
美股投资young
美股投资young
美股科技投资者 ai狂热者 分享每日快讯. all in ai
303Following
792followers
Feed
Livestream
Feed
After this election, news about AI regulation and data center approvals changes daily, and semiconductors are being hit back and forth. I really want to know who knew something earlier than us, so I check those who must publicly disclose their trading records.
I often use three websites.
The first is OpenInsider, which specializes in tracking trades by company executives and major shareholders. The U.S. requires insiders to report trades within two business days via Form 4, and this site organizes all those reports into tables. The most important to watch are "Cluster Buys," where several executives from the same company buy shares with their own money around the same time. A single executive selling stock usually carries little information—many are pre-planned sell-offs or tax payments—but when several people simultaneously put money into their own company’s stock, that signal is much stronger. When viewing, pay attention to the trade type: P means a real purchase, S means sell, A and M are grants and exercises, so don’t get confused.
The second is Capitol Trades, which tracks stock trades by U.S. Congress members. Members must legally disclose their stock trades, and this site neatly organizes those disclosures. You can search by member or by stock, and it’s free. With all the recent noise about AI regulation and chip export controls, seeing which members are trading semiconductor stocks and which committees they sit on is more reliable than just reading their public statements.
The third is Quiver Quantitative, which also has congressional trading data but combines it with government contracts and lobbying expenditures. For example, if a company wins a large government contract and a member of Congress is buying its stock, seeing these lines together gives a fuller picture.
The biggest problem with this kind of data is the lag. Members have a 45-day window to report, so by the time you see it on the site, they may have already completed their trades. Also, the reported amounts are ranges, not exact prices, so you don’t know the exact purchase price. Therefore, using it to copy trades is basically useless; what’s useful is watching the direction: a large group simultaneously increasing a sector, or a committee collectively exiting—these group moves have reference value.
Insider data is the same: executives buying means they think their company is undervalued, but it doesn’t guarantee the stock won’t drop next week. It reflects long-term confidence, not short-term timing.
https://t.co/DJexqc6v6b
https://t.co/xHvnndTgS2
https://t.co/CNZXdAjy5V
The 10-year US Treasury yield is back above 5%, tech stocks are getting hammered every day, but most people can't even accurately say what today's US Treasury yield is, relying entirely on financial media reports.
I only use two sources to check interest rates.
The first is the US Treasury Department's official yield curve page. It publishes the official yields for all maturities after market close every day, from 1 month to 30 years. All media-reported numbers ultimately refer to this. The most important thing to look at on this page is not a single maturity but the shape of the curve. For example, this morning intraday, the 2-year was about 4.80%, the 10-year 5.08%, and the 30-year 5.42%. The short end is down, the long end is up, and the curve is steepening. The short end reflects the market's expectations for the Fed in the coming months, while the long end reflects how much compensation is required to lend money to the US for 10 or 30 years. Last week’s PPI was milder than expected, the short end reacted accordingly, but the long end did not follow, indicating that the market’s demand for long-term compensation is still rising. Companies that borrow long-term money, spend big, and take years to break even suffer the most in this environment; new cloud companies building data centers through financing are typical examples.
The second is TradingView, which I use to overlay interest rates and stock prices. Searching US10Y and US02Y pulls up yield trends, then overlaying QQQ or SOXX on the same chart lets you see if interest rates are suppressing valuations during that period. You can also input US10Y-US02Y directly to plot the yield spread, which is much more useful than just watching a single 10-year number to see if the spread is widening or narrowing.
The benefit of using both together is: the official site’s numbers are accurate for verification; TradingView’s charts are intuitive for viewing trends and relationships.
One easy mistake to make: in early October, the 10-year briefly exceeded 5.3%, while the Philadelphia Semiconductor Index was still high, so just looking at the 5% number alone doesn’t tell the whole story. High rates and AI stock declines happening simultaneously don’t necessarily mean rates caused the drop. In the past two weeks, there were also $NVDA leadership changes, export controls, and $ANT IPOs. You have to put interest rate trends and stock prices on the same chart to know who’s really driving the market during this period.
https://t.co/ZoxEWXjCiZ
https://t.co/DonSkciAv8
$NVDA's GPUs are getting more expensive, but AI usage is becoming cheaper, and the price gap in between will tonight rely on $ANT's investors to fill.
First, let's look at two sets of numbers moving in opposite directions.
At the earnings call, management said Hopper renewed contracts with prices rising up to twice the original. Since the release of the B300, rental prices have increased by over 60%. All cloud providers and new cloud computing power are fully utilized.
During the same period, token prices have dropped nearly 60% in six months.
GPUs are expensive, compute rental is costly, yet the price for selling models is dropping. Caught in the middle are two types of companies: new cloud providers renting GPUs to others, like $CRWV and $NBIS; and model companies buying compute power for training and inference, like $ANT.
These two types of companies are now $NVDA's fastest-growing customers. Outside the four major cloud providers $MSFT, $GOOGL, $AMZN, and $META, Q3 contributed 47.5 billion in revenue, nearly doubling growth. Management also said that cutting-edge model labs account for about 15% of data center revenue this year.
Where does their money come from?
New cloud providers rely on borrowing, a path $NVDA has started to tighten by requiring them to secure financing first before commitments. Model companies rely on fundraising, with the biggest one tonight: $ANT pricing 825 million shares in a range of $70 to $78, raising about 61 billion at the midpoint, with a valuation around 1 trillion dollars, starting trading on Nasdaq tomorrow. $NVDA itself has committed to subscribe for 10 billion.
Following the money down: much of $ANT's raised funds will turn into compute contracts, which then become orders for new cloud and cloud providers, ultimately returning to $NVDA's revenue. Part of $NVDA's confidence in GPU price hikes rests on this IPO.
There is also corroboration. UK new cloud Nscale is building a 45 billion dollar data center for $ANT, and $NVDA's funding support is still under negotiation.
So if $ANT prices smoothly and is stable on the first day, this chain can keep turning. If pricing is at the lower end of the range or breaks below on day one, the first to suffer will be new clouds like $CRWV and $NBIS, then $NVDA.
Does $NVDA itself understand this risk?
From this earnings report, it is already looking for the next batch of payers:
Sovereign AI, revenue growth more than doubled, with Australia's cooperation target to build 2GW by the end of 2027.
Physical AI, meaning robotics and autonomous driving, nearly doubled growth, expected to exceed 25 billion for the year.
Enterprise on-premises deployment, nearly 5 billion in the past 12 months, management expects enterprise revenue to more than double by FY28, benefiting $DELL and $HPE.
CPU, Vera has started selling separately, with FY28 CPU revenue expected to more than double, originally $INTC and $AMD's territory.
Software, signed a formal agreement to acquire Hugging Face in September.
The directions are right, but the scale is still small. Physical AI's 25 billion annual revenue accounts for about 6% of this year's roughly 424 billion revenue. The four major cloud providers generate 60.1 billion in a single quarter. These new legs combined still can't support $NVDA in the short term; the bulk of revenue still depends on cloud providers and model companies paying.
Another change worth noting: cloud industry's backlog orders rose from 2 trillion to 2.5 trillion in three months, with no demand reduction. But rising memory and system prices are suppressing consumer PC demand, with $HPQ hit first. Costs are passing downstream, and whether AI application revenue can catch this at the end is the biggest question for the entire chain.
Next, watch these:
$ANT's pricing tonight and first-day performance tomorrow.
Whether the scissors gap between GPU rental and token prices will continue to widen.
Whether the cloud industry's backlog orders can keep rising next quarter.
Whether Physical AI can deliver 25 billion for the year.
$NVDA controls GPU prices, the market controls token prices. The price gap in between will tonight first depend on whether $ANT's investors are willing to pay.
Really f***ing nonsense
This board of directors
$NVDA keeps buying and getting stuck all the way, I can only say awesome

$NVDA This earnings report is most worth looking at to see where the money flows. Following Nvidia's costs, procurement, and capital expenditures, who benefits and who loses is basically all written in the report.
Ranked by degree of benefit:
First tier, memory. The most direct and certain.
Nvidia's supply commitments rose from 279 billion to 390 billion in one quarter, with the extra 110 billion mainly for multi-year memory procurement. Next quarter's gross margin will drop by 3 points; the company itself says the main reason is memory price increases.
To translate: Nvidia is willing to give up 3 points of gross margin to lock in HBM for the next few years. This is rare in the AI supply chain; previously others sought Nvidia, now Nvidia is placing long-term orders with memory manufacturers in advance. Nvidia personally endorses the pricing power of memory manufacturers this time.
Related: $MU, Hynix, Samsung follow the same logic. Memory manufacturers expanding production also benefit equipment-side $LRCX.
$SNDK should be discounted. NAND is not as directly related to this HBM long-term order, more driven by sentiment.
Second tier, TSMC and advanced packaging.
Rubin sold nearly 20 billion in the first quarter; the company says this is the fastest ramp-up in history. FY28 growth forecast raised from 70% to 75%, management repeatedly says supply is the bottleneck, customers are still chasing orders. Where is the supply bottleneck? 3nm, CoWoS, HBM. TSMC's certainty is second only to memory.
Related: $TSM $ASML $AMAT $KLAC, packaging and testing look at $AMKR $ASX
Third tier, networking and optical communications, but internally differentiated.
Networking business exceeds 21 billion per quarter, growing rapidly. Note Nvidia says CPO (co-packaged optics) is already in mass production, claiming it saves 5 times the power compared to pluggable optical modules.
CPO volume growth benefits laser and optical chip makers. Traditional pluggable optical module makers will lose some market share long term. Optical communications cannot be uniformly bullish this time; it depends on whether you hold upstream chips or module assembly.
Beneficiaries: $LITE $COHR $FN $CRDO
Under pressure: $AAOI, also watch A-share Zhongji Xuchuang and Newisland
Fourth tier, power and data center infrastructure.
Many people overlook this. When analysts asked Nvidia how it will support customers in the future, Kress's top answer was "land, power, factory buildings." Previously Nvidia's money went directly to new cloud card purchases; now it is more likely to invest in underlying data center infrastructure. These companies effectively gain a new financial backer.
Related: $VRT $ETN $GEV $PWR, power generation side $VST $CEG
Fifth tier, servers, neutral bias.
Enterprise on-premises deployment revenue nearly 5 billion in the past 12 months; management expects FY28 to more than double, an incremental gain for server manufacturers. But memory price increases also squeeze their gross margins; revenue rises, but profits may not follow.
Related: $DELL $HPE $SMCI
Below that are the losers.
New cloud, bearish. These companies' business is actually not bad: Hopper re-contract prices up to twice original, B300 rental prices increased over 60% since launch, compute fully utilized. The problem is money. CFO comments specify commitments to new cloud only take effect after the other party secures financing and Nvidia verifies compute; future support will be "restrained and selective." Good business, but reliable partners step back, valuation must be recalculated.
Related: $CRWV $NBIS $IREN $APLD
PC and consumer electronics, bearish. The report directly states memory and whole machine price increases suppress consumer PC demand.
Related: $HPQ, Lenovo follows same logic
Server CPU, slightly bearish. Nvidia starts selling Vera CPU separately, expects FY28 CPU revenue to more than double. This market originally belonged to Intel and AMD, volume still small, outlook not good.
Related: $INTC $AMD
Looking at this circle, the conclusion is clear: this round of AI supply chain profits is shifting from selling compute power to the bottleneck segments. Whoever holds scarce capacity has pricing power. The scarcest now is memory, then advanced process, then power.
Nvidia itself is conceding margin to memory manufacturers; this signal is more straightforward than any research report.
$NVDA Earnings Report Analysis
This week, two things were lost simultaneously: the last bit of revenue from the Chinese market, and the person who could shuttle negotiations between Washington and Beijing.
The timeline is tight.
11/12, Jensen Huang resigned. According to media reports, one reason was his extramarital affair with a woman suspected by U.S. intelligence agencies to be a Chinese spy.
11/16, the U.S. Department of Commerce issued new chip export regulations.
11/17, on the earnings call, Kress announced that shipments of authorized Hopper chips to China have stopped.
Let's first look at the China account.
In Q3, Nvidia still sold $2.4 billion worth of licensed Hopper chips to China. This batch was a special license negotiated specifically between Washington and Beijing, and now it has stopped. The Q4 guidance does not include any revenue from this. Revenue from Chinese data centers officially drops to zero starting this quarter.
$2.4 billion is not large for a company with quarterly revenue of 115.8 billion, but the real trouble is the new regulation on 11/16.
The new regulation has two layers. The first: from now on, selling advanced AI chips to Singapore, Malaysia, Indonesia, Thailand, and Brazil requires U.S. approval. The second, harsher layer: any data center worldwide that might provide large-scale computing power to Chinese companies requires export approval. The threshold is roughly equivalent to 750 H100 GPUs. Data centers already serving major Chinese clients must apply within 90 days, or they will no longer be able to obtain chips.
This regulation blocks Alibaba, ByteDance, and others' fallback options in recent years. If they can't buy cards directly, they would rent data center capacity in Southeast Asia. Now that route is also closed.
The impact estimates vary widely. By customer headquarters, companies in these five countries account for at most 1.3% of Nvidia's revenue. But Jefferies estimates the new rules could reduce Nvidia's data center revenue next year by $44 billion to $62 billion.
Why such a big difference? Because many data centers supplying computing power to Chinese companies are registered in the U.S., Europe, or elsewhere, so the demand from China is not visible on the books. Based on management's FY28 revenue estimate of about 740 billion, Jefferies' figure represents a 6% to 8% cut.
Kress said on the call that demand from Rubin is strong enough that the new rules won't affect total revenue, but admitted Southeast Asian customers might gradually be lost. When asked if the claim of 100% growth next year still holds, she did not answer directly, only saying the new rules need more time for evaluation.
So why did the board establish a National Security and Export Control Oversight Committee at this time?
Because Jensen Huang's biggest role in recent years was personally negotiating licenses one by one between Washington and Beijing. His departure is also related to the so-called spy issue. The board now needs to convince Washington that Nvidia is completely severing ties with this matter. So Chairman Jones made it clear on the call: there is no intention to bring Huang back, and government relations will be handled by a committee of independent directors.
Beijing's response was swift. The Ministry of Foreign Affairs labeled the new rules as "economic coercion," and Malaysia has already requested its data centers to comply. At the upcoming APEC, Trump and Xi Jinping are scheduled to meet. Commerce Secretary Lutnick has also agreed to testify before the Senate Banking Committee.
After all this, Chinese AI companies have only one path left: using domestic chips. Nvidia's complete exit from the Chinese market directly benefits companies like Cambricon, Hygon, and Huawei Ascend.
Next, watch four things:
Whether China and the U.S. will ease chip licensing at APEC.
How many Southeast Asian data centers can obtain licenses within the 90-day application period ending mid-February next year.
Whether Lutnick will reveal stricter follow-up measures during testimony.
Who Nvidia's new CEO will be, and whether Washington will accept that person.
With Jensen Huang gone, Nvidia has lost its translator between Washington and Beijing.
This kind of drop is rarely seen
$NVDA

All cryptocurrencies are rising, but gold and the US stock indices are not following
$XAU contract is quoted at 3867.8, basically flat in the last 24 hours. It fell from 3885 to 3859.6 in the morning, rebounded to 3874.7 at noon, then hit a 24-hour low of 3843.4 at 2 PM, and now has recovered to 3867.8. When the crypto market rallied together at noon, gold only moved by a dozen points.
$QQQ contract is quoted at 664.97, down 1.09% in the last 24 hours. The lowest was 658.21 at 10 AM, rebounded to 666.75 at noon, then fluctuated between 663.8 and 667, still 1.2% below the 24-hour high of 673.18.
$SPY contract is quoted at 740.31, down 1.37% in the last 24 hours. The lowest was 736.75 at 10 AM, rebounded to 742.39 at noon, now back to 740, further away from the 24-hour high of 750.95.
The crypto market is all up today, gold is flat, and the US stock indices were still falling before the open. At the same noon, the three markets moved completely differently.



