Oli.
Oli.
🍓Web3投研 🍑人工智能 🚀《干翻狗庄》系列工具作者
5Following
222followers
Feed
Feed
The current pressure on U.S. Treasury bonds can no longer be explained simply by saying "wait for the Fed to pivot." Reuters reported on October 1 that global bond markets continue to be under pressure, with borrowing costs in the U.S., France, and Japan reaching multi-year highs. Energy prices are driving inflation up, while AI and data center construction are competing for funds, keeping financing costs high for a long time.
This makes me pay more attention to those high-yield products on-chain. Previously, seeing a nice annualized figure made it easy to think that funds finally had a place to go. But when traditional dollar assets can also offer more attractive returns, you really have to carefully compare how much extra you earn by taking on smart contract and liquidity risks.
Of course, long-term bonds also experience price fluctuations due to interest rate changes and cannot be treated as cash. When comparing, you have to consider both duration and risk together, not just pick the higher percentage.
I have no bias against on-chain yields; on the contrary, I hope they become more solid. Interest paid by borrowers and fees generated by real transactions can be discussed; if the main support is extra token subsidies, then we have to keep asking how many people will remain once subsidies decrease.
The most annoying thing about high interest rates is that they make capital picky. No matter how good the project story is, it must explain why users are willing to take on an extra layer of risk. Now, pages that only show annualized returns without explaining the source of the yield don’t excite me as much.
#美债收益率频创新高，长期利率压力未缓解
First, an update on a detail: the nine consecutive days of net inflows into BTC ETF were interrupted on September 30, with a net outflow of about $149 million that day; the previous nine trading days had a cumulative inflow of about $3.1 billion. On the same day, ETH ETF also recorded a net outflow of about $59.6 million. The trending list still shows continuous inflows, but the fund records have already turned a page.
I don't think a single day of outflow is enough to overturn this round of recovery, but this event is a timely reminder: institutional funds will still adjust their positions. They have budgets and deadlines when buying, and when facing rebalancing, redemptions, or risk limits, they will sell. We can't see the motive behind every transaction, so we can't call all inflows long-term allocations and all outflows short-term noise.
BTC and ETH data need to be viewed separately. BTC's previous capital attraction doesn't prove that funds will necessarily follow a fixed route, with the next stop automatically being ETH. That script sounds good, but the funds are under no obligation to cooperate.
I prefer to observe whether the price can hold steady after outflows and whether new subscriptions follow. Continuous inflows with rising prices are easy to understand; after a temporary contraction in buying, if there are still willing buyers, that better shows market support. Relaxing vigilance just because you see the word "institutional" will ultimately leave your own account bearing the drawdown.
#比特币ETF连续9日流入，ETH转流出
Tonight's focus is on the nonfarm payrolls, but don't rush to label the data as either positive or negative. The initial jobless claims in the U.S. dropped to 197,000 on October 1, below market expectations; Reuters also pointed out that companies remain cautious about expanding hiring. This combination is quite awkward: those already employed are temporarily stable, but those looking for jobs may not be doing well.
Few layoffs indicate that companies are still holding on; fewer hires mean companies are not so confident about the coming months. If the nonfarm payroll additions are weak and initial claims remain low, I would tend to interpret this as the labor market entering a low mobility state, which does not directly signal a recession for now. For the Federal Reserve, this state may not be enough to eliminate inflation concerns.
What’s most frustrating in the crypto world is translating every piece of economic data into the same sentence: liquidity is coming soon. If employment is slightly weak, they cheer for policy easing; if employment is slightly strong, they say the economy is resilient. In the end, any outcome can be explained as a reason for prices to rise.
This time, I want to see whether the new jobs have expanded into more industries or if the total number is maintained by just a few sectors. If finding jobs becomes increasingly difficult while price pressures persist, policy shifts will become more conflicted. The first candlestick tonight might be lively, but I’m unwilling to draw conclusions about the entire report based on just those few minutes of sentiment.
#加息预期推迟，9月非农成下一关键
🟢 Oli Daily Brief｜2026.10.02
BTC has returned to $84,600, but this round of gains cannot yet be defined as a full rebound. The reason is simple: BTC and ETH are rising, but the total market cap is still declining; the latest complete ETF data shows net outflows; and US inflation pressure is rising again. Tonight at 20:30 HKT, the US non-farm payrolls report will likely determine the market's next phase direction. 📊 BTC rebounds, but altcoins have not fully followed As of 05:03 HKT: BTC $84,608, 24h +1.13% ETH $2,697.78, 24h +0.62% SOL $118.11, 24h +0.04% Total crypto market cap: $2.896 trillion, 24h -1.90% BTC dominance: 58.64% Fear and Greed Index: 74 — Greed The most notable data here is not BTC's 1.13% rise, but that BTC rose while the total market cap fell by 1.90%. This indicates that funds have not broadly spread across the entire crypto market. BTC dominance remaining above 58% also means current funds still favor top-tier assets. So the more accurate current market structure is: BTC-led recovery + altcoin divergence + localized capital rotation. Not: a full risk-on. The divergence within mainstream coins is also very clear. UNI: +3.18% NEAR: -8.39% NEAR, which was strong yesterday, has directly become weak today.
While Strategy bought BTC in the latest week, it also repurchased approximately $151.7 million in STRC preferred shares. The repurchase amount was even slightly higher than the expenditure on buying coins that week, and this detail attracts me more than just the increase in total holdings.
The two expenditures serve different purposes. Buying BTC increases asset exposure, while repurchasing preferred shares involves financing structure and future dividend burdens. The company needs to consider both the price of the coin and whether the securities it issued are worth buying back.
This makes me feel that crypto treasuries can no longer be studied solely as "large coin holders." They are also financing entities that need to decide when to issue stock, when to repurchase securities, and how much cash to keep for expenditures.
For common shareholders, these decisions may be equally important. If funds are only focused on continuing to buy coins while financing arrangements become increasingly strained, no matter how impressive the company's holdings are, shareholders may not feel comfortable. Conversely, appropriately adjusting the preferred stock scale may also improve subsequent funding arrangements, with specific effects still depending on price and terms.
I would not declare the treasury model mature based on a single repurchase. But at least it shows the company is not just taking one type of action. When the market focuses on whether it continues to buy BTC, it is also worth looking at what is happening on the other side of the balance sheet.
In the future, when reading treasury announcements, I will consider securities issuance and repurchases together. Looking only at the coin purchase line can indeed easily miss decisions that affect returns.
#Strategy再购BTC，多家财库同步增持
NVIDIA has authorized an additional $150 billion stock buyback, while on the other side, model companies continue to raise funds and sign computing power agreements. Putting these news items together makes the differences within the AI industry much clearer.
Some need to continuously purchase equipment and services to gain more users; others sell this equipment, and after receiving cash, can arrange shareholder returns. Both sides rely on AI demand, but their operational pressures and cash flow situations differ.
Therefore, I am increasingly reluctant to treat the entire industry as a single investment. Good model performance does not immediately mean good operating profits; strong supplier orders do not necessarily mean customers can afford all the investments in the future. When researching, one must connect the accounts of both sides.
NVIDIA's large buyback is a sign of management's confidence in its cash-generating ability. It certainly matters to its shareholders. But if this confidence is extended to "all AI companies deserve higher valuations," the evidence is insufficient.
What makes me more cautious this time is portfolio allocation. Looking at several companies bought, with different names, they may all ultimately rely on the same group of customers to continue expanding capital expenditures. On the surface, it looks diversified, but the sources of demand may not be.
I still have expectations for AI development, but when buying stocks, I first distinguish who pays the money and who can keep the profits. After the industry's investments grow larger, this distinction will only become more important.
#英伟达追加1500亿美元股票回购
BTC spot ETFs had a net inflow of nearly $2.4 billion last week, and today I want to break down the total. According to daily data from Farside, BlackRock's IBIT and Fidelity's FBTC together contributed about 80% of the net inflow.
This concentration makes me feel that while funds are returning, they are also choosing their entry points. Fund size and client channels may influence where the money goes, but based on the flow table alone, we cannot determine why each buyer chose it.
For BTC, buying through any fund will create corresponding spot demand. However, this distinction is quite important when judging whether demand is broad. Multiple products receiving subscriptions together indicates a different situation than inflows mainly contributed by two large funds.
Moreover, outflows from one fund and inflows to another may sometimes involve adjustments between products. We can see changes in fund shares, but we cannot directly read all investors' intentions from the table. Writing every inflow as new institutional bullish conviction would be overly confident.
I acknowledge the strength of this buying wave, but when researching, I want to look at more lines of data. If more funds continue to receive subscriptions later, it will make the demand improvement more convincing; if it remains concentrated in a few products long-term, it is worth paying attention to allocation changes in these channels.
The total in the flow table grabs attention, but differences between funds are often more worth reading.
#BTC现货ETF周流入创近一年新高
OpenAI is reportedly discussing at least $30 billion in new financing, targeting a pre-money valuation of about $1.4 trillion. The amount is large, but it is still in financing negotiations and cannot be considered a completed deal.
What makes me curious is what exactly the investors are prepared to buy in this round. Ordinary readers seeing the valuation might easily imagine everyone entering at the same price and taking the same type of shares. The specific rights arrangements of private financing, however, can only be judged after the terms are disclosed.
For example, whether the shares have preferential rights, and how they convert upon sale or listing, will affect the actual returns. Knowing only the $1.4 trillion valuation is far from enough to determine whether an investment is worthwhile. It also cannot be used directly to price all "OpenAI-related assets" in the market.
From the company's perspective, continuing private financing has its conveniences, allowing it to supplement funds before going public. But for those waiting for public financial information, this also means continuing to rely on scattered reports to judge how much profit revenue growth can actually retain.
I am interested in product progress, but not so easily excited by constantly refreshed valuations. Successful financing certainly gives the company more confidence to invest; whether investors make money depends on the entry price and subsequent operations.
This time, I prefer to wait for actual transaction news. It would be best to see more financial data and financing terms then, and less discussion focused solely on valuation rankings. The price of a company deserves to be studied more seriously than just by ranking lists.
#OpenAI拟1.4万亿美元估值融资300亿美元
According to The Information, Anthropic disclosed a SpaceX computing power agreement with a maximum amount of up to $84.5 billion. The word "maximum" must be retained and should not be directly written as the amount already paid.
More interesting than the amount is the relationship between the two cooperating parties. SpaceX owns xAI and is developing its own models; Anthropic, however, needs its computing power. This indicates that despite fierce competition in models, the currently deliverable computing resources remain sufficiently scarce, and business cooperation must continue as usual.
Anthropic has previously confirmed using the full computing power of Colossus 1, while also using AWS Trainium, Google TPU, and NVIDIA GPUs. For a model company, this choice is very pragmatic: whichever platform can provide capacity on time is worth serious consideration.
I actually think this makes AI competition more concrete. Besides model performance, companies also need to solve whether capacity is sufficient and whether services can run stably. Users will not accept waiting in line indefinitely after paying just because your relationship with the supplier is complicated.
But purchasing from suppliers who also develop models raises concerns about how business operations are isolated and whether there will be enough options during contract renewals. These issues cannot be glossed over with the phrase "strong alliance."
This agreement gives me a more intuitive sense of the bargaining power of computing power suppliers.
#Anthropic披露845亿美元SpaceX算力协议
Trump signed an executive order requiring U.S. administrative departments to rename AI as SI, meaning "Super Intelligence," in official communications and websites. It sounds like the entire industry has suddenly upgraded.
But reading the definition section of the executive order made me want to laugh a bit: at this stage, SI still refers to AI technology covered by the existing legal definition. The document also requires new legislative definition proposals within 60 days. The name changes first, but how to define this technology still needs further study.
This move can certainly generate attention and may change government propaganda and corporate marketing wording. But for ordinary users, opening the same model tomorrow won’t automatically reduce error rates just because the abbreviation changed; price and capability still depend on product updates.
What worries me more is that the word "Super" might raise people’s expectations of system reliability. Many are already willing to entrust important decisions to models, and if the name implies it far surpasses ordinary intelligence, users might check results less.
The industry needs to explain capabilities and also explain limitations. What truly makes me willing to increase trust are repeatable tests, clearer responsibility arrangements, and having someone to address issues when they arise. The new name in the executive order can be discussed, but treating it as a technological breakthrough is a bit ridiculous.
In the future, when SI appears in corporate promotions, I’ll probably take a closer look: what exactly has the product updated, or did they just update the introduction page.
#特朗普签署行政令将AI更名为SI