沪上老徐

Everyone knows that the most expensive component in an AI server is the GPU. But how many tiny capacitors, each smaller than a fingernail, are crammed into a next-generation GPU rack? The answer is forty to sixty thousand. These are MLCCs (Multilayer Ceramic Chip Capacitors), the "rice" of the electronics industry—unremarkable yet indispensable. A phone has hundreds, a car has tens of thousands, and an AI server directly requires tens of thousands. The current absurdity is this: No matter how expensive GPUs are, NVIDIA can manufacture them. Yet this tiny capacitor costing just a few cents has recently become the bottleneck strangling AI development...

Last night, after the Iran news broke, the Nasdaq 100 plunged 3% intraday, and the watchlist page was a sea of red. However, $2x Long VIX Futures ETF(UVIX.US) surged from just over $4 to $4.65 in one go intraday, rising as much as 16%, with over 87 million shares traded for the day; in pre-market trading today, it's up another 6% to $4.38. A quick explainer for newbies: VIX is the fear index, a measure of "market nervousness" calculated from S&P 500 option prices—the more panicked the market, the higher it spikes. So here's the question: VIX is an index...

Today, the US stock $Navitas Semiconductor(NVTS.US) closed at $28.88, down 9.2%, while the Hong Kong stock $HUA HONG GRACE(01347.HK) closed at $169.70, up 11.4%. At both ends of the same wave of AI data centers + new energy vehicles, the directions are completely opposite: Navitas Semiconductor was founded in Silicon Valley, USA in 2014, initially focusing on gallium nitride (GaN) power semiconductor integrated ICs...

Today, a new ETF in the US stock market left me stunned again — $Leverage Shares 2X Long CBRS Daily ETF(CBRG.US) surged 20.7% in a single day, while the $Cerebras(CBRS.US) it tracks rose 10.4% over the same period. One is doubling in speed, the other is moving at a normal pace. It might look like some kind of mystery, but it boils down to one sentence: CBRG is a 2x long leveraged ETF of CBRS. Cerebras Systems...

On May 20, $Bloom Energy(BE.US) signed a $2.6 billion energy agreement with Nebius. On May 24, it secured a fuel contract for a 328-megawatt AI data center. The stock price hit a 12-month high on May 21...

The 113-year-old $IBM(IBM.US) is considered by many to be long overdue for retirement. Last Friday, the U.S. government announced a $20 billion investment in quantum computing, with IBM alone taking $10 billion. On the same day, its stock price opened 3% higher, surging to $262, before being pushed back down to close at $254...

先讲一个台湾半导体业的旧梗。1980 年那批从美国硅谷回台湾创业的工程师里，张忠谋去了台积电（TSMC），曹兴诚去了联电（UMC）——这是台湾"晶圆代工双雄"。后来台积电一路狂奔到 3nm 制程世界第一，UMC 卡在 28nm-14nm 这一档没法越级，市场习惯把它当"老二"看，估值长期是台积电的零头。但 2024 年到现在的故事变了。这帖讲清楚 $United Microelectronics(UMC.US) 这一个月凭啥涨 44%...

先抛一个反直觉的事实：$Arm(ARM.US) 这家公司自己一颗芯片都不生产，今年涨幅却在 iShares 半导体 ETF（SOXX）三十个成分股里排第二（+169.77%），仅次于 Intel，把美光、台积电、AMD、英伟达全部踩在脚下。更离谱的是，北京时间 5/22 凌晨 04:00 收盘，ARM 单日 +16.16% 收 .23、市值首次踩破 .58%...

Have you guys heard of $Leverage Shares 2X Long CBRS Daily ETF(CBRG.US)? Most likely not, because it has a total float of less than 600k shares, and its trading volume on 5/20 was only $4.2 million. But its underlying asset CBRS has been a frequent headline in the past two weeks—it just IPOed on 5/14, was hailed by analysts as the "Next Nvidia," and Cathie Wood bought 100k shares in a single day on 5/20...

First, let me ask everyone something — on 5/20, LRCX rose 6.84%, and LRCU rose 14.41%, almost exactly double. If you happened to buy LRCU yesterday, you'd be grinning from ear to ear today. But here's the question: how long can you hold onto a stock that "doubles in a day"? Let me explain the background clearly before making a judgment. $Lam Research(LRCX.US), its full name is Lam Research Corporation. What does it do? To put it simply, it's the "Blue Sky Technical School excavator" in the semiconductor world — no matter what chips the wafer fab is making...

2024 年 7 月 CrowdStrike 那次全球蓝屏事件，900 万台 Windows 设备同时崩溃、航空公司一天损失上亿美元、CRWD 一周市值蒸发 200 多亿。但 18 个月后的今天，CRWD 股价是 $617，距历史新高就一步之遥。按常理这种"自己产品把全球网络打瘫"的公司应该挂掉了，但它不仅没挂，整个网安赛道反而越涨越稳...

Last Wednesday (5/14), an AI chip company called Cerebras Systems went public on Nasdaq under the ticker CBRS$Cerebras(CBRS.US). This event didn't attract much attention domestically, but it generate...

Let's look at a set of comparison numbers. The iShares Software ETF $iShares Expanded Tech Software Sector ETF(IGV.US) rose from $86.63 at the close on 5/1 to $92.87 at the close on 5/18, a cumulative increase of +7.2% over 17 trading days. In the past 5 trading days, it closed in positive territory for 4 days, and on 5/18, it even continued to rise +1.19% against the Nasdaq (-0.51%). During the same period, the semiconductor ETF SMH...

I often see people asking what $Snowflake(SNOW.US) actually does for business. Let me explain briefly: In a nutshell, Snowflake is a "data warehouse cloud platform"—it helps large companies put their data on the cloud for unified management and analysis, essentially building a super database that can be networked for enterprises. How it makes money: It charges based on usage, not by seat or fixed monthly fees. The more computing power used and the more data stored, the more you pay. So its revenue is directly tied to the customer's data usage. The more companies adopt AI, the greater the data processing volume they use...

$Cerebras(CBRS.US) Can you imagine a chip as big as an iPad? A normal GPU, like the NVIDIA H100, is about the size of a postage stamp. But there's a company dedicated to "not cutting wafers"—Cerebras, which makes an entire 12-inch wafer (about 30 cm in diameter) into a single chip, called the Wafer Scale Engine (WSE). A single chip packs 900,000 cores and can run AI inference at 10-20 times the speed of a GPU. This company just had an IPO on Nasdaq...

Have you ever thought about: why can a Bitcoin mining company suddenly sign a 4 billion contract with NVIDIA? To put it bluntly, it's because these mining companies possess the scarcest resource in the AI era—electricity. Last year, everyone was still laughing at crypto miners as "cyclical gambling," but this year, they've transformed into hot commodities for AI data centers. The reason is simple: training a Frontier Model doesn't require smarter engineers, it requires bigger power plants. AI data centers often start at 500MW, 1GW...

$Credo Tech(CRDO.US) It's a matter of less than a week, and a single stock can swing more than 10%. Its market cap is over $30 billion, not some small, ignored stock. In the AI optical communication sector, the volatility is just this high. CRDO is Credo Technology — not the Credo for credit cards, but the Credo that's "Broadcom's little brother." I'll explain it in plain language...

$Hewlett Packard Enterprise(HPE.US) closed at $32.07, up 6.16% for the day. This isn't a huge wave — but after-hours it was pulled to $33.85, up another 5.5%. Then I saw in the news that activist investor Irenic Capital joined. This company is Hewlett Packard Enterprise. It sounds like a name from the last era, why would anyone suddenly want to buy it in the AI cycle? I think...

$Maxlinear(MXL.US) This stock was still just over $20 in mid-April, and by May 11th Eastern Time (early morning of May 12th Beijing time), it had reached $102—almost a fivefold increase in a month. The most frequently asked question is: "What does it do? Why is it surging so much?" I'll first dig into its background, then walk you through what happened over the past month step by step. MaxLinear is involved in optical communications + analog mixed-signal chips. It started out making modems for broadband TV set-top boxes, which sounds pretty boring, right...

A buddy asked about $Ambarella(AMBA.US) — it closed at .72, up 4.48% on May 11th EST, and only gained 7% for the week. It doesn't look like a blockbuster stock like MXL, but it's always mentioned in semiconductor sector discussions. Today, I'm going to lay all its cards on the table and talk about it. First, a question: What's the most expensive resource in the AI era? Most people would say computing power, which is why NVDA has skyrocketed. But let me look at it from another angle—the flip side of computing power is energy consumption. An H100 at full load draws 700 watts...

In this wave of AI data center boom, everyone is eyeing NVIDIA's H100 and B200, a few are watching Broadcom's ASICs, but—who is making money from that network cable extending from each GPU, the network responsible for making tens of thousands of GPUs "talk to each other"? Most people's first thought is ARISTA. But actually, $Cisco(CSCO.US), this networking relic founded in 1984, has quietly risen 27% year-to-date, and is up 29% since its Q2 earnings report...