Oli.

Oli.

🍓Web3投研 🍑人工智能 🚀《干翻狗庄》系列工具作者

5Following
280followers

Feed

Oli.
Oli.
🟢 Oli Daily Brief|2026.11.17
BTC has fallen back to $64,500, SOL has dropped below $72, but the total market capitalization of the global crypto market has increased by 1.17%. Meanwhile, the Fear and Greed Index has risen from 10 to 12, indicating a slight improvement in market sentiment, though it remains in extreme fear. The most noteworthy today are three seemingly positive messages that still cannot prove a market reversal. First, the US SOL spot ETF showed a preliminary net inflow of $10.8 million in the latest trading day, while BTC and ETH-related products still temporarily recorded net outflows. Second, Strategy disclosed to the SEC that it purchased an additional 485 BTC last week, bringing its total holdings to 850,849 BTC. Third, the Solana Breakpoint conference reported new progress in payment applications: Zelle demonstrated an on-chain confidential transfer valued at $1.25 million. ETF subscriptions, institutional accumulation, and payment technology applications all sound like positive signals. But the problem is that they represent three completely different demands. ETF subscriptions reflect fund capital flows. Strategy accumulation reflects a single company's treasury allocation. The payment demonstration represents technical validation and does not equate to actual customer transaction volume. What is more concerning is that, while these messages appeared, large BTC transfers have again turned into net inflows to exchanges, the global stablecoin supply has continued to contract over the past 7 days, and Solana's on-chain transaction activity has not grown in sync with the Breakpoint conference. Therefore
Oli.
Oli.
Seeing the phrase “高盛预计2027年AI投资放缓,” the first thing to ask is, which line is slowing down? Goldman Sachs' recent global outlook nominal AI investment forecast still rises from about $1.1 trillion in 2026 to about $1.8 trillion in 2027. In another study, it discussed that the boost AI investment provides to S&P 500 earnings growth may gradually weaken, while the productivity improvement effect might slowly catch up. These two points are much easier to understand when applied to company accounts. Cloud providers can continue buying more equipment next year, and equipment vendors will still have revenue. But the profit expectations lifted by a surge in orders last year are already very high this year. By the next earnings report cycle, how many new orders must equipment vendors secure for the market to keep raising expectations? On the cloud providers' side, they will start to bear more depreciation and will have to see if the income from renting out new cabinets can keep up. The same AI expenditure, recorded on the buyer’s and seller’s books, could represent two completely different moods. I'm not so pessimistic about next year's AI market, but it's also hard to get excited for a long time just over “capital expenditure increase.” What I most want to see now is companies that have spent money clearly stating how much computing power was used, how many work hours were saved for clients, and how much money was ultimately recovered. When these numbers can be presented on the table just like procurement amounts, the accounting for AI investment will be considered to have reached the second half. #高盛预计2027年AI投资放缓
Oli.
Oli.
The rumor about the merger of Tesla and SpaceX is eye-catching with the timing in December. Both companies have disclosed that their boards have established special committees to evaluate the potential deal, and media reports say the announcement could come as early as December. Rather than guessing the exact announcement date, I'm more interested in knowing which projects the shareholders of each company will fund for the other going forward. Take data centers as an example. If the two companies merge and cash is allocated by the same group, cars, energy storage, satellites, and AI will all compete for the budget. If car sales do well and energy storage earns more money, the group might use that to expand projects with longer payback periods. Conversely, if satellite communications contribute more cash, that might be redirected to robotics or factories. The savings from combined procurement costs can be calculated, but once the money is centralized, it becomes harder for external shareholders to predict which projects will be prioritized. I understand why people like to imagine these businesses combined. But from a shareholder's perspective, I would care a lot about whether we can still see the capital expenditures and cash flows of each business separately in the future. As long as these accounts are clear enough, at least we can judge which projects are worth continuing to invest in. If in the end there is only a nice-looking group growth curve, it might be very difficult to trace whose money is being spent on those businesses that require continuous cash burn. #特斯拉与SpaceX合并谈判取得进展,最快12月宣布
Oli.
Oli.
NVIDIA's revenue last quarter was $96.2 billion, with guidance for the next quarter at $108 billion, plus or minus 2%. So if it crosses the $100 billion mark this time, I will probably sigh along and then immediately turn to the cash flow and balance sheet. The market has long priced in expectations above $100 billion; what really causes divergence is what the collections and inventory look like corresponding to such revenue. The more complex the full cabinet system, the easier it is for component procurement, assembly, and delivery to span quarters. Inventory buildup could be for stocking up in advance for a new platform, or it could be that some components have arrived while the full system is still waiting on other parts. Accounts receivable rising with sales scale is also normal. I will compare the revenue growth with accounts receivable growth and listen to management explain whether payment terms have been extended or if customer prepayments have changed. Looking at just one item alone can easily mistake normal expansion for bad news. What I most want to hear is management clearly explain the stocking rhythm for the next-generation platform. For example, how much demand remains for the old platform, which inventories can continue to be used as the new platform ramps up, and which need to be quickly digested. Crossing $100 billion in revenue is already very impressive. To grow beyond this scale, each generation change requires committing a large sum of money upfront. The more impressive the numbers, the more I want to scrutinize where that money is being allocated. #英伟达冲击千亿美元季度营收纪录
Oli.
Oli.
The news from Hormuz is a bit annoying. On one hand, they say reaching an agreement isn't that necessary; on the other, new reports of oil tankers in danger keep emerging, with some details still lacking independent verification. Every time oil prices move with the headlines, what I really want to see are the freight rates and war risk premiums for the tankers. Whether shipowners are willing to take the next cargo often requires them to decide earlier than market watchers, and if they make a mistake, they have to pay the price themselves. Here's a detail that's easy to overlook. When risks rise, shipowners may demand higher freight rates, and insurance companies might shorten the validity period of their quotes. Buyers then have to switch ships at the last minute or delay shipments. By that time, even if the public quotes for the same type of oil don't change much, the landed cost for Asian refineries could still increase. If refineries start switching to crude from other origins, they also have to consider whether the oil type is compatible with their existing equipment. Even if the oil is purchased, the products refined and the profits may not be as originally planned. Next, I'll be watching to see if shipowners dare to accept normal orders spanning several weeks. Occasionally, a ship passing through will certainly ease the market. But if the next batch of cargo still requires expensive war risk insurance or contracts leave ample room for cancellation and delay, I find it hard to believe that a few conciliatory words can bring down both freight and risk costs together. #特朗普淡化对伊协议必要性,霍尔木兹航运风险持续
Oli.
Oli.
When discussing AI data centers, it's easy to misunderstand power issues as simply a matter of how much electricity is lacking. However, in July this year, the opposite happened in Northern Virginia. After a disturbance in the local power grid, nearly 4000 megawatts of data center load unexpectedly disconnected from the grid. Normally, these campuses have stable power consumption, but when a batch of server rooms simultaneously switched to backup power, the grid had to immediately handle the sudden drop in demand. PJM subsequently proposed reliability requirements for large loads, with this incident being one of the backgrounds. This event made me reconsider the grid connection commitments of data centers. A cloud provider said their project obtained a power usage quota, but I also want to know how long the server rooms can sustain during grid fluctuations, what tests the backup power switch must undergo, and how recovery is staged after a failure. Doing well in a single campus is not enough; if several nearby campuses use the same protection logic, encountering the same fault could cause them to take the same action simultaneously. Therefore, I am increasingly indifferent to AI infrastructure news that only announces installed capacity. There are photos when chips arrive, but running server rooms at full load depends on joint debugging with the power grid. The former can be scheduled for a release date, but the latter relies on repeated testing. When customers truly hand over long-term training tasks, the provider that can handle both power outages and recovery well will have a better chance of securing the next long-term contract. #AI数据中心加速扩张,电力保障成为关键
Oli.
Oli.
The U.S. is discussing expanding AI chip export restrictions this time, and my first reaction is that the legal departments of cloud providers might be busier than the sales teams selling chips. Two senators have suggested that recipients already approved to export advanced chips should apply for a separate license if they want to provide remote access to Chinese companies. Currently, this is a recommendation to the Department of Commerce; whether it will be adopted and which customers and services it will cover depends on the official documents. What I most want to see is how cloud service contracts will change. When a company opens an overseas cloud account, the contracting party, the payer, and the team actually using the computing power sometimes belong to different companies. If in the future cloud providers are required to verify the actual user, they will have to ask more detailed questions when opening accounts, and even switching project teams within the client could trigger a re-examination. Clauses that no one usually pays close attention to—such as who guarantees user identity, how long compliance reviews take, and how prepaid fees are handled during service interruptions—will suddenly determine whether a deal goes through. The market likes to ask how many fewer cards NVIDIA will sell, but I think that calculation can wait a bit. Cloud providers first need to figure out how many customers can open accounts as originally planned and which projects will have to delay signing. For a company that makes money by renting out racks as quickly as possible, a few extra weeks of review time could change when it dares to place the next batch of purchase orders. #美国拟扩大AI芯片出口限制,海外算力租赁或受影响
Oli.
Oli.
When BTC and ETH both fall, I first turn off one of the USD quote charts and switch it to ETH/BTC. Both can fall due to a stronger dollar and worsening risk appetite, and watching for a long time it’s easy to mistakenly think you’re tracking two independent trades. When a real rebound happens, whether ETH can outperform BTC better reveals whether the market is willing to take on more risk. Last Friday, the US BTC spot ETF still saw net outflows. The ETF doesn’t trade over the weekend, but the coin price moves anyway, and the time difference between the two markets can easily cause misjudgments. Even if ETH bounces more strongly over the weekend, you have to see after the US session opens on Monday whether the ratio can hold. If BTC stops falling first and ETH/BTC continues to dip, I interpret that as buyers only willing to return to the most liquid asset. If BTC holds steady and ETH can maintain relative gains through subsequent retests, then the market starts to get interesting. The most frustrating part of continuous pullbacks is that you finally guess a rise correctly, but the next day the coins in your hand give back the gains exactly. I can’t guess the bottom, and for me, there’s no need to guess the absolute lowest point. First, see if strength can spread from BTC, then see if it can hold through several sell-offs after spreading. If neither happens, I remain skeptical about any so-called turning point. #BTC与ETH持续回调,市场何时迎来转机?
Oli.
Oli.
In the discussion about Anthropic's IPO, the subscription multiple has almost overshadowed the other contents of the prospectus. I flipped to the page about voting rights and actually paused there longer. According to the preliminary prospectus design, the founder entity holding Class F shares has 50.1% of the voting rights on general shareholder matters. The outside investors are competing for Class A shares, so who can decide the company's direction after listing has been quite clearly arranged before issuance. This is especially worth pondering for a company with such large computing power expenses. If the management wants to sign a multi-year cloud service contract or continue to suppress profits to speed up model iteration, public shareholders can express their attitude in the market, but the space to change the course through voting is much smaller. Those optimistic about the founder team's long-term commitment might be willing to accept this deal. But if the subscription is too hot and the issue price is pushed up, investors should carefully consider how much operational autonomy they are willing to pay for. My interest in this issuance mainly lies after pricing. At that time, the excitement during subscription will settle into a price per share, so investors can better evaluate revenue growth, computing power expenditure, and the shareholder rights they can obtain together. I can understand paying a premium for growth. But if it's so expensive that even the management has no room for mistakes, I feel it's a bit overwhelming. #AnthropicIPO进展:获数倍超额认购,估值或上调
Oli.
Oli.
The US PPI rose by 0.3% month-on-month in October, below market expectations. It was originally a number that could ease concerns, but looking deeper, it feels a bit awkward. Commodity prices rose by 0.5%, services by 0.2%, with outpatient hospital service prices still climbing. The total looks better, but several items related to consumer inflation send mixed signals. This time, I’m more inclined to wait for the core PCE. Some medical service prices in it use PPI data, which differs in scope from the core CPI that everyone usually watches. The rise in outpatient services alone can’t explain the entire PCE, but it reminds me that using the total PPI to immediately cool down the December rate hike expectations might be premature. Especially since earlier data on employment and CPI were on the hotter side, the interest rate market needs more subcomponents to weaken together before there’s reason to change the original judgment. The most vulnerable time in the crypto space is when data comes in below expectations, BTC spikes, and people immediately treat it as if the rate trade has reversed. I prefer to watch whether the two-year US Treasury yield remains low by the time the US stock market closes. If short-term bonds recover their initial drop, no matter how comfortable BTC’s surge candle looks, I find it hard to have much confidence in it. #美国10月PPI低于预期,12月加息预期仍强劲