应用程序零崩溃,我首先得保证它能轻松拉取崩溃报告,理解哪个应用版本对应哪次 Git 提交,并赋予它访问 Windows、macOS 和 Linux 环境的权限,使其能够真正独立地复现并验证修复方案。 深度(Depth) 有时候,你只是想把某一项具体任务执行得极其出色。在“深度”上拥有雄心,意味着促使模型去使用该领域最顶尖专家所采用的工具、技术、媒介和专业知识——并持续深挖,直到连那些专家都能从你的产出中获得启发。 继续沿用 Acme Inc 的季度业务回顾例子。假设 Acme 公司正面临用户流失(随着时间推移客户流失)的困境。我发现结合以下技巧非常有效: 坚持使用业内顶尖专家所使用的工具: 当被要求“表现得像个专业人士”时,模型往往只是换了套行业术语,而没有真正改变工作方法。要在深度上解决问题,我们必须使用顶尖数据科学家和统计学家同款的工具。当 Claude 倾向于生成简单的 Markdown 表格时,坚持要求它构建数据库和统计模型。你不需要预先知道专家用什么,Claude 通常心里有数。 设定看似离谱的目标,并告诉模型这完全合情合理: 你需要将网页加载时间从 10 秒缩短到 5 秒?不妨直接要求达到 200 毫秒。如果模型觉得某件事不可能办到,重申这完全是可解的,它只需要继续探索尝试。 挑战既有认知: 尤其在商业领域,许多我们坚信不疑的事情其实只是对现实的粗糙近似。将原始信息与数据提供给模型,让它以对抗性的视角去检验既有假设:“我们一直认为用户流失是由于上手引导(onboarding)体验不佳导致的。请尝试从原始数据中复现并验证这一推论。” 创造新知识: 这一点可以顺理成章地承接对既有认知的挑战——一旦 Claude 已经深入数据腹地,让它去探寻盲区:“基于系统日志构建一套能够复现去年业务情况的客户模拟系统。然后找出我们尚未察觉的最早流失信号,并证明该信号能在未参与训练的新数据上完成有效预测。” 引入外部裁判: 切勿让模型对自己打分评判。我相信随着模型智能的提升,这一问题终会解决,但在 2026 年 10 月的当下,我强烈建议引入外部验证机制。最理想的方式是提供一个可供模型直接运行并实时获取反馈的评估工具。 我经常使用的一个简短提示词(Prompt): “我希望这项任务更加深入,而不是范围更大。在重写任务前,请先列出:(1) 该领域的顶尖专家会做哪些我目前没有要求的事情;(2) 他们会从哪些最原始的材料入手,而不是使用已被加工整理过的数据;(3) 他们会使用哪些具体的方法和专业工具;(4) 除了你和我之外,还有什么客观指标或外部手段能证明结果是正确的。完成以上分析后,重写我的提示词,精确提出这些要求。” 广度(Range) 简而言之,在“广度”上拥有雄心,意味着我们从多个不同领域的视角来审视手头的工作——引入比过去在成本考量下所能承受的更多跨界学科。这并不是多么独到的洞见,每一部盗贼冒险电影(Heist movie)在前 30 分钟里都在忙着集结性格迥异、各怀绝技的团队成员。随着 AI 工具让多智能体系统(Multi-agent systems,包含多线程、编排器或单纯的子智能体)的运行变得越发轻而易举,要求 Claude 跨界引入额外领域往往能带来惊艳的效果。 将其付诸实践并不简单,需要你发挥创造力。直到我看到同事 Alex Albert 使用 Blender 渲染室内设计,我才想到可以用 Claude 为我的个人主页构建 3D 虚拟环境——尽管我以前做过电子游戏!然而,这也是我见证最多神奇瞬间发生的地方。 具体该如何操作?我通常采用以下方法: 尝试用动词替换名词: 这是来自汤姆·凯利(Tom Kelley)与乔纳森·利特曼(Jonathan Littman)著作《创新的艺术》(The Art of Innovation)中的一个经典技巧。季度业务回顾不是“一套 PPT”,而是“在 30 分钟内,让事务繁忙的人相信一个关于因果关系的故事,并在不确定性中做出决断”。提炼出动词之后,不仅能让我、也能让 Claude 更容易理解哪些领域能为这一目标带来独特且有价值的视角。 组建你的特遣队: 假设这套 PPT 必须达到极高水准,否则外星人就要毁灭地球——人类会组建一支怎样的梦之队?为了保证产出质量,我会拉入一位数据分析师、一位调查记者和一位编辑;如果需要现场演讲,或许还要加入动态视觉设计师和音频设计师;甚至还可以组建一支对抗性的“红方团队”,邀请一位做空机构研究员和一名法务会计师加入。这里很容易不小心滑向纯粹的角色扮演,这只会让上下文变得古怪而不会实质提升产出质量。使用这个方法的核心是寻找对应的学科领域与专业方法:不要让模型去“扮演”一个做空机构空头,而是让它像做空者调研仓位那样去执行具体的调研动作。 上述所有维度的雄心,都要求模型拥有可安装软件的计算机环境,以及对各类相关数据的访问权限(无论是通过 MCP 协议连接器还是其他途径)。目前已有大量工具让这成为可能,我非常建议大家配置使用。 历史早有先例 以上就是我对如何变得更有雄心的具体建议。如果你愿意多停留一分钟,我还想聊聊为什么我认为在当下这显得尤为必要。 人类自古以来就常常受困于自己对“可能发生什么”的认知定势——更糟的是,我们通常只在极其狭窄的专业领域内信任自己的能力。众所周知,米开朗基罗当年极不情愿去画西斯廷礼拜堂天顶画——因为在他看来,自己是一名雕塑家,而不是画家。 在调研关于“雄心”的课题时,我发现心理学中对此有一个专门的概念:自我效能感(Self-efficacy)。简而言之,自我效能感高的人更坚信自身的能力,从而往往能在现实中取得更大的成就。研究表明,自我效能感是可以被激发的:洛克(Locke)和拉瑟姆(Latham)深入梳理了长达 35 年的目标设定研究,得出明确结论:被设定具体且高难度目标的人,其表现显著优于那些仅仅被告知‘尽力而为’的人。Stripe 联合创始人帕特里克·科利森(Patrick Collison)长期维护着一个“快速榜单”(Fast list),记录了人类在被赋予宏伟目标和严苛期限时所完成的种种奇迹。其中一些被充分印证的思维策略包括: 通过相信事情可行去攻克难关: 1939 年,乔治·丹齐格(George Dantzig)在加州大学伯克利分校的一堂统计课上迟到了。他把黑板上的两道题抄了下来,以为那是课后作业并按时交了上去——而实际上,那是当时统计学界尚未破解的两大著名世界难题。 目睹他人做成之后去突破极限: 数十年来,人类一直试图打破“4 分钟跑完 1 英里”的极限。罗杰·班尼斯特(Roger Bannister)于 1954 年 5 月首次打破该记录,但他作为这一纪录唯一保持者的时间仅仅维持了 46 天。如果你热爱跑步,会发现类似的突破正在马拉松破 2 小时的征程中上演:塞巴斯蒂安·索维(Sabastian Sawe)在今年 4 月的伦敦马拉松赛中跑出了首个官方破 2 成绩;而在同一场比赛中,亚军也同样突破了 2 小时大关。 通过允许自己失败去成就大事: 大家似乎最喜爱的商业书籍是《身心合一的奇迹力量》(The Inner Game of Tennis,又译网球内心戏),而我上周才发现这本书。该书的核心观点是:如果你拼命去用力,你反而打不出最好的比赛状态。我们工程师或许还记得 Linux 最初的发布宣言:“这只是一项业余爱好,不会搞得像 GNU 那样庞大而专业。” 自我效能感并非新问题,但在今天它变得前所未有地关键。置身于能力每隔数周就会发生飞跃的世界,我们需要一套全新的思维与方法,让我们敢于去做那些曾经未曾考虑过、甚至不敢相信可能办到的事。 如果你发现了更多能激发雄心的新技巧,欢迎随时和我交流! 本文题图来自弗朗西斯科·戈雅(Francisco Goya)的作品《飞行方式》(Modo de volar)。
Hi, I'm Felix. I'm an eng lead working on Claude. I love watching people take flight and do things they never thought possible. We often describe that as "being more ambitious" when using AI, which is tricky advice to follow. I wrote a tutorial: felixrieseberg.com/how-to-be…
88
160
24
802
112,927
Bloomberg Intelligence: Traditional Cloud & AI ROIC SpaceX expected to have the strong ROIC of ~25% by 2030. $AMZN $CRWV $SPCX $NBIS $CRWV
8
6
1
71
10,576
Ryan Wang 🇹🇼 retweeted
Anthropic Q3 numbers from today’s investor meeting, Per Bloomberg reporting. Revenue: $16.8 billion Adjusted gross margin: 85% Adjusted operating margin: 11%+ Free cash flow: Positive Adjusted operating profit: Positive
78
32
21
382
166,730
Ryan Wang 🇹🇼 retweeted
I've been traveling the world a bit. One of the most incredible experiences was getting to visit Taiwan w/ @JensenHuang. It was wild 🤯.. huge crowds of engineers, insanely good street food at night markets, and of course a lot of good conversations across many days. I caught some of it on video, see below. Alright, that's it from me, love you all ❤️ Timestamps: 0:00 - Highlights 0:46 - Intro 3:25 - Computex: crowds, engineers, tech, future of AI 9:12 - Chat with Jensen: keynote prep & nerves 13:03 - Chat with Jensen: Taiwan's tech history 19:47 - Raohe night market 29:37 - Chat with Jensen: memories from childhood 33:44 - NVIDIA party 40:29 - Lunch: braised pig's feet 43:09 - Chat with Jensen: GPUs, AI, and competition 55:07 - Ningxia night market 1:10:14 - Taiwanese breakfast 1:29:54 - Outro
447
1,124
218
12,344
825,178
Ryan Wang 🇹🇼 retweeted
zerohedge
@zerohedgeOct 13
Just how full is the hyperscaler/AI IG debt pipeline: there have been zero deals since the start of September. And SPVs not doing better. As a reminder, debt becomes capex, becomes revenue, becomes EPS. Without debt, there is no EPS growth.
zerohedge
@zerohedgeOct 13
"Until The AI CapEx Debt Creation Machine Turns Back On": Goldman Says Credit Is Calm... For About Two More Months zerohedge.com/markets/until-…
33
131
8
648
160,520
Ryan Wang 🇹🇼 retweeted
Anatomy 应该是这周最火的 Skill 能将复杂的机械结构变成一个风格非常独特的等轴立体动画。 然后我在想,用这个风格做那种科普视频是不是很爽? 于是结合了 Grok bot,还用即梦生成了一些图片和视频, 最后用这个 skill 做了个“芯片是如何制造的”科普视频,讲解得挺清楚的
Introducing Anatomy Describe a machine in one line. Claude brings it to life. - Interactive isometric wireframes you can rotate, inspect and explore - WebGL shaders for fire, water, plasma and more - Real 3D motion, with individual parts that rotate and move Install it as an agent skill. Fully open source.
50
60
7
397
35,010
Ryan Wang 🇹🇼 retweeted
ASIC Update: TPU v9 Timeline & v10 Opportunities • We push out TPU v9 Humufish expectations by one quarter
Core A slipped ~3 weeks → MTK Core B/I/O mods. Sampling still Nov; expect tape-out in Jan 2027, MP tape-out 2Q27, shipments 1Q28. • Constructive on EMIB-T: root causes of substrate yield (45-50%) identified → potential re-acceleration. • Google still undersupplies its own models. Gemini 4 trained on ~200k TPU v7, but heavy external allocation (esp. Anthropic) in 1H26. More internal in 2H26 helps, yet compute gap widens — Anthropic + OpenAI each +5–10 GW in 2027 vs few GW for Google. • Broadcom unchanged: 460k CoWoS wafers in 2027, shipments late 2026.
MTK Zebrafish volumes higher into 2027–28, partially offsetting v9 timing. • TPU v10 (4Q28/1Q29): Icefish (MTK + GUC 3DIC on A14) the only confirmed mainstream SKU. Merope uncertain; Broadcom faces AMD+Marvell competition. EMIB-T momentum rising for v10 • Near-term headwinds for MTK; mid/long-term still like MRVL, MTK (100% v9 + potential new DC customers), and Intel (EMIB-T). #MTK #TPU #Broadcom #INTC #MRVL
1
25
11
219
56,594
Ryan Wang 🇹🇼 retweeted
bloody....last 21 trading days - MOMENTUM is CHANGING as REGIMES seem to be changing 1. top & bottom factors 2. most "unusual factor moves" 3. momentum constituent changes I'm focused this am on what is changing in TMT MOMENTUM... check out the TMT MOMENTUM - NOW vs TMT MOMENTUM (6m ago) Six months ago (4/10): - Optical and networking names led the top of the ladder: AAOI +356%, LITE +158%, VIAV, CIEN, GLW, COHR. - Chip equipment (UCTT, FORM, AEIS, TER) and memory/storage (SNDK +155%, WDC, STX) filled most of the rest. - In Major Theme terms, 9 holdings were Robotics & Automation, which is where this taxonomy puts chip-equipment makers, and only 1 was Cybersecurity. *software left for D-E-A-D Now (10/09): - The top is software and security: TEAM +115%, RNG +89%, OKTA +66%, PAYC +59%, CRWD +46%, then QLYS, TWLO, PANW and FTNT - Cybersecurity went from 1 holding to 6, and 5 of those are in the top 20. Digital Transformation went from 2 to 6 - Robotics & Automation (chip equipment) fell from 9 to 3 - The only AI-hardware names still near the top are HPE (+56%), DELL, LITE (+44%) and MRVL
Absolute destruction in the AI space while SPX is unchanged and VIX futures are actually slightly down! I don’t know what this crazy rotation means, but it is nasty. Bet @JaredKubin has plenty to say on it.
4
9
2
120
23,405
Ryan Wang 🇹🇼 retweeted
玉晶光(3406 TT)Nomura AI 光學互連(Optical Interconnect)需求,特別是被動式光纖連接器與 CPO FAU(Co-Packaged Optics Fiber Array Unit),將在未來 5 年重塑傳統光學鏡頭產業。 台灣手機鏡頭廠(玉晶光、大立光)在高精度塑膠射出、光學設計、主動對位、鍍膜與檢測能力上具備先天優勢,加上無塵室設施與穩定自由現金流,有望切入此新市場並享有較高估值溢價。 新業務機會:被動光纖連接器 + CPO 玉晶光已取得被動光纖連接器 ferrule 訂單,預計 2026 年底或 2027 年初開始出貨(公司未公開評論,渠道檢查所得)。 可供應 CPO FAU 相關零組件(鏡頭、稜鏡、V-groove、FA 等),並有機會承接 FAU 組裝,與 FOCI(3363 TT)等合作供應關鍵元件。 獲利貢獻預估: - 2027F:低於 3%(CPO FAU 約 0.7%) - 2028F:約 16–20%(CPO FAU 約 16.8%) 光纖連接器 ferrule 假設 ASP USD 0.5–1、年出貨 2,000–3,500 萬顆,營收占比約 2–3%,毛利率高於公司平均。 CPO FAU 貢獻基於 NVIDIA Rubin Ultra / Feynman 等平台採用率與市占假設。 為何鏡頭廠具備優勢? 技術演進路徑為:MPO → MMC(US Conec 專利)→ EBO(Expanded Beam Optical,非接觸式擴束連接器)。 EBO 採用精密聚合物 ferrule + 內建 TIR 鏡片,對精度要求極高(常低於 5μm,甚至 3μm)。傳統連接器廠商的塑膠射出能力多在 >10μm,較難達標;而手機鏡頭廠的高精度模具、塑膠射出、光學設計、鍍膜、對位與檢測能力恰好匹配,且已具備無塵室與充裕現金流。 EBO MSA(2026 年 3 月成立)成員包含 Microsoft、Meta、NVIDIA、AMD 等,Microsoft 已在 Azure 採用,顯示產業趨勢明確。 蘋果鏡頭業務展望 - 舜宇光學(Sunny,2382 HK)在 iPhone 18 Pro 可變光圈(VA)主鏡頭出現良率問題,可能間接利好玉晶光(大立光為直接受益者)。 - 預期 iPhone 鏡頭出貨在 2026 年底至 2027 上半年受記憶體成本上升與 Duo 機型分流影響而較慢;2027 下半年隨 20 週年機與第二代 Duo 上市後回溫。 - 進一步展望:2027 下半年 Pro 主鏡頭再升級,2028 下半年 Pro Max 可能導入新潛望鏡頭。 主要風險 1. iPhone 需求不如預期 2. 競爭導致 ASP 壓力 3. 後鏡頭市占率進展不如預期 4. 其他 Android 客戶滲透較慢 5. CPO 與被動連接器實際採用進度、客戶訂單與量產時程仍具不確定性
Genius Electronic Optical (3406 TT), a Taiwanese smartphone lens maker, has likely secured an order for fiber-optic connector ferrules, according to Nomura. The bank’s channel checks point to shipments starting in late 2026 or early 2027. Genius declined to comment. Ferrules hold and align optical fibers inside connectors. Nomura says newer high-density designs require tolerances below 5 micrometers, and sometimes below 3. Traditional connector makers typically mold plastic to tolerances above 10 micrometers, while camera lens makers already work to the tighter tolerances these designs require. Nomura assumes an average selling price of $0.50 to $1 per ferrule and annual shipments of 20 to 35 million units. On that basis, it expects ferrules to account for 2% to 3% of Genius’s revenue in 2027 and 2028. The bank estimates that a rack with 60 to 80 GPUs or AI ASICs would need 600 to 1,000 ferrules if it used fiber connectors for optical scale-up. Nomura sees a bigger opportunity in components for co-packaged optics, including V-grooves and prisms used in fiber array units. It expects these components to contribute less than 1% of Genius’s earnings per share in 2027, rising to 17% in 2028. The bank raised its price target to NT$1,415 from the NT$568 target it set in October 2024.
8
1
77
8,892
Ryan Wang 🇹🇼 retweeted
As industry attention turns to substrates, Japanese and Taiwanese ASP charts look like they're telling completely different stories. Japan pricing is increasing, but is not up nearly as sharply as the pricing data coming out of Taiwan, which has tripled since mid-2025 and is at an all-time high. Same industry. Same AI cycle. So what's going on? (1/5)🧵
13
14
3
95
28,797