Oli.

Oli.

🍓Web3投研 🍑人工智能 🚀《干翻狗庄》系列工具作者

5Following
280followers

Feed

Oli.
Oli.
Tesla and SpaceX are reportedly advancing merger talks, with an announcement possible as early as December. Both companies' filings confirm the establishment of special committees, and the potential deal is still under evaluation. What makes me suspicious is that Tesla's own disclosures show that exploring deeper cooperation isn't limited to a corporate merger. If some things can be done together by signing agreements, why merge the entire companies? This question ultimately has to be answered by the dealmakers. Imagine if both sides really need to share certain computing resources or jointly purchase equipment. Long-term contracts can already specify who pays and who has priority usage rights, and can include exit mechanisms if a project fails to meet requirements. To make cooperation run smoothly, the two parties might indeed need closer arrangements. But if the plan is to put cars, energy storage, satellites, and other businesses all into one acquisition, shareholders have reason to ask which benefits can only be obtained through a merger and how much extra cost that entails. If a deal document is released, I would first look at how the board compares the merger route versus continuing cooperation. Ideally, I want to see what was specifically discussed and why this option was chosen. The exact day in December when the announcement is made will certainly be used by the market for trading. For shareholders, the pages worth reading are often those that come after the announcement and after the stock price has already jumped, when people are willing to read carefully. #特斯拉与SpaceX合并谈判取得进展,最快12月宣布
Oli.
Oli.
NVIDIA's revenue guidance midpoint for this quarter is $108 billion, surpassing $100 billion has long been expected. What I actually want to hear in the earnings call is some details about the networking business. As model training moves to larger clusters, no matter how many GPUs are purchased, the data transfer speed between them must be fast enough. If customers are willing to upgrade switches, network cards, and interconnects together, the amount of money a data center spends on NVIDIA could continue to increase. This revenue also faces competition. Large cloud providers have their own networking solutions; they can use NVIDIA's accelerator cards but will also choose different interconnect technologies across different clusters. A major customer purchased a complete system this quarter, but as the cluster expands next quarter, will they still be willing to purchase the same network configuration? I think this question is more interesting than the earnings just exceeding a few hundred million dollars; it relates to how much NVIDIA can sell in the customer's next data center. So, if revenue truly surpasses $100 billion, I will be happy at first, then look at where the growth comes from. If the guidance for the next quarter can still be raised, and management can clearly explain which types of clusters the networking products have entered and whether the demand comes from more customers, this report card will be more convincing. If we only hear that GPUs continue to sell well but don't hear how the supporting business is growing, my expectations for the following quarter will be a bit more reserved. #英伟达冲击千亿美元季度营收纪录
Oli.
Oli.
There has been a substantial change regarding renting high-end GPUs overseas. The U.S. Department of Commerce's Bureau of Industry and Security has issued a temporary final rule, starting to include remote computing power provided by some overseas data centers under export control considerations. The public text also lists new licensing requirements for exports of advanced semiconductors to parts of Indonesia, Malaysia, Singapore, Thailand, and Brazil. A few days ago, there was still discussion about whether action would be taken; now operators have to recalculate project schedules based on the text. Which type of projects am I most worried about? Those whose data centers and power supplies have already been negotiated and planned for phased deployment. The first phase may have already started, but subsequent batches of equipment have not yet arrived. If an additional license is required for later shipments, the costs for data centers and labor must still be paid during the waiting period, while the computing power provided to customers may be delayed. Even if no chips are missing, a delay of several months in deployment will affect cash flow compared to the initial project plan. Next, when looking at announcements from cloud computing power companies, I will pay attention to whether they clearly state "how much capacity has been contracted" and "when it can be delivered." How old contracts are handled and which terms apply to new orders must be reviewed case by case according to the applicable scope. The worst scenario is the company only reporting a long-term total capacity, but when customers ask how many cards can be activated next month, there is no clear date. This time, the timing difference alone can eat up a lot of profit. #美国拟扩大AI芯片出口限制,海外算力租赁或受影响
Oli.
Oli.
Seeing that the U.S. government might label a batch of AI companies as “national champions,” my first reaction is that the name sounds like an award, but the real test might be whether several competitors can successfully close an overseas deal together. The U.S. previously received 78 joint applications for its AI export plan, focusing on packaging chips, cloud, models, and applications for export. The new round of reports mentioned names like Nvidia, Amazon, and Microsoft, but the official list of selected companies is yet to be announced. When it comes to actual overseas customers making purchases, the issues are far more complicated than just the list. In a single solution, who delivers the computing power, which cloud runs the model, where the data is stored, and what happens if the supplier needs to be changed after three years? If customers still have to negotiate financing themselves and sign contracts separately with each company, it’s hard to say how much hassle the so-called integrated solution really saves. If several companies can handle deployment, training, and after-sales service together, it will be much more attractive to clients without large technical teams. What interests me most about this is where the first batch of projects will land and who will pay for them. Being selected can generate some momentum, but whether the project can be replicated depends on whether the customer is willing to buy a second phase after the first deal is done. Only then will there be something to calculate. I think it’s a bit premature to boost company revenues just based on the list. #特朗普政府拟公布AI“国家冠军”计划入选名单
Oli.
Oli.
As of November 13, the US spot BTC ETF has seen net outflows for seven consecutive trading days, totaling about $1.9 billion. I looked into the breakdown for the last day, and $185.4 million of the net outflow came from IBIT and FBTC, while all other recorded products were zero that day. So, a number that looks like the entire sector is withdrawing en masse is actually mainly driven by two large funds. What difference does this make? Seeing products of various sizes redeemed over several consecutive days makes me more worried that selling pressure has spread. If the total volume is mainly determined by a few leading products, a sudden positive number the next day still requires checking how many funds have shifted together. A single large fund subscription can pull the total back into positive territory, while the rest of the funds might still be stationary. ETF data tells us which products have net subscriptions or redemptions, but it cannot see who is making the moves, nor whether there are positions elsewhere hedging. In recent days, I have been more critical of the phrase "institutions are back." Even if the next capital bar turns green, I want to observe for a couple more trading days to see if net redemptions shrink and if more products have subscriptions. The continuity over seven days is worth noting, and reversing this selling pressure also requires some continuity. Just looking at a nice total number, I am not easily excited right now. #BTC现货ETF连续7个交易日净流出
Oli.
Oli.
The US October PPI rose 0.3% month-over-month, below expectations. When you see this on the chart, it's easy to jump to the conclusion that "December rate hike pressure is easing." But there's a detail in this table that's easy to overlook. Excluding food, energy, and trade services, the final demand price still rose 0.5% month-over-month, the same as last month. The total number has indeed cooled down, but from another perspective, the improvement isn't that obvious. Especially for trade services, which track the gross margin changes of wholesalers and retailers. Imagine a retailer discounting to clear inventory; the price they pay for goods can be a bit cheaper, but the merchant's profit margin might also be squeezed thinner. After the promotion ends, whether prices can stay low next month depends on procurement costs and demand. Seeing the overall PPI decline, I would ask more: which segment is absorbing the pressure, and how long can it continue to do so? So my judgment on December expectations hasn't changed for now. This data allows for more debate, but to make the market clearly drop rate hike bets, we need to see several subsequent readings moving in the same direction. The most frustrating thing in crypto is when the market gives you ten minutes to celebrate, then takes back those gains. If only the total number surprises and other data lags behind, it's hard for me to treat the short-term relief as the end of rate hike pressure. #美国10月PPI低于预期,12月加息预期仍强劲
Oli.
Oli.
Regarding the news about Hormuz this time, I will read Iran's statement and the maritime reports separately. Iran claims that two oil tankers hit mines. The UK Maritime Trade Operations office also issued two tanker incident reports on November 16, one of which described a tanker being hit by an unidentified projectile while exiting the strait. The reports indicate that navigation risks remain, but the specific causes of danger need to be verified later; these pieces of information cannot yet be combined into a single conclusion. From the shipowner's perspective, the situation is actually quite straightforward. Even if someone tells you the channel is still passable, as long as there are doubts about whether underwater threats have been eliminated and whether rescue can be timely in case of emergency, it is difficult to decide when the next ship can pass. Fully loaded ships cannot leave, and buyers will delay receiving cargo. Empty ships are also reluctant to enter, reducing the number of vessels available to transport oil. This drag on both ends affects more than just the number of ships passing on a given day. What I care about now is whether fully loaded tankers can pass through the strait continuously for several days, and whether there are clear navigation arrangements after incidents. Occasionally, a ship passing smoothly is certainly good news, but refinery production scheduling depends on the next ship and the one after arriving on time. Oil prices can be traded in advance to ease expectations, but crew and shipowners must make daily decisions based on the day's risks. This ongoing situation is not so easy to resolve. #伊朗称霍尔木兹两艘油轮触雷,通航安全再受关注
Oli.
Oli.
From early October to this pullback, the total market capitalization of the crypto market has decreased by about $800 billion. This number is quite striking, but when you actually look at the trading interface, I think the most frustrating thing might be something else. You thought you could still sell at that price yesterday, but when you check the order book today, you find that the buy orders have thinned out considerably. Market capitalization is estimated based on the latest transaction price for circulating tokens. When the transaction price moves down a bit, the remaining unsold coins must also be revalued accordingly. Over the weekend, BTC once approached $62,000, and the market simultaneously saw large leveraged liquidations. Forced liquidation orders rush to execute, and if the quoting side withdraws their orders further away, the same sell order could push through several price levels. At this time, the unrealized loss on the account and the actual loss upon closing the position often have a gap. This time, I’m more interested in seeing whether the order book can recover after the pressure subsides. A single candle pulling BTC back can easily give the illusion that the issue is over. If the next day normal-sized sell orders can still cause significant slippage for major coins, and the bid-ask spread for smaller coins remains wide for a long time, it indicates that the market’s capacity to absorb selling pressure has not yet recovered. Price gains will convince people to chase in, but the trading experience will tell you how wide the exit door really is when you want to leave. #加密市场缩水8000亿美元,这轮下跌有何不同?
Oli.
Oli.
Anthropic's IPO subscription orders are reported to have reached nearly 7 times the basic issuance size. When I saw this number, the first thing I did was to check the prospectus to see exactly how many shares could be sold during the listing days. The initial offering plan is 825 million shares, with underwriters having an additional overallotment option of up to 123.75 million shares. The lock-up arrangements for existing shareholders are also tiered, with some shares eligible for early unlocking after meeting certain conditions. The supply and demand seen on the first day is just a small segment of this company's supply and demand for the coming years. This can easily get people excited. Those who didn't get allocations might want to buy more after the market opens, leading to concentrated short-term buying. But if the pricing is raised, some funds originally willing to subscribe might change their minds. Later, when unlocking occurs, shareholders who have held shares for a long time will have opportunities to cash out. Each group will have different price expectations for transactions. So when the market opens, I will first note the actual issuance price and then observe the trading in the first week. If the first day is very lively but trading volume shrinks afterward and the stock price drops slightly when facing selling pressure, I will discount that "7 times" figure. Filling orders during book-building is one thing; how much the market is willing to pay to take over existing shareholders' stakes will only become clear after listing. I am more interested in the latter price. #AnthropicIPO进展:认购订单接近发行规模7倍
Oli.
Oli.
These past two days, people have been guessing whether Jensen Huang will return. The latest report says that Nvidia's board has not considered letting him resume the CEO position. The company has publicly confirmed that Colette Kress will continue as interim CEO, and the official candidate will be sought both internally and externally. Rene Haas from Arm has been named by the media, but the board previously said they are still selecting a headhunter and have not formally contacted any potential candidates. Putting these pieces of information together, the official announcement is still far off. I think the hardest part for the board is that recruitment and investigation have to proceed simultaneously. Before an external candidate takes on this job, they will definitely want to ask which matters the board will handle during the independent investigation, to what extent management can speak publicly, and which questions must wait until the investigation concludes. Until these issues are clarified, even if compensation is agreed upon, the candidate may not be willing to sign immediately. Therefore, at this earnings call, what I most want to hear is whether the company can provide a timeline for the next public update. The interim management team can continue to push the business forward at the original pace, but the board should also let employees and shareholders know which matters have been settled next. The most exhausting thing is that every few days a new name pops up, the person involved says nothing, and the market has already scheduled their start date. #英伟达下一任CEO人选会是谁?