都叫我TANG老师

最近 #GLW 康宁这个股票也是走的很强，今天我们从 #GLW 康宁护城河，在AI时代的卡位和风险都分解下
康宁 Corning（GLW）确实有比较强的护城河，但要分清楚：它不是 Nvidia / ASML 那种“单一核心产品近乎卡脖子”的护城河，而是材料科学 + 工艺know-how + 客户共研 + 大规模制造 + 长认证周期叠出来的护城河。
在 AI 时代，康宁的卡位明显变好了，核心是：AI 数据中心从“算力瓶颈”逐渐走向“数据搬运瓶颈”，光纤/光连接/光子封装的重要性上升。
康宁强在哪里？
康宁最核心的能力不是“做玻璃”这么简单，而是： 特殊玻璃和陶瓷材料配方 光纤、显示玻璃、Gorilla Glass、汽车尾气陶瓷载体，都依赖材料配方和长期工艺积累。
超高一致性量产 光纤不是实验室能拉出来就行，要低损耗、低缺陷、大批量、长期稳定。这类工艺靠几十年数据和设备调校，不容易快速复制。
客户共研模式 康宁历史上和 Apple 做 Gorilla Glass，现在和 Meta、Nvidia、Amazon 等做 AI 数据中心光连接。
这种模式一旦进入客户架构，后续黏性很强。长认证周期数据中心、通信、汽车、显示这些行业都不喜欢频繁换材料供应商。因为材料一旦出问题，损失很大。
美国本土制造优势 现在 AI 基建有供应链安全和本土化要求。康宁在美国扩产，正好符合 Nvidia、Meta 等大厂需求。
AI 中的卡位是什么 AI 数据中心里，GPU 很重要，但 GPU 之间要通信。模型越大、集群越大，数据搬运越痛苦。
以前很多短距离连接可以用铜线，但 AI 集群规模变大后，铜会遇到问题：距离稍长损耗大发热高功耗高电磁干扰线缆粗，影响散热和布线
光纤用光传数据，优势是：带宽高低损耗低延迟更适合长距离和高密度布线更适合大型 AI 数据中心 它不是直接卖 GPU，也不是卖 HBM，而是卖 AI 数据中心的“光连接底座”。
#GLW 和 AI 的关系有多深？
可以分三层： 第一层：确定性最高，光纤/光缆/连接器。 AI 数据中心越大，光连接需求越高。康宁已经拿到 Meta、Nvidia 等客户订单，这是当前最实在的收入来源。
第二层：中期机会，硅光和 CPO。 CPO 是把光学组件更靠近交换芯片/计算芯片，降低功耗、提升带宽。如果 AI 网络从传统可插拔光模块进一步走向共封装光学，康宁有机会往价值链上游走。
第三层：更长期，玻璃基板/先进封装/光电融合。 未来芯片封装可能更多用玻璃基板、光互连。康宁在玻璃材料上有天然优势，但这部分商业化节奏还没那么确定。
最后主要风险 股价已经反映很多 AI 预期，2026 年康宁股价涨幅很大，市场已经把 AI 光连接故事打进去不少。
后面如果订单节奏低于预期，波动会很大。 客户集中 Meta、Nvidia、Amazon 这类大客户订单好处是规模大，坏处是议价强、节奏集中。 AI capex 周期风险 如果云厂削减 AI 数据中心资本开支，康宁会受影响。光纤不是全链条最高利润环节 AI 光模块、交换芯片、CPO 里面有很多玩家。康宁如果只卖基础材料和线缆，利润弹性不一定等同于 Nvidia。 传统业务拖累 康宁还有显示玻璃、汽车环保、手机玻璃、生命科学等业务。这些不全是 AI 高增长资产。
我的结论 康宁是一个AI 基建链里比较优质的“卖铲子”公司，回调都值得建仓 ！

Yesterday, ahead of Micron's earnings report, I suggested that after the report drops, we could build a position at the mid-Bollinger Band during the dip, and sure enough, it hit the mid-Band before launching a strong 20% rebound.
#MU After hours on Wall Street, Micron's earnings were absolutely explosive. They locked in over 20% of future capacity for the next 3-5 years with long-term contracts, securing the gross margin and alleviating fears around cyclical stocks. I think the market will start to reassess this sector. Last night, the earnings report coincidentally dipped right to the mid-Bollinger Band, and it's currently rebounded by 20 points in after-hours trading, around the 1200 price level.
When I took apart an AI server, I saw the two most valuable components: one is the #GPU and the other is the memory. From a tech perspective, the GPU is the engine, while the memory is the fuel tank.

#SK海力士 Official decision to issue NASDAQ ADR: 17.9 million new shares, raising capital up to 45.5 trillion KRW
▶️ ADR Issuance Overview Board Resolution Date: June 24, 2026 (All six external directors present) Issuance Type: New Stock DR (Third-party placement method) Listing Exchange: NASDAQ (NASDAQ Global Select Market) Expected Listing Date: July 10, 2026 Expected New Stock Listing Date: July 29, 2026 Subscription and Payment Date: July 14, 2026
▶️ Issuance Scale Maximum New Stock Issuance Cap: 17.9 million shares Reference Issuance Price: 2.555 million KRW per common share (based on June 23 closing price) Total DR Issuance Amount (Reference): Approximately 45.4534 trillion KRW Conversion Ratio for Each DR to Base Shares: 0.1 shares (1 DR = 0.1 common shares) Final issuance price will be confirmed after book building

#AI With things heating up like this, I bet many folks still don't get how it works behind the scenes.
Why when we type a piece of text does AI generate corresponding results?
Let’s break down the complete flow from input to output, which is what we call the inference stage, because that's what's really happening when we use AI daily. Here’s an overview of the whole process (7 major steps).
You input a question (Prompt)
Tokenization
Embedding
Transformer layer processing (core computation)
Output logits → sample the next word
Repeat step 5 until generation completes

From the data breakdown, let's see which component is the most valuable. Below, we analyze the latest GB200 setup with 72 NVIDIA cards, breaking it down into parts to assess the value of each.
Currently, a complete NVIDIA GB200 setup with 72 cards is approximately 20-30 million RMB. Let's take the midpoint, calculating at 25 million RMB; this is the composition of the GB200 setup with 72 cards.
72 B200 GPUs, 36 Grace CPUs, 36 GB200 Superchip compute nodes, liquid-cooled rack with NVLink/NVSwitch interconnects, and external InfiniBand/Ethernet networking.
B200 GPU compute chip/package, excluding HBM, estimated at 8 million, accounting for 32%, with 72 B200 cores holding significant value, including GPU die, advanced packaging, substrate, etc.

Yesterday, US stocks took a nosedive, and as the pure HBM leader in the market, #MU dropped 15 points. As long as the big players in capital expenditures don't hit the brakes, the semiconductor sector is still set to moon. The major cloud providers are all throwing cash like there's no tomorrow to keep up in this AI race, and right now, there's no sign of a slowdown. In the capital markets, there's actually one key indicator to watch, and that's the flow of money. Semiconductors aren't about financing; it's about the cash flowing in. While the big firms are still raising funds and pouring money in, the volatility in semiconductors is only about valuation adjustments, not earnings adjustments. After the close on Wednesday,

Are there any stocks in the US market that are purely focused on MLCC? The answer is no, there aren't any particularly pure plays. The closest in the US market is Vishay Intertechnology, #VSH, but it’s not a pure MLCC company; it's a mixed bag of passive components and discrete semiconductors. Today, let’s dive into VSH, but first, let’s understand what MLCC is.
MLCC is a multi-layer ceramic capacitor, heavily used in smartphones, cars, servers, GPUs, AI servers, power supplies, and motherboards. AI servers particularly drive demand for high-capacitance, high-reliability, high-temperature, high-voltage, low ESL MLCCs. Recently, the market has been buzzing about how AI servers are causing longer lead times and price hikes for MLCCs. PCGamer cited DigiTimes, stating that a single AI server board might require thousands of MLCCs, with demand far surpassing that of regular servers. Murata has also been reported to have price increases of 15%-35%.

This week on the 26th and next week on the 1st-2nd are times when US stocks tend to be pretty volatile. The main reason is that it's the end of the quarter for pension funds, plus half-year rebalancing strategies. Honestly, the last few trading days of each quarter usually see some significant swings. Now that it's June, this volatility is going to be even more pronounced because the end of June is quite special – it's the end of the month, end of the quarter, end of the half-year, and funds/pension rebalancing index adjustments all happening at once. After options positions are reset, we hit a window right before the new half-year allocation cycle. So, at the end of June, you're likely to see some technical buying and selling that isn't driven by fundamentals. That's why we saw a reversal candlestick on March 31st when pension funds started buying stocks. But why does this impact US stocks? Because pension funds, target date funds, insurance capital, and sovereign funds are massive players. They don’t just buy #Nvidia today because they’re bullish and then short it tomorrow if they turn bearish. They manage risk ratios. If stocks rise too much and their stock allocation exceeds the target, they’ll sell; if stocks drop too much and their stock allocation falls below the target, they’ll buy. This target ratio is generally around 60% stocks and 40% bonds. If stocks soar this quarter and bonds stay flat or even drop, the mix might skew to 65% stocks and 35% bonds. So, by the end of the quarter, they need to: sell stocks and buy bonds. To illustrate with an example, let’s say a pension fund has $100 billion in assets, targeting a 60/40 split: initially, $60 billion in stocks and $40 billion in bonds. If this quarter stocks rise by 10% and bonds drop by 2%, stocks become $66 billion and bonds drop to $39.2 billion, giving a total asset of $105.2 billion. The target stock position should be: $105.2 billion × 60% = $63.12 billion, but now stocks are at $66 billion, so they need to sell $66 billion - $63.12 billion = $2.88 billion worth of stocks and then buy bonds. Some might ask, why not just hold on to the stocks if they’re soaring and wait for them to go even higher? This ties into the concept of the pension fund’s “liabilities.” If the pension fund's assets are already sufficient to cover future retirement payments, say a funded ratio of 110%, they might not want to take on too much stock risk. So, they would reduce stock exposure and add bonds to lock in profits and reduce volatility. This is called de-risking, and locking in the yield from risk-free government bonds is crucial. That's why US Treasuries serve as the anchor for all assets. To sum it up in one sentence: the rebalancing at the end of the quarter/half-year for pension funds is big money pulling their portfolios back to target ratios. In this environment where stocks are strong and pension fund conditions are high, the end of June is more likely to see technical pressure to sell stocks and buy bonds. However, this is usually just a short-term disturbance and doesn't mean a reversal in the US stock trend. As long as you judge that the trend hasn’t worsened, these fluctuations caused by pension fund rebalancing generally present good entry points for mid-to-long-term positions.

About #kaito mastering it, just three steps.
The first purely relies on talking and desperately flattering the project party.
The second pledge #KAITO pledging has risks, I pledged when it was 30 dollars #ATOM and getting back is a long way off.
The third is participating in the platform's new share offerings; there will be another tutorial on new share offerings later.