@Mojo_flyin @intel Given this development trend, it is highly likely that Intel's factory will become an independent entity, providing services to the entire domestic semiconductor industry in the United States.
English
techQuicker
651 posts
@OutspokenGeek AMD has no its own manufacturing facilities and relies entirely on TSMC, which is its biggest vulnerability. This dependence runs counter to U.S. national strategy, and the associated risks are evident.
English
Do not forget that $AMD isn't sitting still and the Venice 2nm Zen 6 server variant will even things out. $INTC is so messed up in its server roadmap that it's pointless to even compare till they get back on track. Yes $INTC is going to suck till the modern cores are available.
English
Turns out Vera beating x86 thoroughly was complete BS.
Actual real world (if chaperoned) benchmarks are available from @phoronix whom I respect a lot -
phoronix.com/review/nvidia-…
While there isn't a close equivalent, 2x EPIC 9455 is reasonable proxy and Vera is only ~1.2x in perf.
OutspokenGeek@OutspokenGeek
@jukan05 Also, whatever the heck does "1.5x faster, 2x perf" mean? Makes me question the person's basic understanding of benchmarking.
English
@jukan05 When the first deepseek moment occurred, Nvidia was unfairly sold off, and I bought Nvidia.
English
@treasureh8nter It is estimated that this trip to Taiwan aims to guide the upstream and downstream supply chain to set up factories in the United States.
English
What happened to all the $INTC investors pushing for “Made in America” and saying Taiwan is a geopolitical risk?!?! 🤭🤭🤭
Lip Bu Tan is meeting with KEY SUPPLIERS in Taiwan 🇹🇼 ahead of Computex this year.
Why does Intel have key suppliers in Taiwan?!?! What happened to the geopolitical risk? 🤣🤣🤣
“Intel CEO Is Set to Be in Taiwan This Weekend, Focusing On Meeting With Key Suppliers Including TSMC”
Muhammad Zuhair@mzuhair123
$INTC CEO Is Set to Be in Taiwan This Weekend, Focusing On Meeting With Key Suppliers Including TSMC "On the eve of COMPUTEX 2026, the world's three AI chip giants have gathered in Taiwan. Following visits to Taiwan by AMD CEO Lisa Su and NVIDIA CEO Jensen Huang, Intel CEO Lip-Bu Tan will also arrive in Taiwan this weekend to exchange views with senior executives in the supply chain. Jensen Huang, who is still in Taiwan, is estimated to have stayed for more than 10 days, meeting with senior executives of the Taiwanese supply chain. Industry insiders also revealed that Chen Liwu is expected to arrive in Taiwan this weekend. In addition to internal meetings and a "tentative" meeting with senior executives of TSMC, he also has three secret meetings scheduled for the Taiwan-based supply chain project. On June 1, Intel is expected to host a supply chain cocktail party in the evening, inviting partners who have collaborated with Intel in the upstream and downstream sectors for many years. On June 2nd, a keynote speech at COMPUTEX will be delivered in the afternoon. In the evening, a closed-door exchange will be held with executives from companies such as the "Five Kings of Electronics," ASUS, and Advantech to discuss their plans and prospects for AI servers, PCs, and American manufacturing. It is understood that, in addition to geopolitical factors, the demand for AI is simply too great, prompting customers to actively seek secondary sources to diversify risk. In fact, what truly attracts customers to Intel is not just its technology, but also its overall supply capabilities in CPUs, ASICs, advanced packaging and processes, and local manufacturing in the United States." $TSM
English
EFB（Elevated Fan-out Bridge）是 AMD 自研的工业界首创的多芯片 GPU（Multi-die GPU）2.5D封装设计.
EFB是与TSMC CoWos-L(B200/B300)相当的技术, 它们本质上都是通过“RDL（重布线层）+ 局部硅桥（Silicon Bridge）”的组合，在便宜的有机基板上实现局部的超高速互连。
封装厂（如日月光）拿到的是已经做好的、表面平整的普通有机基板。接着，直接在基板表面的特定位置放上硅桥（此时硅桥凸出来了）。然后，利用扇出（Fan-out）技术，在凸起的硅桥和周围的空隙之上铺设重布线层（RDL）。最后，主芯片跨坐上去，焊接在 RDL 表面。此时的RDL层, 线宽及密度要求大大缩小, 很老一代的光刻机就可以完成, 要求比较高的使用stepper来定位.
TSMC生产了die, 硅桥, 与CoWoS-L不同的是, 最后日月光完成基板+硅桥+RDL以及后端封装.
更具颠覆性的是，AMD 已经联合力成（PTI）通过了行业首个基于方型大面板的 2.5D Panel-based EFB 认证。
最近一次AMD 10B投资台湾, 其中一部分就是EFB工艺的投资, EFB工艺缓解了TSMC CoWoS的紧张局面, 让AMD的产能得到更大的发挥空间.
AMD 1000 in 2026.
中文
@ashishtuli Correct. If AMD do not establish its own robust supply chain now, AMD will be dependent on others, especially by overly relying on Taiwan's industrial chain, which could become a fatal problem in the coming years.
English
“Fabless” used to mean asset-light.
Now it increasingly means prepaying, co-investing, and locking up years of wafer and advanced packaging capacity just to secure supply.
You may not own the fab, but you are still paying to make sure it exists for you.
The label is still “fabless.” But it is not truly asset-light anymore.
AMD@AMD
Today, we announced more than $10B in investment across Taiwan’s ecosystem to scale advanced packaging and accelerate next-gen AI infrastructure, from 6th Gen EPYC CPUs codenamed “Venice” to our Helios rack-scale platform including Instinct MI450X GPUs, with multi-gigawatt deployments beginning in 2H 2026. Additionally, AMD and TSMC have hit another major production milestone, with Venice EPYC CPUs ramping on TSMC 2nm technology in Taiwan with future plans to ramp production at TSMC’s Arizona Fab. More on the news: bit.ly/4tJrUkR
English
@dnystedt It seems that AMD is taking a reverse approach; while everyone else is moving their supply chains back to the United States, AMD is going in the opposite direction. Is this a gamble?
English
AMD Helios server racks with new Instinct MI450X GPUs will be out in the 2nd half of this year, media report, with help from Taiwan partners including Wiwynn, Wistron, and Inventec, with Sanmina involved in co-developing the racks. $AMD $SANM #Wiwynn #Wistron #Inventec money.udn.com/money/story/56…
English
这是一张公司从成立到IPO的累计净亏损图。
SpaceX到目前为止，亏损了370亿美元，比第二名滴滴打车的124亿的三倍还多，也比Uber、Rivian、Airbnb三家加起来的亏损总和还大。
中文
1999年先涨CPU，后来涨路由器，最后涨光纤——AI正在走同一条路
科技牛市从来不是一次性定价的。
1990年代那轮，市场花了整整十年，不断发现下一层瓶颈，不断重新定价。每一次定价，都是一轮新的机会。
现在的AI，正在走同一条路。
1990-1995：PC时代
市场逻辑很简单：每个人都会拥有电脑。
最强的是Intel、Microsoft、Dell。CPU就是整个科技产业的核心。
对应今天：NVDA第一阶段，2023-2024年的故事。
1995-1997：互联网商业化，市场发现电脑需要连接
Windows 95、Netscape、AOL出现后，市场意识到：原来电脑之间要连接。
网络设备需求爆炸，Cisco、Lucent、Nortel开始上涨。
瓶颈从CPU转向router、switch、network infrastructure。
对应今天：AI从GPU转向networking、scale-out、interconnect。这一步正在发生。ANET、MRVL、ALAB、CRDO。
1998-1999：市场发现真正的瓶颈是带宽
互联网流量爆炸，大家突然发现：数据根本传不动。
于是全世界开始疯狂铺fiber、DWDM、optical transport。
最疯狂的阶段来了——JDS Uniphase、Corning、Ciena、Juniper开始parabolic move。
对应今天：optical、CPO、SerDes、DSP、scale-up networking。市场开始意识到真正的瓶颈不是compute，而是data movement。
LITE、COHR、CIEN和NOK。这一层正在被重新定价。
1999后期：即使有网络，还是缺数据中心
市场再一次发现：physical infrastructure不够。
Equinix、Exodus、Global Crossing开始最后一轮疯狂。
市场从数字世界重新回到物理世界。
对应今天：电力、冷却、数据中心、并网。算力工厂需要的不只是芯片，还需要土地、电力、物理空间。
2000最后疯狂：应用层
基础设施都炒完后，万物互联网化。Yahoo、Amazon、Qualcomm、eBay最后疯狂。
市场逻辑变成：互联网会改变一切。
对应今天：AI agents、robotics、physical AI、autonomous economy。这一层还早，但会是最后最疯狂的阶段。
映射到今天的时间线。
2023-2024：GPU/Training → NVDA、AMD
2025：Networking/Interconnect awareness → ALAB、MRVL、ANET、CRDO
2026：Optical + Scale-out + Power bottleneck → LITE、COHR、IREN、VRT、power infra
2027+：AI Application/Agents/Robotics/Physical AI → software + embodied AI + autonomous economy
还有些重要的部分比如说存储之类的，没提到不代表不重要。
这也是我过去几个月推荐顺序背后的逻辑。
第一步推了连接层——MRVL、ALAB。AI瓶颈从算力转向互联，这两个卡在最难绕开的位置。
第二步推了基础设施层——IREN、Oracle。物理资产开始被重新定价，算力工厂需要电力和数据中心。
第三步是应用层——CRM、NOW、MSFT、SNOW。AI基础设施建好之后，真正的变现层是企业软件。这一层现在筹码最干净，叙事正在反转。
TSLA如果你把它看成physical AI的代表，它属于第四阶段最后爆发的那类资产。
历史上最夸张的阶段，往往发生在市场意识到：原来整个基础设施都不够的那一刻。1999年的答案是光纤。2026年的答案可能是：电力、连接、和企业软件。每一层瓶颈被发现，都是一轮新的机会。
#AI基础设施 #光互连 #半导体 #数据中心
中文
anthropic 估值超过 OpenAI 真是大大被高估了。
OpenAI 眼光很好，提前锁了不少算力，相比起来 anthropic 的算力根本不够。
而且 anthropic 很反华，从短期看反华排除了中国 ip，缓解了算力压力，但从长期看放弃中国市场并不明智。
anthropic 的工程师文化过重，虽然很多码农喜欢这种腔调，但这并不是什么好事。
codex 走生态的路线绝对是高明的，我认为不管 ai 这个行业发展的怎么样， 多 OpenAI 空 anthropic 一点问题没有。
不过现在这俩公司都没上市，没有太好的多空手段，或许可以考虑去 polymarket 上买是否。
中文
Agentic AI时代，CPU再次站到台前。Tenstorrent的Jim Keller和那颗Ascalon RISC-V 高性能CPU很值钱，花落谁家，谁家股价涨一波。如果高通拿下来，服务器级CPU的那个逻辑还有人怀疑吗？不会的。
中文
@imnotharsh This is likely related to Terrafab's technical framework.
English
I just don't think $INTC Intel ends up spending that cash on such a high risk bet. Prob better they don't for a more secure roadmap. The balance sheet will be in tremendous scrutiny I would imagine.
That said, also would not be all that up in arms if they do. Signals some aggressive bets from Intel.
If they didn't go heavy after SambaNova, don't think they will do it here.
Maybe IP licensing, co-design, or some level of foundry use at best.
Jukan@jukan05
Personally, I think this is a must-win deal for both Intel and Qualcomm. Whoever gets it will save an enormous amount of time. I hope Qualcomm ends up buying it.
English
「18 個月後台灣可能不再重要。」
出現在我螢幕上的爆論！誰說的？
這是個從工程師轉行做 VC 的人，突然在 podcast 裡說：「18 個月後台灣可能不再重要。」
這根本不是在談台灣會不會被取代。
他叫 Chamath。
以前寫 code，後來發現講故事比蓋廠還能讓錢快點成長。
他 2017 年投 Groq，一家美國做 AI inference 的公司。2025 年 NVIDIA 給了 200 億美元的授權，他套現大概 30-40 億。這家公司做的東西，本來就是要挑戰 NVIDIA 跟 TSMC 那條供應鏈的。
所以他現在講「美國 1-2nm 快自給了、Neuralink 那種機械臂會改變一切」，其實不是在預測晶片未來。
他是在保護自己手上的牌。
他現在在募 30 億新基金，LP 裡不少是 Trump 那邊的錢。他公開支持關稅、支持美國製造。這句話一出來，新基金的募資故事就順了。
他還有家 8090，用 AI 幫企業把老軟體重寫。他這東西很吃算力。如果市場相信「美國自己會有晶片」，他的故事就好賣。
反方聲音其實很直接。
有人說第一島鏈、日本航線不是 18 個月能搬走的。
有人說蓋一座先進廠要 10-15 年，不是蓋倉庫。
有人說台灣不只是 TSMC，還有整條 ODM、伺服器、封裝的生態，斷掉的話全球硬體供應會直接卡住。
有人直接說：「你把 Neuralink 的機械臂拿來比晶片製造，真的不懂半導體。」
這些不是情緒，是現實。
真正有趣的是，他根本不需要台灣股價真的崩。
他只要讓市場「開始相信」台灣可能被取代就夠了。
因為一旦這個敘事站住，Groq 的估值會被重新定價、他的新基金更容易募到錢、8090 的 compute 成本故事也會變好聽。
這就是典型的投機者打法。
不押單一股票，押「故事會怎麼改變資本流向」。
所以下次再聽到這種話，先別急著問他說得對不對。
先問：他現在在做什麼？他正在保哪一塊？
答案通常比他說的還清楚。
你覺得呢？
The All-In Podcast@theallinpod
Chamath: Taiwan Loses Its Strategic Importance in 18 Months @chamath: “ We're 18 months from Taiwan not being an important moment of conversation the way it is today. Why 18 months? Because we are at a point where we're probably 1-2 nanometers away from being able to do what we need Taiwan to strategically do for us. And so as we scale up our chip fabs, as we get more capacity, and interestingly, there are these orthogonal technologies being developed. I don't know if you guys saw, but Neuralink was showcasing a machine that is literally operating at the almost nanometer scale to do the brain operations for the implantation, all automatically. When you have the dexterity and the capability mechanically to make these things, the real reason then is a very different one than what it is today. Today, it's economic. And if you take that off the table, I think we'll have a very different attitude to Taiwan.”
中文
特朗普强调 如果中国真的入侵台湾，美国很可能帮不上忙。
以及认为台湾将美国的芯片工业一滴不剩的“偷走”了。并且似乎在假设其可能放任中国攻取台湾，强烈要求所有芯片公司要在此之前搬迁到美国，否则后果自负。
$TSM 他这样说可以把台积电砸得更低一点吗？
说实话，台湾芯片工业都搬走了，美国才真的不会再保护台湾了。
Clash Report@clashreport
Trump on Taiwan: When you look at the odds, China is a very, very powerful, big country. That’s a very small island. Think of it; it’s 59 miles away. We’re 9,500 miles away. That’s a little bit of a difficult problem. If you look at the history, Taiwan was developed because we had presidents that didn’t know what the hell they were doing. They stole our chip industry.
中文
@imnotharsh By 2028, domestic chip production reaching over 40% is Trump's goal. Let's estimate Intel's production volume and market capitalization by then.
English
This is the one thing people don’t get about $INTC. One quarter flips all FCF & P/E related bearish sentiment.
The balance sheet by end of this year will flip the equation quick, thanks to multiple Foundry wins. $AAPL, $TSLA / $SPCX SpaceX, $NVDA, soon to be confirmed $GOOGL, and likely $META and $AMZN on the roadmap.
The re-rating is not complete.
Nothing has even happened yet.
Long. Live. Intel.
bubble boi@bubbleboi
On the thread guy podcast I said if Intel only does the Feynman I/O die SKU they are making 4.8B in packaging revenue… for one product. I estimated also if they did all of Feynman (including compute die) that it could be between 40-80B in revenue. I did NOT have Vera Rubin Ultra in my model but it makes sense. Intel w/ EMIB is the only way on earth you can pull off 4 reticle sized dies in one package. Huuuuuge vindication.
English
就增量以及预期差而言:
2026 / 2027 是ASIC 和 CPU之年.
对应的是 $AVGO, $AMD $ARM, 其市场份额以及总量双份增长.
#INTC 没有double beat: 它的份额可能会降低.
Sean@sean_________
Mizuho $NVDA/CoWoS slides - $AVGO: (1/3)
中文
@mzuhair123 This reflects the mindset of these Taiwanese companies, which still cling to the illusion of keeping advanced technology confined to Taiwan.
English
Pegatron's Chairman Says that Taiwan Shouldn't Worry With All the Spotlight Going Towards $INTC, As It Takes off Pressure From TSMC
Intel is rumored to be manufacturing some of Apple's orders, while SK Hynix is also reportedly considering partnering with Intel due to TSMC's tight capacity. In response, Pegatron Chairman Tung Tzu-hsien stated that, from an industry perspective, the global market and order volume are large enough to accommodate competition from various companies. There's no need to create panic or euphoria over one or two order rumors; TSMC still holds the majority of Apple's orders.
Tung Tzu-hsien pointed out that stronger competitors don't necessarily mean a downturn for TSMC. When Intel was very strong, TSMC was doing just fine, so there's no need to worry excessively about TSMC just because Intel is stronger. TSMC is like a Michelin three-star restaurant; even if someone doesn't eat at a three-star restaurant today and switches to a two- or one-star restaurant, it doesn't mean the three-star restaurant has lost its value. Intel was almost stripped of its stars; now we need to see if it can return to its previous level.
Tung Tzu-hsien further explained that the world is big enough to accommodate two companies. TSMC's current focus is on fulfilling its orders, so the real issues it needs to address are the talent, electricity, and succession planning required after the factory expansion.
Tung Tzu-hsien said that, in short, it will not be easy for the United States to rebuild its precision manufacturing capabilities, because over the years, a large number of related talents and supply chains have flowed to TSMC and the Asian semiconductor industry chain. If the United States wants to rebuild its complete wafer manufacturing capabilities, "it will also have to work hard."As Tung Tzu-hsien previously mentioned, the AI competition is fierce, and Taiwan has always been an arsenal for tech giants in the AI race.
Take TSMC as an example; TSMC also receives orders from major companies such as AMD, Intel, and Tesla. Intel is also a competitor of TSMC, but it still places orders with TSMC out of necessity. In short, Taiwan's supply chain can participate in most of the competition among these tech giants, playing a rather unique role as a "neutral arsenal."
English
思来想去，想来思去，纠结了好久，还是把所有的子弹给了intc。
原因1:不想太分散仓位。集中仓位才能有超额收益。
2.存储我估计我是拿不住的，买了估计也是赚个鸡腿就卖飞了，我对存储的认知严重不足。
Buy the dip，长期持有，Discord在个人资料。
美股仙人@hanking66
非常纠结，到底是继续加intc好还是建仓dram好？🤔🤔幸福的烦恼 Buy the dip，长期持有，Discord在个人资料。
中文