小楼

The war for altcoin ETFs is over.

Together, the SOL ETF and the HYPE ETF account for 80% of all altcoin ETF trading volume (excluding BTC/ETH). SOL’s AUM is $904M, and HYPE’s is $350M. The remaining few hundred altcoins share just 20%.

Do you think institutions research the value of each altcoin one by one? No. Institutions care about only two things: liquidity and brand. SOL has ecosystem scale; HYPE has social capital. As for the rest—LINK, AVAX, DOT, NEAR—at the ETF level, they don’t even deserve to have their names on the list.

SOL ETF saw net inflows of $5.83M in a single day, and a cumulative $1.146B. But look at SOL’s price: up 1.17% over the last 7 days, $77 → $78. With $1.146B in institutional money flowing in, the price doesn’t move at all.

What does that mean? The buying power of ETFs is fully absorbed by the market makers’ arbitrage and hedging. Retail buys the SOL ETF → the market maker buys SOL spot → simultaneously shorts as a hedge in the futures market → the price stays flat. Your money goes into the ETF, but the price doesn’t rise—because institutions are “smoothing” your buys for you.

“ETF-ization” of altcoins isn’t a signal of “institutional entry”; it’s a signal that liquidity is being siphoned off.

#Solana $SOL #ETF #altcoin

Open today’s cryptocurrency futures trading volume ranking: BTC $48B, ETH $32B—normal.

3rd place: SNDK (SanDisk) $5.4B. 4th: SOL $5.2B. 5th: BANK $4.7B. 6th: SK Hynix $3.2B. 7th: SOXL $2.9B. 8th: XAU (Gold) $2.8B.

6 out of the Top 10 are not cryptocurrencies.

Do you think crypto futures platforms are exclusive casinos for the coin world? They’ve already stopped separating crypto from TradFi. SanDisk’s daily futures volume is higher than SOL’s. BANK (bank index) volume is nearly twice that of XRP. Gold is running $2.8B in daily volume on a crypto futures platform—you don’t need to go to CME for gold; you can do it where you trade BTC.

This isn’t “tokenized stocks.” This is a merger of trading venues. When crypto-native perpetual contract platforms start devouring the volumes of traditional financial derivatives, the boundaries disappear. The upside is liquidity; the downside is you can no longer make decisions by looking at crypto data alone—SanDisk’s earnings and BTC’s funding rate are now influencing each other in the same order book.

#DeFi #加密期货 #TradFi

BlackRock withdrew 1,789.6 BTC from Coinbase Prime, worth $119M.

That same morning, an anonymous address sold 1,862 ETH for $1,923—bought five months ago for $2,685, down $1.42M and cutting a 28% loss.

The world’s largest asset manager is withdrawing coins; the world’s smallest retail traders are cutting losses. Same market, same direction (up)—two different actions: one is hoarding, the other is running.

BlackRock’s $119M withdrawal is most likely routine ETF settlement—they don’t care about today’s price because they’ll be buying again tomorrow. And that ETH retail trader selling at $1,923 is exactly $100 above the ETH 7-day low of $1,821—he made it through the deepest part, then let go just before dawn.

This isn’t a matter of luck—it’s a structural issue: institutions’ capital costs are perpetual (ETF management fees), while retail traders’ capital costs are time-limited (emotional capacity). What you can’t outlast isn’t the market—it’s yourself.

#Bitcoin $BTC #Ethereum $ETH #聪明钱

ETH long/short ratio is 1.70, BTC long/short ratio is 1.03.

In the same exchange, with the same batch of users, at the same second—people betting “up” on ETH are betting “down” on ETH at 1.7 times the size, but for BTC it’s almost a 50/50.

This isn’t disagreement; this is split personality.

ETH is up 3.34% this week, slightly more than BTC’s 2.72%. But the speed at which leveraged longs pile up far exceeds the speed of price rising—an L/S of 1.70 paired with a 1.15% annualized funding rate means longs are adding leverage for free. When funding is close to zero and L/S spikes to 1.70, there are only two possibilities: either shorts are waiting for a breakout to reverse and harvest, or market makers are using zero-cost incentives to encourage you to keep adding positions—then pull the plug when you’re most comfortable.

BTC’s 1.03 is the real consensus: nobody has conviction. ETH’s 1.70 is noise—the more unified the “belief,” the closer it gets to a reversal.

#Ethereum $ETH #Bitcoin $BTC #Leverage structure

Day by Day One More Step: TSMC’s U.S. investment ramped up again to $265 billion. On the same day, Kimi K3 released open source with 28 trillion parameters, and its API price is one-third of Claude Fable 5. Anthropic permanently opened access to Fable 5 but cut the quota to 50%, and compensated users with 100 units—this shows the impact of K3 has already become real.

Put the three clues together:
- TSMC pours $265 billion into expanding production in Arizona—an arms race on the AI chip supply side
- Semiconductor ETFs saw $46 billion of inflows this year—capital is adding positions
- K3’s price has dropped to one-third at the door—demand is scaling up, but unit prices are collapsing

AI investment is shifting from “who builds the strongest model” to “who builds a strong enough model for the lowest cost.” The profit of front-end models is being squeezed, and profits are concentrating in the infrastructure (TSMC, HBM) and downstream applications.

The Gold Rush isn’t driven by people digging for gold—it’s driven by those selling shovels. But now even the shovel sellers are getting squeezed by price competition.

#AI #台积电 #Kimi

Daily Effort One Enlisted: The SK Group Chairman Says Next Year’s AI Semiconductor Demand Will Rise by +60–100%, Overall Semiconductor Demand by +50–60%, and the Amount of New Supply Is Almost Zero.

The supply-demand gap will continue to widen. Material supply and facility construction have become new bottlenecks, and even submarine cables are in short supply.

On the same day, the U.S. Semiconductor ETF attracted more than $46 billion in 2026 inflows, with expectations of setting a record for the largest annual net inflow—more than double the total over the past eight years. Just last week, it saw inflows of $2.3 billion.

Retail investors are chasing while institutions are buying. The KOSPI fell 5%, triggering the sidecar, but the semiconductor ETF is still pulling in money. The memory fundamentals haven’t changed—what has changed is the composition of participants and leverage. The valuations boosted by retail investors using 3x leveraged ETFs were dismantled by regulators, while institutions are steadily picking up shares at the bottom with real capital of 40 billion won.

The more people who shout about the peak of the semiconductor cycle, the more it indicates that we haven’t reached it yet.

#半导体 #HBM #AI demand

Rise one more step, every day: South Korea releases the "Korean Won Internationalization Roadmap"—offshore won settlement network to enter trial operation in September, with full launch scheduled for January next year. At the same time, it will build digital-asset payment infrastructure, paving the way for won-denominated stablecoins, and start a government bond tokenization pilot next year.

On the same day, the KOSPI fell 5%, triggering the sidecar mechanism. One side is the capital market collapsing, and the other is upgrading financial infrastructure.

South Korea’s算盘: When the KOSPI drops 28%, retail investors are still buying—showing local capital has enough resilience. If retail investors’ faith in won-denominated assets isn’t broken, then seize the moment to push the won into a freely convertible currency: build an offshore settlement network, issue won stablecoins, and tokenize government bonds. Faster than Japan.

What South Korea is doing: rebuilding financial infrastructure through tokenization. Whether the KOSPI falls or not doesn’t matter; what matters is who is taking the next financial system standards-setting power.

#韩元 #代币化 #South Korea

One Push, One Soldier: Binance and Bybit saw $2.3 billion in stablecoin outflows over the past 30 days. BTC liquidity is drying up.

Since the beginning of the year, exchange stablecoin reserves have been steadily declining—investors are withdrawing their funds and exiting. BTC has tested the $60,000 level for 165 straight days. It surged to $80,000 in May but failed to hold, and then fell back. It’s not sell pressure slamming the market—it’s that the buyers have disappeared.

As for the ETF side, last week BTC inflows were only $75.67 million—an asset with a market value of $1 trillion had less than $100 million come in over a week. Meanwhile, the ETH ETF saw inflows of $105 million. Are funds choosing ETH over BTC? No—funds didn’t really pick either; ETH just has a smaller base, so the number looks larger.

Moving stablecoins out of exchanges is like withdrawing bullets from the gun chamber. Without ammunition, any rebound is an empty shot.

#BTC #稳定币 #liquidity

Daily Push Forward by One Soldier: US 30-year Treasury yield rises to 5.06%, the highest since 2007.

At the start of 2022, it was 2% for the same maturity. Over three years, the risk-free rate has flipped 2.5x, meaning the discount rate for risk assets has surged in tandem. BTC hovering around $60k for 165 days wasn’t without reason—when you can earn 5%+ with no risk, the allocation threshold for speculative capital is pushed much higher.

Even tougher: big tech companies are competing with the US government for money. AI infrastructure needs to raise funds via bond issuance, and fiscal deficits also need to be financed via bond issuance—both sides together drain liquidity. Long-term interest rates are not controlled solely by the Federal Reserve; they are pushed up from the supply side of corporate funding and Treasuries.

A 5% risk-free yield is BTC’s ceiling—not because it’s bad, but because the certainty next door is too tempting.

#BTC #美债 #Macroeconomics

Step by Step: A Single Soldier at a Time: In July, the portfolio position fell 42% from its peak; all profits from May to June were wiped out. The most heavily weighted memory stocks contributed the core portion of the decline. Gains and losses come from the same source—this is proven thoroughly.

Two decision errors:
1. At the end of June, seeing the top signal, I sold only 15%—I knew it was the top but still held on; I was greedy.
2. During the downtrend, I bottom-picked too quickly, and I even bought leveraged products—this downtrend wore down the position severely.

At heart, it’s inertia thinking. The first few pullbacks this year were profitable when I bought the dip, so I developed the conditioned reflex of “buy when markets fall.” But this time the magnitude and the affected range both exceeded expectations.

The KOSPI has already dropped 28% from its historical high on June 22, and it has been falling for four straight weeks. The Bank of Korea also raised rates, and the FSC effectively cut off the ammunition for leveraged ETFs. The memory fundamentals haven’t changed—HBM supply is still tight, LTA remains in place, and orders haven’t collapsed—but the self-reinforcing liquidation from de-leveraging doesn’t require fundamentals to deteriorate; it only needs the capital structure to break.

The key question isn’t whether “memory is still scarce,” but when the market’s expectations for the growth rate of memory will return to normal. The day the narrative turns is the starting point for repair.

#内存 #韩股 #盈亏同源

Hold a step forward for one soldier: When an industry begins to attract large numbers of “non-industry participants”—social check-in events, concept dissemination, and fresh waves of Tsinghua and Peking University graduates—it means the attention-rent phase has arrived.

The speed at which the attention dividend fades is far faster than that of the technology dividend. The places with the most funding, the highest valuations, and the greatest media exposure are where the most entrepreneurs, investors, and KOLs gather. The movement of capital is guided by the movement of people.

Most people think it’s really low, and industries that Tsinghua/Peking graduates look down on and refuse to go into often still have great potential. When the Tsinghua/Peking effect starts marketing an industry, it’s basically at the end of the road—so intensely “involuted” that ordinary people can’t even make a living.

In 2021, the metaverse was like this; in 2025, so is AI. But don’t rush to flee—this kind of heat can still make money for another three to five years. The key is to set your sights on three to five years later, and find the next “low” place before the attention tide recedes.

#AI #注意力红利 #Investment reflections

Daily pushing forward one step at a time: ETH ARR is $36.5 million; using a market cap of 225 billion, the PE is about 6,000x. At the peak in December 2021, this figure was 28x.

From 28x to 6,000x—it's not that ETH changed; revenue collapsed. In 2021’s DeFi Summer, on-chain fees were a money-printing machine. By 2026, Layer 2 has pushed Gas down into the dust, and income shrinks along with it. The result of technological progress is that token holders become poorer.

If you value ETH using cash flow, you’ll find: ETH is not a stock—it’s a technology commodity. The value of a technology commodity isn’t in its net profit, but in how much it is used as underlying infrastructure by protocols. HBM memory chips right now look like an HBM storage leader at the top of an AI cycle—demand is real, but the valuation has already been priced in.

The cycle is turning the 28x PE story into the reality of 6,000x, then waiting for the next round: “This is Infrastructure.”

#ETH #估值 #cycle

Day by day, one step at a time: Circle CEO Tarbert has sold CRCL a total of 10 times since June 2025, cashing out $30.77 million, and has never added to his position.

Previously, in an interview with Fox, he said, "The company's stock price will naturally take care of itself," emphasizing that "Circle is in it for the long haul."

What "long haul" means: executives sell first. Ten sales with no add-ons—each one is a pure exit. The story in the stablecoin race is indeed getting bigger, but after listing, insiders tell a different story—one that comes from voting with their feet.

Note: this is not an automatic execution hint from an SEC 10b5-1 plan; it’s active, discretionary dealing. Stablecoins are a good business, but a good business doesn’t necessarily mean a good price, and a good company doesn’t necessarily mean a good valuation. When insiders know the cost of the "long haul" better than outsiders, the long-haul narrative is meant for retail investors.

#Circle #CRCL #stablecoin

Day by Day: FSC raises single-stock leveraged ETF margin from 10 million to 30 million won, accepts only cash, limits purchases to 20 shares per order, suspends new IPO listings, and bans advertisements.

This is not regulation—it’s a steel lock placed on retail investors’ 3x leverage.

The KOSPI has fallen for four straight weeks and is down 28% from its historical high on June 22. Citigroup downgraded South Korea from Overweight to Neutral, while upgrading China—citing that South Korea’s “AI winners concentrated rally” will spread to a wider set of sectors.

Retail investors used 10 trillion won to bottom-fish and hold the downside. But the counterparty to retail has changed: foreign capital is running, institutions are cutting exposure, and the ammunition for leveraged ETFs has been directly cut off by regulators. Without the 3x leverage amplifier, even if semiconductor earnings hit new highs, the “retail investors疯狂拉估值” playbook is hard to replicate.

After this epic round of selloff, the money HBM has earned is still real—but the fuel that pulled valuations back up was taken away by regulators.

#KOSPI #杠杆ETF #Semiconductors

Daily Progress, One Soldier at a Time: Over the past three months, Korea’s largest exchange, a leading fintech platform, and a traditional securities firm have all moved into OP Stack.

Three completely different enterprises, one shared infrastructure—each customized as needed.

This isn’t a PR piece about “going on-chain.” It’s real, industry-grade adoption. Korea’s financial infrastructure is being rebuilt with OP Stack—exchange, payments, and securities, all on one chain across three scenarios.

We discussed tokenization’s endpoint yesterday: it’s not Web3 that swallows TradFi—it's TradFi treating blockchain as a better database. Korea is carrying out proof of this.

#Optimism #OPStack #Korea

Day by Day One Push: BTC’s structure is splitting—medium whales holding 100–1,000 BTC are selling, while big whales holding 1,000–10,000 BTC are taking.

Meanwhile, a 40x-leverage mega whale is going long 1,662 BTC, with a position size of $107 million—making it the largest BTC long on Hyperliquid. The entry average price is $63,958, and the liquidation price is $63,143.

The medium whales are unloading, while the leveraged mega whale goes all in. The big whales are “catching” with spot, and the leveraged mega whale is “catching” with its life. The risk level of these two kinds of “buyers” is completely different: spot buyers are betting on belief and patience; a 40x buyer is betting on margin and a countdown.

If the sell orders from the medium whales get exhausted, circulating supply will shrink and pressure above will ease—but only if the liquidation price of $63,143 isn’t touched. It’s less than 3% away from the current price.

#BTC #巨鲸 #Hyperliquid

Daily push, one step at a time: Databricks provides managed services for open-source models like Kimi, with Asia’s GPU capacity nearly exhausted. Demand continues to grow in Japan, South Korea, the United States, and India.

Kimi’s strength isn’t only on the consumer side—GPU demand on the B2B side is also surging. Databricks’ customers want compute power; Databricks needs to buy GPUs. Buying GPUs requires massive fundraising—every link in the chain is saying “not enough.”

In its quarterly report, Alibaba reported cloud growth of 40%+. Meanwhile, Databricks’ Asia capacity is running out. The arms race for AI compute has shifted from “whose model is better” to “who has more GPUs.” This is a war on the supply side; the demand side has never been the problem.

#AI #GPU #Databricks

Daily Strength: Korea’s pension money pours 370 billion won this month to bottom-pick SK Hynix, and the KOSPI dips below 6,500 for the first time in three months.

Institutions net sold 92 billion won, retail investors bought 35 billion won, and foreign investors bought 51 billion won—large institutions are on the move, while retail and foreign capital are the ones catching up. The pensions are not “dumping bombs”; the selling wave is coming from foreign investors with 11 trillion won.

Yesterday we said KOSPI volatility outperformed BTC—today we add one more cut: the “semiconductor peak” narrative has smashed SK Hynix down 33% in a month, but pension funds moved in and picked up 370 billion won. Korean retail investors buying semiconductors—like Chinese retail investors buying AI chips—is the same belief wrapped in a different package.

Volatility in traditional markets is starting to look more and more like a meme: one wild bet, one belief, and a semiconductor leader down 33%.

#SK海力士 #KOSPI #Semiconductor

Day by Day: Install 4 Front-End Plugins for Codex 5.6—finally, the resulting webpages don’t look like “plastic parts from an AI pipeline.”

The problem with AI writing code is this: while the logic may be correct, the aesthetics default to whatever the system ships with. The default is mediocre. Those 4 plugins, in essence, inject “an engineer with taste’s instincts” into the AI workflow—not to make the AI smarter, but to turn the AI’s default output from “it runs” into “it looks good.”

That’s the real bottleneck of AI-assisted development: it’s not capability—it’s taste. Capability can be scaled, but taste can’t. There are AI coders everywhere; there aren’t yet AI designers with taste.

#AI #Codex #Front-End Development

Build one more soldier at a time: a useful mental model for judging an AI company’s valuation—within the complete business loop, how many steps still have to call the most expensive frontier model?

As long as this ratio keeps falling, the expected ARR growth rate for AI giants is being eroded.

Analogy: in an elite core revenue-generating team at a company, how many top-school graduates are required? If the number keeps getting smaller, the premium of a top-school diploma is shrinking.

After Kimi K3’s backend surpasses GPT-5.6, and open-source model parameters bloat— the moat of a frontier model isn’t “stronger,” it’s “cheap enough to be strong.” When a free model like K3 is good enough for 80% of scenarios, $5 per million tokens is no longer a floor price—it becomes a ceiling price.

The key variable in AI investing isn’t who builds the strongest model; it’s how quickly the “good enough” threshold is moving downward.

#AI #投资 #大模型