趋势感知

$ETH The current trend has been hovering near a resistance zone. Liquidity is drying up. If everyone is shorting, be careful. This level is where long (bullish) strength is gathering and forming consensus, so a violent surge of 7%–15% is easy to trigger. Be cautious about shorting.
$BTC Also, BTC is relatively weak. The upside room isn’t clear; if it does rise, this leg of the rally should be driven by ETH.

Jeff Dean resigns, must post overnight. The super idol at work for more than 10 years—my architectural design philosophy is largely inspired by Jeff Dean. He has carefully read every engineering paper he’s written; Jeff can be said to be a pioneer in internet technology (distributed computing frameworks, big data computing frameworks, AI, etc.).

Since Jeff joined Google in 1999, many of Google’s core architectures and technologies have come from Jeff. In terms of engineering output, he is the most prolific scientist in internet infrastructure technologies and AI foundational architecture—no one else comes close. Therefore, his departure caused Google’s stock to drop by 7% almost immediately.

After Jeff left, he will found an AI company focused on scientific discovery: Discovery Loop. You can think of it as a vertical version of Claude Code and Lobster (we can consider Lobster to be Agent Loop). But this Agent focuses specifically on scientific discovery—automating scientific and technological R&D exploration (through AI autonomous thinking and planning), and accelerating the efficiency of fundamental science exploration.

Previously, people have kept asking me what startup opportunities in AI exist on-device. I’ve always believed that, at the moment, all-purpose Agents and their tools that assist white-collar tasks don’t really make sense. On the one hand, some capabilities can be directly embedded during the post-training stage of large models; on the other hand, AI’s own path is to replace people, while the business model of Agents that assist humans is directly disproven. The characteristics of on-device startup opportunities might be: combine hardware + models, then go deep into a vertical area within a specific industry chain—e.g., Coding (Coding Agent Loop), Jeff’s Discovery Loop focused on scientific discovery, embodied intelligence (Tesla Optimus robots).
$GOOGL

$AMZNB Amazon said there’s no need to worry that huge compute-capital expenditure won’t pay back. Based on calculations, the payback period is 3 years. After 3 years, the investments are marginal returns. So tonight, Amazon surged 15%, which also lifted Alibaba. However, Alibaba is still following the logic I mentioned earlier: an AI-themed narrative plus a technical rebound. In the long run, AI’s moat may be strong, but personally I don’t feel optimistic. Wait until the rebound reaches around 140, then use short-selling to hedge positions in AI-bullish targets. (Mainly two benefits: first, other AI-related concept stocks may rise more than Alibaba—for example, Google. Second, when the AI narrative hits short-term resistance and falls, those stocks’ downside will certainly exceed that of other companies. So Alibaba is a relatively good hedge candidate as an AI-narrative concept stock.)

The ceiling on monetizing knowledge is probably best exemplified by the 165-page long article “Situational Awareness: The Decade Ahead” published by Leopold in 2024. At the time, the 23-year-old won $250 million in funding by deeply quantifying the AI development path through a single paper. He then invested in AI compute infrastructure; by around April 2026, the scale reached $24 billion, and at the July peak it hit $45 billion. It took just two years. Unfortunately, yesterday, due to a semiconductor selloff and excessive leverage, he was liquidated. The remaining $16 billion position was taken over off-exchange by a securities firm. Even more heart-wrenching: the day after he fell—today—the semiconductor sector violently rebounded, with the average gain of related stocks reaching 30%, and the KOSPI (Korean index) surged 18% (an unprecedented, huge jump in history).
Why could he get $250 million in investment based on just one paper? What exactly did the article write? Let me break down the key points:
The article quantifies the capabilities of large models as: model intelligence ≈ algorithmic progress × compute scale × data quality × engineering efficiency. And among these, compute is the most important. This ties to the famous scaling laws from the OpenAI paper at the time—namely the double-exponential law. From the standpoint of compute, we can understand it as: exponential growth in compute can lead to linear growth in model capability, and ultimately achieve AGI. So I summarized several core predictions from the article:
1. Over the next 10 years, we will enter a massive capital expenditure cycle. Compute spending will grow exponentially. The prediction was that 2026 capex would exceed $500 billion, and that it would expand at a 2x pace, reaching $800 billion by 2030. In reality, this year’s capex plan is already close to $1 trillion (US/Korea), far exceeding his prediction.
2. For a single training cluster, he predicted it would reach “tens of billions of dollars” in 2026—around 1 million H100-equivalent GPUs. Due to physical constraints, a single cluster today can hardly reach that; the maximum scale is about 200,000 GPUs, below expectations, but the direction was completely correct.
3. He predicted that around 2027, AGI might be achieved—and that programmers would no longer need to manually write code. This is too aggressive. Programmers might not need to write code manually by 2026, but AGI could be delayed to around 2030.

For investment purposes, the other points aren’t that important, so I won’t go into them.
$SNXXB

Last night, the Federal Reserve did not raise interest rates, in line with expectations, but the QQQ didn’t rise. Now QQQ is extremely sensitive. Earlier I mentioned that by the end of the month I would be watching conditions; it seems that no clear rebound signal has appeared yet. Instead, the market is full of panic. Positive news can’t spark a rally, and negative news will inevitably trigger a sharp drop. This is a typical bear-market pullback signal. Currently, the QQQ Nasdaq index has already pulled back by 10%. If it goes through a secondary large-cycle pullback, then there’s still about another 10% downside. The market is ever-changing and unpredictable, so we’ll keep the “bottom hold” unchanged. If we gradually add positions, we should be a bit more conservative at first—wait for clear rebound signals before we enter with full weight.
Also, the fundamental story behind AI remains essentially unchanged. In my view, what we’re seeing is just capital taking profits for the moment and emotions driving the selloff. The market worries that AI compute-capex won’t translate into corresponding returns on the application side. After all, on the capital side, spending is on the scale of tens of trillions of dollars, while the application side doesn’t yet have a matching revenue structure—this concern about a bubble is normal.
But in the long run, the AI trend is unstoppable. A short-term pullback is precisely our opportunity to add positions. We’re betting on AGI—ASI will ultimately arrive!

Changxin Technology will officially be listed on the STAR Market next week. I placed an on-chain short bet at $7.9 a week ago. That corresponds to the A-share offering price being up by 6.2x. Currently I’m up 24%. So, if on the 27th the closing price rises 6.2x relative to the offering price, what loss would I face? On that day, at what price would the closing be, and by how many x would it be above the offering price?
Also, in the short term, I’m mainly short; in the long run, as AGI gradually breaks through the barrier and Changxin’s technology keeps making breakthroughs, a value of 100 trillion isn’t impossible (everything hinges on AGI being created: physical-model technology, autonomous driving, and embodied intelligence become fully普及, and more than 50% of office workers’ jobs are replaced by AI).
Lastly, a quick rundown of Changxin’s fundamentals:
Changxin Technology—Hefei’s high-tech benchmark enterprise—mainly produces DRAM. HBM is still under development, and it’s expected that meaningful returns will take until 2028. In terms of DRAM technology, it’s 2 generations behind the three giants. Its current market share is about 7% (Samsung, Micron, and SK hynix account for 90%). Production costs are higher than …
For ownership, the Hefei municipal government holds 44%. You have to admit, Hefei’s government has the backbone. According to the prospectus, this year profit is expected to reach around 120 billion RMB. For 2026 Forward PE, using the offering price, it’s about 5x; in the same period, SK hynix should be around 5.6x. But I believe the two companies are not truly comparable. If the capital market were fully open—no current sanctions, no trade war—and storage is valued strictly under institutional capital’s valuation logic, it should be around 3–4x PE. Therefore, if we don’t consider Changxin’s technology lag, the market cap should be around 300 billion RMB (PE valuation: the US > South Korea > China).
In addition, Changxin’s profit mainly comes from price increases. If prices revert back to the pre-price-increase level, Changxin’s profit could drop from 120 billion RMB to under 10 billion RMB. This is why storage-industry valuations are relatively low: the industry is highly cyclical, and it’s also a standardized product. Once everyone ramps up production capacity, and if demand doesn’t keep growing, competition will cause profits to fall exponentially, even turning into losses.

#美联储料7月29日维持利率不变 Without a doubt, the most likely outcome is that things will remain unchanged. I analyzed the Federal Reserve’s decision in a livestream before. Next, I will analyze it from two dimensions: the standard answer and the non-standard answer:
In June, the FOMC kept the federal funds rate at 3.5%–3.75%. Currently, the core CPI is around 2.6%. Nominal interest rates are already higher than core inflation, so by this measure, monetary policy is not loose.
Of course, some people might argue that the U.S. government pays $1 trillion in interest on Treasury bonds every year—more than military spending—which accounts for 15%+ of fiscal expenditure, and that the total size of Treasuries is already around $40 trillion. They might say that rate hikes could accelerate a collapse of U.S. Treasuries. But I want to say that none of that is important—this isn’t the core point. The fundamental logic of how the global economy operates is within the dollar system. Treasury bonds equal dollars; the larger the Treasury market, the stronger the dollar. Of course, this requires a delicate balance—like an aerial acrobat performing on a tightrope. If you can’t control it, everyone knows what the consequences will be. The FOMC is the key institution that maintains this balance. Therefore, the focus isn’t on how much interest or how large the Treasury balance is; what matters is maintaining balance. For the U.S. government, the size and the interest have never been the most important factor; it’s not a single indicator. It’s about whether the overall set of indicators and their linkages are healthy (for example: the U.S. nominal GDP, fiscal revenue, global GDP, and the fact that global GDP growth also reflects the strength of demand for the dollar; improvements in global technology productivity increase dollar demand; inflation expectations, and so on).
Also, to maintain the balance I just mentioned, I believe the FOMC will definitely take action. In this critical period of competition over national destiny in the AI narrative, the probability of rate hikes is relatively low. Over the next few years, enterprises’ AI capital expenditures will shift from corporate earnings to financing and debt. In this situation, rate hikes could cause the AI bubble to burst early. Currently, U.S. AI compute capital expenditures exceed 700 billion u; next year, they are expected to reach 1,000 billion u. The scale is huge. Therefore, I think the room for action includes: structural divergence in CPI data—for example, independently tracking the AI industry chain; a mild balance-sheet reduction (quantitative tightening); and using tough talk (a strategy commonly used in the “wolf is coming” style—saying it but not doing it, which reduces market expectations, especially in the stock market, to prevent it from getting overheated). In short, they will definitely do something, but the likelihood of rate hikes is low.

The market slightly rebounded, and profits filled the entire screen. In the current period of extremely volatile trading, is AI trend quant still usable? For example, the Korean stock index: $KORU . Recently, over the last 7 days, when statistics were taken at the 4-hour dimension for volatility, more than 80% of the K-lines exceeded 5%, and the total volatility exceeded 300%. That means in a perfect scenario, you could capture profit from every fluctuation, potentially earning 300%. But in reality, when I analyzed my quant orders, and calculated the profit ratio based on the mid value of the position, I only captured about 28.5%—so I didn’t even reach 271.5%. The efficiency is below 10%. Therefore, volatility efficiency will be the “North Star” metric for optimization going forward.
Of course, it’s not all without advantages. The risk control is simply too strict—positions and entries are conservative—resulting in lower profit margins. Naturally, the drawdown rate is also lower. In recent weeks of intense market choppiness, the maximum drawdown over the last month has not exceeded 2%. Monthly compounded annualized return is 150%, and the Sharpe ratio is 2.1, which is considered excellent. With this Sharpe ratio, it’s entirely possible to appropriately increase the entry size per trade so that the compounded annualized return can be raised to 300%. All open-and-close orders are clearly visible in the platform’s public-domain paired-trading project; if you’re interested, you can download and study it yourself. (Given that over the past week, Korean semiconductors have seen daily swings exceeding 10%, and yet the overall return rate still hasn’t reached 10%, it’s indeed too low.)

$BTC Short follow orders, multiple fills, no short momentum. The market is currently entirely driven by long momentum (but very weak). Related crypto stocks have all rebounded sharply, $CRCL $MSTR , meaning that in the short term the 4H chart has formed a long-term resonance. First, look for above 68000

#韩国散户杠杆持仓降至三个月低点
A few days ago, when I said that the QQQ technical setup might pull back, I shorted QQQ Nasdaq. Yesterday it didn’t continue to fall. I’m now considering whether to take profit. Currently, the 4-hour trend is in chaos and it’s hard to see clearly.
Yesterday I “ate the瓜” from Big V and also went long on the Korean index and semiconductors—Samsung and SK Hynix—and they’ve started to show some profit. Now I’m thinking about whether to take profit on part of the position or continue betting on a rebound. It’s unclear whether the market has bottomed out. I checked some related data yesterday: the rate at which foreign capital is exiting hasn’t slowed down—there are still tens of billions of dollars in net outflows every day. Korean retail leverage is still very high, not low. Compared with the June peak, it’s fallen by about 25%. Although leverage has declined, it’s still far above the level at the beginning of the year

After the recent sharp drop in semiconductors, a VIP member of Ren Zeping got liquidated and surged to the top of the hot search. I briefly looked into it. This Zeping isn’t an ordinary person. He has a PhD from Renmin University, and his work credentials are impressive: he served as a chief economist at multiple securities firms. He was even invited by Evergrande’s Xu Jiayin with a 15 million yuan annual salary to become Evergrande’s chief economist. I checked his official account—his content mainly focuses on macro trends, and he’s very sharp. In general, he only aggressively calls trades after a trend has already formed (in other words, hindsight “proving” after the fact). For example, this year: after gold had already been falling for a few months, he started saying gold was in a declining cycle and told everyone not to add positions. Then once AI semiconductors started rising, he began aggressively calling for semiconductors. (He claims the impact of the AI revolution is 100 times that of the Industrial Revolution; that pullbacks are “golden pits”; that “a thousand pieces of gold can’t buy a bull’s return.” His language is extremely aggressive. This may cause many of his VIP members to experience FOFO—fear of missing out while still positioned—and take overly aggressive positions. I wonder whether Evergrande’s earlier aggressive style was influenced by his aggressive style… hhh)

After all that gossip: since he’s basically a retail investor and all low-leverage positions got liquidated, does that mean leveraged capital has already been forced to sell at low levels, and whether selling pressure has reached a stage peak? Is the downside space limited going forward? I’m planning to start accumulating some long positions in South Korean semiconductor stocks.

$AKE , Continuing my previous logic, we have already reached an interim goal. We’ve currently pushed up by 10x. Next, we’ll look toward 20x—i.e., price moving toward 0.004. The main force is still at work; don’t worry, just follow along. But be careful with leverage. It’s possible they may first drive the price below 0.001, though the highest price will most likely break 0.004. As for whether to go toward the 100x target next, it depends on the K-line trend.

$QQQ.ETF Has the Nasdaq index 4h technical setup already formed a bearish trend structure?
In the past half month, bullish momentum has consistently been trying to break higher, but it has been tightly suppressed by the trendline the whole time—this suggests that over the recent period, bullish capital has been exiting.
On the fundamental side, the semiconductor sector is still in a downtrend. The AI narrative has been ongoing, but massive capital expenditures from cloud providers have made the market worry about future profitability. The AI application layer has not yet demonstrated strong profitability, which further amplifies market concerns. In addition, the SpaceX listing has pulled liquidity from the market, OpenAI has postponed its IPO, tensions in the Iran–Israel geopolitical conflict have intensified, and CPI inflation has warmed up—these are all factors that have weakened the recent stock index.
On the price action, bullish and bearish momentum is also fading. After a prolonged back-and-forth between bulls and bears, if price cannot break upward and bullish momentum runs out, it could lead to an even larger pullback. Recently, do not take an oversized position; wait until the end of the month to see the K-line develop, then observe the market outlook structure.

$AKE has a big main force in operation, and it should still be able to move higher. I entered late, so I bought a little casually. Let’s see if I can try to bet on another ~3x from the current level, meaning an overall move up of about ~15x. It’s pure math-probability betting, with no technical logic.
Let me add more of my thoughts:
1. The “shanzhai” (copycat) main coin being traded, LAB, this round has already ended. The market liquidity has been released. Refer to my earlier judgment about LAB (the main force’s last harvest, preparing to withdraw from the pool). Then there’s a chance that another super “whale-controlled” coin will appear (the logic is simple: in a bear market, it’s hard for multiple super whale-controlled coins to show up at the same time; liquidity is insufficient). This is the opportunity.
2. Then AKE. From a technical expert perspective, I looked at the trading team behind it. The K-line they drew is quite impressive—not something an ordinary quantitative team could draw. If you’ve worked in real industry, led technical teams, or been involved deeply, you can tell the underlying technical team is top-tier in the world, and that they also have the financial strength (my guess is they belong to the Binance top-tier trading team).
3. The third point continues from the second: the bullish momentum is showing very strong strength. It doesn’t look like a typical “shanzhai” harvest where they just pump 2x–3x. Instead, a single 4h or 1h K-line blasting the shorts and running a single-wave harvest usually yields limited profit—because the 4h timeframe can only consume the liquidity currently sitting in the order book. Of course, if you run into a coin like this, don’t chase it. It’s not a hundred-bagger coin that’s meant to be the next round of manipulation by the main force. But with AKE, it’s clearly built on the foundation for a long-term pull-up and repeated harvesting (they already have capital, technology, and liquidity—there’s no reason for them not to act).

#美国6月CPI降至3.8%
At the early hours of July 15 Beijing time, the U.S. June CPI year-over-year came in at 3.5%, below expectations of around 3.8%-3.9% and also below the prior reading of 4.2%. This was mainly due to a decline in oil prices. As a result, although the Nasdaq rose, momentum was relatively weak. Recently, however, oil prices have been quickly driven higher by the U.S.-Iran conflict, which will affect subsequent CPI data. Since early June, QQQ (Nasdaq) has been in a 45-day pullback, and the 4-hour chart formed a converging triangle with a technical pattern breakout signal. Combining fundamentals, I think it is likely to break upward, but with limited momentum—meaning the upside may not be very large. You can buy on the long side; take profit at new highs. Bet on the next month: QQQ breaks to new highs.

$BTC BTC You went early. Today, in the @听澜321 live room, I mentioned that I’m bullish toward the area around 64000, and then consider whether to do a short sell near there. I initially intended to take advantage of the range: if it drops, continue going long. I planned to look toward around 64000 first, but I sold too early—missing the move. In the short term, the market is just like that: there’s always a 20% chance that the market behavior will be outside expectations. In other words, two standard deviations of volatility can lead to volatility that’s beyond what’s easily predictable. That’s a universal rule of the universe—no one can predict it. So that’s why you should understand why you shouldn’t trade the spot market rashly, right? It’s easy to sell too early in a bull market. Of course, this is a contract—so it doesn’t really matter.

$KORU $SKHYNIX 4h Market structure and overall trend: I started recommending yesterday that you begin building your positions gradually. At this moment, don’t be afraid of the drop—others are fearful while you’re greedy. Fundamentally, there’s nothing wrong. This round of decline in the Korean stock market is driven purely by liquidity and sentiment.

In terms of liquidity, it’s mainly the leveraged funds that are being sold off in a chain reaction. Koreans really like playing with leverage. Recently, the 2x single-asset leveraged product has gotten so out of control that it’s driving everything crazy. This kind of leverage amplifies volatility—when prices rise, it creates a spiral upward force; when prices fall, it creates a spiral downward force. Why? You can ask AI, and it will give you the answer.

Why hasn’t the fundamental picture changed? The logic is very simple: the AI narrative is still intact. If the AI narrative were to collapse, the first things to collapse would be U.S. stocks—NVIDIA, TSMC, and Google. So if the upstream hasn’t collapsed, but the downstream is collapsing now—does that sound reasonable? That’s why I say it’s purely liquidity and sentiment-driven. Also, over the past year alone, the Korean stock market has risen 4 to 5 times. The profitable positions are now somewhat crowded and need to be released. All of this is normal.

$SKHYNIX $KORU left side standing still short and many,
$BTC continue to short, 15f line continues to look downward, the target is not determined yet—wait until it drops and then we’ll talk
$
Betting on the semiconductor index to stabilize and rebound this week; follow the trend’s rhythm

$BTC I’m amazed by the things these old 6 are saying in their朋友圈 (circle)—they say BTC has a bit of “old-school old geezer” vibe, and ask what young people are playing. Come to think of it, that actually seems true. Back then, BTC was mostly played by teens and people in their 20s. Lately, I hardly ever see kids around 20 playing crypto anymore. So is it true now that it’s really the “middle-aged folks” and “old geezer” crowd who are playing? Prove you’re young—drop a comment.