J

579 posts

J

J

@Investor2212

Investing for my future. Not Financial Advice

Katılım Aralık 2024
35 Takip Edilen111 Takipçiler
J retweetledi
Macro_Lin | 市场观察员
$INTC Intel切入NVIDIA先进封装供应链的时点可能从2028年Feynman提前到2027年Rubin Ultra,CoWoS独家依赖出现松动。 UBS的具体说法是NVIDIA给Rubin Ultra做产品分级,2-die保留在TSMC CoWoS-L,4-die改用Intel EMIB-T。CoWoS-L做4-reticle级die的良率和interposer尺寸都吃紧,前段时间Commercial Times报Rubin Ultra缩回dual-die,源头就是这个物理瓶颈。 EMIB-T在bridge里加TSV做垂直供电,解决了标准EMIB在HBM4级供电下的电压droop问题。封装尺寸支持到120x180mm,panel利用率约90%,CoWoS interposer的wafer利用率约60%。Bernstein估算EMIB-T单颗封装成本百元级,CoWoS千元级。 NVIDIA此前已确认在Feynman上让Intel承担I/O die(18A或14A),并在EMIB封装上拿约25%份额。Rubin Ultra 4-die SKU如果被确认属实,Intel封装进入NVIDIA供应链的时间表就从2028年提前到2027年。
Jukan@jukan05

UBS: NVIDIA may offer two SKUs for Rubin Ultra: a 2-chip version and a 4-chip version. The 4-chip version will likely use Intel’s EMIB-T. $INTC

中文
2
13
118
52.9K
J
J@Investor2212·
@LogicalThesis Agreed. If you haven't yet, check out the superfan episodes on Peacock. Every episode of each season re-edited and released with like 5-8 minutes of extra footage per episode. Hilarious stuff that wasn't included in the original.
English
1
0
1
48
Logical Thesis
Logical Thesis@LogicalThesis·
If you're ever feeling stressed or down, just put on The Office. Can't tell you how much this show improves my mood. Ultimate comfort zone. Constant smiles and laughs. Great wind down at the end of a tough day/week. Michael Scott is a top 5 tv character of all time.
English
10
1
50
4.1K
J retweetledi
Jukan
Jukan@jukan05·
UBS: NVIDIA may offer two SKUs for Rubin Ultra: a 2-chip version and a 4-chip version. The 4-chip version will likely use Intel’s EMIB-T. $INTC
English
19
33
480
166.4K
J retweetledi
ImNotHarsh | 📈💸
ImNotHarsh | 📈💸@imnotharsh·
BREAKING: Ming-Chi Kuo reports $AAPL Apple has initiated projects at $INTC Intel's 18A-P foundry using Foveros packaging. This will be for low-end/older processors in iPhones (80% of orders), iPads, and Macs, with 2026 testing, 2027 ramp-up, 2028 growth, and likely a shift to 14A and additional products, expected to be ready by 2029. Apple is strategically developing Intel as a capable long-term supplier by simulating full product-line integration, aiming to reduce TSMC dependency amid AI-driven capacity constraints and maintain bargaining power. Intel gains a high-stakes opportunity to rebuild its foundry business with Apple's demanding volumes and standards.
ImNotHarsh | 📈💸 tweet media
English
8
48
272
26.3K
J
J@Investor2212·
$msft only thing red on my watchlist rn... Need this thing to do something other than just red day after red day.
English
1
0
2
232
PeterLamberg
PeterLamberg@PeterLamberg·
$MSFT Bad news: this didn’t recover as other tickers did in the afternoon. Good news: it didn’t drop further and is still in the same bull-flag-ish pattern. On your lucky day, this can reach 435-ish and give a 1500% gain. Not impossible at all. 🤞🤞🤞🤞
English
1
0
3
348
J retweetledi
Macro_Lin | 市场观察员
聊一下Intel的EMIB-T,这个技术可能在未来两年改变先进封装的竞争格局。 先说背景。AI加速器的封装面积越来越大,一颗芯片装不下的计算和内存,需要把多颗die封装在一起协同工作。当封装面积超过单次光刻曝光的极限,就需要用”局部硅桥”来连接相邻的die。这个思路,TSMC和Intel各自走出了不同的实现路径。 TSMC的CoWoS-L,L代表LSI(Local Silicon Interconnect),用局部硅桥做相邻die之间的高密度互连,全局布线靠InFO RDL扇出层完成,整个工艺在圆形晶圆载板上加工。Intel的EMIB,同样是局部硅桥,但嵌入的是矩形有机基板,全局布线靠基板上的金属层完成。两者底层思路相近,但制造平台和全局布线的密度存在本质差异。 早期EMIB有一个结构性短板。硅桥里没有垂直通孔,电源从基板底部送不穿桥,只能绕路。偏偏最需要电的PHY电路就坐在桥的正上方(信号线越短越好,所以PHY放在die最边缘),电源反而要从die中心横向传过来。路径长,压降大,瞬态响应差。HBM3e时代还能应对,到HBM4接口宽度翻倍、pin速率推到6.4Gbps以上,这条供电路径就成了瓶颈。 EMIB-T就是解这个问题的。T代表TSV,在桥里打铜通孔,电源从基板直接垂直穿过桥送到上面的PHY。同时桥里集成MIM电容和接地铜网格,就地稳压滤噪。供电路径从横向绕行变成垂直直连。 这个改动释放了两层约束。第一层是供电能力,可以支撑HBM4/4e级别的功耗需求。第二层是封装面积上限,Intel目标是2026年做到120×120mm封装、集成8倍光罩面积的硅,2028年推进到120×180mm、超过12倍光罩。 跟CoWoS-L的竞争,核心差异在制造平台。CoWoS-L在圆形晶圆载板上完成全部工艺,面积越大,晶圆边缘的浪费越大,RDL层的均匀性和翘曲控制难度也非线性上升。EMIB-T在矩形有机基板上多放桥,每个桥的工艺独立重复,整体良率虽然也随桥数量指数衰减,但衰减速率稳定可控。封装越大,EMIB-T在成本上的优势越明显,良率的可控性也越高。反过来说,CoWoS-L的InFO RDL层提供了更高密度的全局布线和电源分配能力,加上NVIDIA等客户的设计生态已经深度适配CoWoS工艺,迁移成本极高,这是CoWoS短期内不可替代的护城河。 目前的状态是,标准EMIB量产良率90%,Intel表示已接近传统FCBGA封装水平。EMIB-T验证良率也到了90%,但这是小批量数据,量产目标98%。郭明錤的评价很准确,从90%到98%比从零到90%难得多。2026年EMIB-T进产线验证,2027年可能出现第一批产品(Jaguar Shores),Google TPU等外部客户也在评估EMIB-T方案,2028年才是真正的放量年。当然,Intel近几年的产品节奏有过多次延期,这个时间线能否如期兑现本身就是一个需要追踪的变量。 对TSMC的影响,我的判断是分流,不是替代。NVIDIA跟TSMC的全栈绑定短期内动不了,但Google、Meta这些做自研AI ASIC的客户,以及Apple等寻求封装供应链多元化的厂商,封装面积需求大、成本敏感、又有分散供应链的动力,EMIB-T给了它们一个真实可用的备选。Bernstein估算,EMIB平台封装每颗芯片成本在几百美元量级,而Rubin级别加速器上CoWoS-L封装成本约900到1000美元。即便只从边缘蚕食,打破CoWoS垄断定价权这件事本身,对整个产业链的影响也远超实际转移的份额。 SK hynix这两天也传出在跟Intel合作研发EMIB封装。HBM供应商主动拥抱EMIB生态,说明连它们自己都不想被CoWoS产能卡住出货节奏。 Intel的EMIB-T,正在把先进封装的竞争从”谁的技术更好”转向”谁能在超大面积上同时控住成本和良率”。从目前的市场叙事来看,EMIB-T的结构性优势应该还没被充分讨论。
Macro_Lin | 市场观察员 tweet media
中文
68
69
350
44.6K
J retweetledi
Intel
Intel@intel·
We're thrilled to partner with @Google on something we've been building with them - Googlebook. Premium, powerful devices designed for Intelligence. We can't wait to get it into your hands this fall. Learn more at ms.spr.ly/6013vuzjt #Googlebook #NEXT #Intel
Intel tweet media
English
110
398
2.4K
514K
J retweetledi
bwatts 
bwatts @_bwatts_·
Dark pool activity worth watching from today’s session 👀 $SPY saw a $737–738 cluster, 3 prints, $4.43B combined notional. $QQQ printing a $707 cluster, 2 prints, $916.76M combined. Standalone print at $713 adds another $516M. $MSFT printing $4.22B at 88.71% off-exchange volume. MASSIVE print for Microsoft. $GOOGL $SNDK $META all printing 65%+ dark pool volume. $AAPL $AMZN $MU $TSLA rounding out the notable prints. These levels are best used as support / resistance 🤝 Data from QuantData. Custom dashboard. Full breakdown 👇
bwatts  tweet media
English
3
5
32
3.6K
J retweetledi
ImNotHarsh | 📈💸
ImNotHarsh | 📈💸@imnotharsh·
This is bullish $MSFT. Microsoft’s stake in Anthropic and continued stake in OpenAI gives it the hedge it needed. Azure is still the primary cloud provider, unless and until it cannot meet capacity demands. Microsoft owns ~27% of OpenAI’s for-profit entity (valued at ~$135B at the last recap). By removing the uncapped drag on OpenAI’s finances, the deal gives OpenAI more flexibility to raise capital, partner more broadly, and potentially IPO, which directly boosts the value of Microsoft’s stake. Satya is smart. He knows what he is doing.
Shay Boloor@StockSavvyShay

OpenAI reportedly capped $MSFT total revenue share at $38B under its revised deal. That could give OpenAI more flexibility to work with $AMZN and $GOOGL as it builds toward a potential IPO.

English
4
5
105
26.8K
J retweetledi
Jukan
Jukan@jukan05·
I didn’t realize this, but it looks like AWS may be using EMIB for a lower-end version of Trainium through MediaTek. Similar to the TPU v9 program handled by MediaTek, next-generation Trainium seems likely to use both EMIB and CoWoS. $INTC
English
18
41
536
113.1K
J retweetledi
ImNotHarsh | 📈💸
ImNotHarsh | 📈💸@imnotharsh·
$INTC Intel Foundry's EMIB tech may be used by $AMZN Amazon for lower-end Trainium chips as well. Intel's partnership with MediaTek, which I have highlighted before is already the link between $GOOGL Google and Intel for TPU v9e, but seemingly also for AWS. Intel is embedded within the hyperscaler network.
Jukan@jukan05

I didn’t realize this, but it looks like AWS may be using EMIB for a lower-end version of Trainium through MediaTek. Similar to the TPU v9 program handled by MediaTek, next-generation Trainium seems likely to use both EMIB and CoWoS. $INTC

English
1
9
61
7.8K
J
J@Investor2212·
Massive dark pool buying in $MSFT !
Alphatica@alphaticaio

DARK POOL UPDATE | May 11, 2026 $19.4 billion in off-exchange flow. 744 prints ≥$5M. Net: +$1.84B buy. Friday the dark pools said -$2.7B net sell. Today they flipped to +$1.84B net buy. Dark Pool Buyers: $MSFT +$2.12B (28 prints) $MU +$1.13B (53 prints) $CVX +$890M (6 prints, NEW on the dark pool) $AAPL +$876M (57 prints) $TSLA +$335M (35 prints) Dark Pool Sellers: $NVDA -$1.55B (49 prints) $QQQ -$855M (43 prints) $SPY -$732M (37 prints) Where the two tapes agree: $MSFT: +$6.73B on the lit tape. +$2.12B in the dark pools. Both tapes say buy. The Microsoft reversal is confirmed across every venue. Last week $MSFT was being sold on both tapes. Today both flipped. $NVDA: -$8.02B on the lit tape. -$1.55B in the dark pools. Both tapes say sell. The $NVDA distribution is happening everywhere. No venue is showing buyer interest. $MU: +$2.99B on the lit tape. +$1.13B in the dark pools. Both tapes say buy. The memory trade is being accumulated across both venues. Where the two tapes disagree: $SPY: +$1.74B on the lit tape. -$732M in the dark pools. Bought in public. Sold in the dark. Second straight session with this pattern. $QQQ: +$1.49B on the lit tape. -$855M in the dark pools. Same thing. Bought in public. Sold in the dark. The index divergence is now a pattern. Two consecutive sessions where $SPY and $QQQ are being bought on the lit exchanges and sold off-exchange. The dark pool is the shadow tape. When the two disagree on the indexes for multiple sessions, one of them is wrong. $CVX appeared in the dark pools for the first time. +$890M across just 6 prints. That's $148M average per print. The Iran deal rejection is driving oil-related dark pool flow. *When a name shows up in the dark pool at $148M per print, that's not a hedge fund. That's a sovereign wealth fund or a pension rebalancing. The lit tape shows you what the institutions are doing in public. The dark pool shows you what they're doing where they think nobody is watching. Now you're watching.

English
0
0
0
133
J retweetledi
ImNotHarsh | 📈💸
ImNotHarsh | 📈💸@imnotharsh·
$INTC Intel Foundry may have just won a portion of AI6 manufacturing, on top of the AI7 and beyond roadmap for Terafab $TSLA, originally expected to be fabricated by $TSM. “…under strong pressure and encouragement from the Trump administration, the portion of AI6 that was originally assigned to TSMC will reportedly be transferred entirely to Intel Foundry.”
Jukan@jukan05

Interesting. Tesla’s current AI5 chip, codenamed Helios, is reportedly being foundried by TSMC and Samsung, and the next-generation AI6 was also originally expected to be split between those two foundries. However, under strong pressure and encouragement from the Trump administration, the portion of AI6 that was originally assigned to TSMC will reportedly be transferred entirely to Intel Foundry. $INTC

English
6
14
94
16.9K
J
J@Investor2212·
@tradertvshawn I would not be shorting $INTC... Stock has been absolutely on fire. News out today on partnership with SK Hynix & their adoption of intel's EMIB tech.
English
0
0
1
95
J retweetledi
J retweetledi
ImNotHarsh | 📈💸
ImNotHarsh | 📈💸@imnotharsh·
NEWS: SK Hynix is reportedly in R&D with $INTC Intel on 2.5D packaging, specifically testing Intel’s EMIB technology to integrate HBM memory with logic chips for AI accelerators. EMIB serves as an alternative to TSM’s CoWoS, potentially allowing SK Hynix to expand beyond memory supply into turnkey HBM solutions amid high AI demand. The collaboration bolsters Intel’s foundry packaging ambitions and fits SK Hynix’s strategy for U.S.-based 2.5D facilities to enhance supply chain resilience.
ImNotHarsh | 📈💸 tweet media
English
6
19
102
15.6K