计然

South Korean stocks rebound 3.3% in a day; intraday gain of 4.7% triggers a buy-side sidecar: still haven’t recovered the previous day’s losses, and remain 37% below the June high

【What happened on October 21】
- The KOSPI closed at 5,735.66, up 3.29%. It peaked at 5,816 intraday, up 4.75%, triggering a buy-side sidecar around 10:00 Beijing time (automated buy orders paused for five minutes). It was the first buy-side trigger in two months. According to EBN, including sell-side triggers, this was the 52nd such event this year.
- Samsung closed up 5.7%, after gaining as much as 7.8% intraday; SK Hynix closed up 5.8%, after rising as much as 8.7%. The KOSDAQ gained just 1.4%—the rally was in semiconductors, not across the whole market.
- Korean media attributed the gains to foreign buying of semiconductor stocks and easing AI concerns. There was no specific new news—and the previous day’s −3.73% drop hadn’t either.
- Korean stock contracts on Binance moved in the same direction, but less sharply (read at 17:00 Beijing time; the day’s session had not closed): Samsung contracts +2.6%, Hynix contracts +2.6%, Korean stock 3x ETF contracts +0.8%; $BTC +0.3%, lagging behind.

【The view over a week】
- October 15, circuit-breaker day: closed at 5,554; October 16: 5,668; October 19: 5,768; October 20: −3.73%, back to 5,553; October 21: +3.29%, up to 5,736.
- Today’s rebound didn’t recover yesterday’s losses: at 5,736, the index is still 0.6% below its October 19 close of 5,768.
- The widely circulated figures—“+4.7%, Samsung +7%, Hynix +7.5%, $160 billion”—were intraday peaks at 10:16 Beijing time. By the close, the gains had been cut roughly in half. The numbers are right; the timing matters.

【The view over a year】
- The KOSPI’s highest close of the year was 9,115 on June 22. Today it stands at 5,736, down 37%.
- October 1: 7,004; down 18% in three weeks. From October 8 to 15, down 16% in one week.
- This selloff has hit the AI and memory supply chain. Samsung and Hynix are the KOSPI’s two largest constituents: if they rise 7%, the index can gain 4%; if they fall 7%, the index can also lose 4%. Since the start of October, the KOSPI’s daily direction has largely matched the direction of these two stocks.

【What to watch】
- Can the KOSPI reclaim 5,768, its October 19 close? If not, today’s move was just a technical rebound after the 3.7% drop.
- Micron’s and SK Hynix’s upcoming earnings and guidance.
- The relationship between Korean stock contracts on Binance and BTC: on October 17, 18, and 20, all three sharp drops began with Korean stock contracts, with crypto following 5 to 25 minutes later. Today it was the other way around: Korean stocks rose 3%, while BTC gained just 0.3%. The correlation seems to work only on the way down.

Down 3.7% one day, up 3.3% the next—the second move of this magnitude in a week. Did SpaceX’s earnings restore confidence in AI stocks, or was it just short covering?

BlackRock’s IBIT saw $665 million in one-day inflows—the biggest in over a year!!!

【What happened that day】
- On Monday, October 19, U.S. spot BTC ETFs saw total net inflows of $527 million, their biggest day since September 22
- Breakdown: “IBIT +$665 million,” Fidelity’s FBTC −$84 million, Grayscale’s GBTC −$55 million, and the other eight funds were flat. Excluding IBIT, the total was −$139 million
- According to The Block, $665 million was IBIT’s largest single-day inflow since October 2025
- $BTC was down 0.22% that day, closing at 81,938 and still range-bound

【The next day】
October 20: IBIT +$30 million, total +$23 million. No follow-through.

【Putting it in monthly context】
- All ETFs combined saw inflows of $2.65 billion in September, including $1 billion in a single day on September 21
- So far in October, the total is −$65 million: IBIT +$885 million, the other eleven funds −$949 million
- In other words, since October, money in ETFs hasn’t been flowing in—it’s been moving house: from other funds to IBIT, with the overall total unchanged. That $665 million day was the biggest day for fund rotation, not for new money coming in
- One related point: on October 20, BlackRock filed for its third Bitcoin ETF, a downside-buffered product. This type of product is designed for investors who aren’t comfortable holding BTC directly

【How to read this number】
- “The biggest single-day inflow in over a year” is true; “the start of a bull market” is another matter. A single-day inflow reflects one buyer’s decision; monthly net inflows are what show the trend—and October is still negative
- Why is money concentrating in IBIT? It has low fees, the best liquidity, and the most active options market, so institutions commonly move funds there when rebalancing. I haven’t verified this explanation from a primary source, so treat it as speculation
- JPMorgan’s ten-year baseline for BTC that same week was 10% annualized returns and 55% volatility. That lines up with the ETF rotation this month: institutions are allocating, not scrambling to buy

【What to watch】
- Can the combined total turn positive over the remaining eight trading days of October?
- Will the funds outside IBIT have a day when they’re positive in aggregate?
- Will BTC break out of the 79,700–87,200 range?

$665 million was the biggest one-day inflow in a year, and the only meaningful one in October. Do you think ETF money is flowing in, or just moving house?

$BTC has been trading sideways for three weeks, with volatility compressed to its narrowest in 9 years: the range is about to break, but which way? Could a U.S.–Iran ceasefire be the catalyst?

【How narrow is the range?】
- 23 days since September 28: closes ranged from 81,491 to 86,530; intraday prices ranged from 79,800 to 87,220
- The 30-day range (high minus low, divided by the low) is 9.5%—among the narrowest 1% of the 3,323 trading days since August 2017; the all-time narrowest was 6.4%
- 14-day realized volatility is 26.4%, at the 13th percentile over the past 2.7 years

【What happened historically after volatility compressed this much?】
Since 2017, there have been six non-overlapping instances when the 30-day range first compressed below 10%. The first five:
- July 23, 2020: +21.4% over the following 30 days
- July 21, 2023: −12.4%
- October 13, 2023: +38.0%
- September 16, 2025: −7.4%
- August 1, 2026: +25.1%
In all five cases, a one-way move of more than 10% unfolded within 30 days: three to the upside and two to the downside. The sixth is the current period, which began on September 20. It is up +2.3% so far and has already lasted longer than the previous compressions.
The sample is only five instances, so all we can say is that “volatility is likely to return”—not which way.

【U.S.–Iran: first, a correction to one assumption】
There is no ceasefire. Three things are happening at once:
- The U.S. rejected Iran’s ceasefire proposal in late September. According to the WSJ at the time, the expectation was that bombing would resume after the midterm elections; there have been no new airstrikes since.
- On October 18, Iran proposed that “if the U.S. accepts its terms, the Strait of Hormuz will reopen within 7 days.” On the 21st, Iran’s foreign minister said the strait would remain closed until the U.S. accepted. Reuters reported on the 20th that negotiations were deadlocked.
- Attacks in the strait have not stopped: two cargo ships were sunk on October 14, and the number of attacks on the 17th reached a wartime high.
Prediction markets put the probability that “the U.S. will not strike Iranian territory again before the November 3 midterm elections” at 80.5%. That prices in a “pause in U.S. strikes,” not a ceasefire agreement, and it says nothing about Iranian attacks on shipping.
So a “ceasefire” is not the catalyst—it does not exist yet. The real potential catalyst is reopening the strait. Prediction markets put the odds at 0.8% by October 31, 5.5% by November 30, and 11.5% by year-end; the probability that daily transits through the strait will be between 0 and 5 ships on October 31 is 90.5%.
The transmission chain is: reopening → Brent falls from 109 → energy puts less upward pressure on CPI → rate-hike expectations decline → bullish for BTC. This chain has only a one-in-ten probability; it is not the base case. The risk in the opposite direction is greater: negotiations collapse completely, bombing resumes, and oil surges again.

【Upcoming dates to watch】
- October 28: FOMC; Kalshi puts the odds of an October rate hike at 14%. Strategy’s special shareholder meeting is also scheduled for that day.
- October 31: October monthly close
- November 3: U.S. midterm elections
- November 11: MSCI announces its list of companies to be excluded for non-operating-company status, including MSTR
- Anthropic publicly files its IPO prospectus. The Wall Street Journal reported on the draft’s contents on October 20; the date is still unknown.

【What to watch for a break】
- Upside: a close above 87,220, accompanied by a simultaneous increase in open interest and spot trading volume
- Downside: a close below 79,732, the October 15 perpetual-futures low
- More fuel on the downside: funding rates have been positive for 15 consecutive periods, and perpetual-futures open interest is elevated at around 95,000 BTC. If the market breaks down, cascading liquidations of longs would provide more fuel than short liquidations.
- The three recent sharp drops (October 18 at 05:00 and 21:00, and October 20 at 09:00, Beijing time) all started with Korean equities and memory contracts on Binance falling first; crypto followed 5 to 25 minutes later. The Asian AI chain may determine the direction sooner than U.S.–Iran developments.

BTC has been trading sideways for three weeks, with volatility compressed to its narrowest in nine years. Historically, periods like this have almost always been followed by a move of more than 10% within 30 days—but breakouts have happened in both directions. A ceasefire is not the catalyst; it does not exist yet. Reopening the strait is the potential catalyst, and the market gives it only a one-in-ten chance. Which way are you betting?

Is this cycle’s leading coin $ZEC finding a floor and stabilizing, and gearing up to rally again?

In a nutshell: There are signs of stabilization, but an “upswing” is not yet confirmed. After hitting 1,049 on October 18, it posted two consecutive up days and closed at 1,257 on October 20 (+9.1%; ZEC was the only one to rise that day), reclaiming the 1,200 level. However, it has yet to break above the 1,300–1,350 upper end of the trading range in place since October 8. The pattern of progressively lower lows has paused, but has not reversed.

【Price (Binance spot, daily)】
- The cycle high close was 1,653 on September 26, with an intraday high of 1,698
- On October 8, it fell 10.6% in a single day (1,328 → 1,187, with an intraday low of 1,113), then entered a wide trading range of 1,049–1,349
- Lows moved progressively lower: 1,158 on October 12 → 1,125 on the 14th → 1,075 on the 15th → “1,049 on the 18th, the cycle low”
- Rebound: closed at 1,152 on the 19th (+1.9%) → “closed at 1,257 on the 20th (+9.1%)” → around 1,261 at approximately 12:00 Beijing time on the 21st
- Still down 24% from the September 26 high close
- Levels to watch: October 11 high of 1,300 and October 12 high of 1,349; the October 20 intraday high of 1,280 remains below both. A break above 1,300–1,350 would signal a breakout from the trading range

【Relative to BTC】
ZEC/BTC: 0.01377 on October 18 was the cycle low; 0.01514 on the 20th and 0.0152 on the 21st. The September 26 level of 0.01958 was the highest since September 2019; it is now 22% below that level. It is also recovering relative to $BTC , but has likewise not returned to its highs.

【What drove the October 20 green candle? (Binance perpetuals)】
- Retail traders added shorts as the price rose: the retail long/short ratio (by account count) fell from 1.02 to 0.61 over three days, and was still 0.62 in the morning of the 21st
- Shorts were liquidated: on the 20th, $4.79 million in ZEC short positions and $1.4 million in long positions were liquidated. It was the second-biggest day for short liquidations since October 9; every day from October 13 to 19 saw more longs liquidated than shorts
- New positions were opened: open interest rose from 455,000 to 471,000 coins, up 16,000 in a day; market buys dominated, with a taker buy/sell ratio of 1.07
- Whales reduced their longs: the large-trader long/short ratio went from 1.75 → 1.60 → 1.35, but remains above 1
- Funding turned slightly negative at the 08:00 settlement on the 21st: −0.0017%
- Sector comparison: XMR was down 1.3% and DASH was up 0.3% that day. This was not a rotation into privacy coins; it was ZEC’s own move. The specific catalyst has not been verified, so no attribution is being made

【How to read this】
- The fuel for September’s rally was “retail traders heavily short, with shorts repeatedly squeezed.” On October 8, the retail long/short ratio moved back above 1, and that fuel was temporarily spent; what followed was two weeks of a gradual decline and choppy trading
- Retail traders have now turned bearish again (0.61), so fuel is building back up. That is a structural reason it “could rally again”
- Two caveats: ① The retail long/short ratio for a single coin cannot be used to time the market; it describes positioning, not future direction. ② Large traders have been reducing longs during the same period, the top of the range has not been broken, and the pattern of lower lows has not reversed

ZEC found support at 1,049 and, with a +9% green candle, climbed back above 1,200. Retail traders have started shorting again, making the positioning look a lot like it did before the September rally began. But it has yet to break out of the 1,050–1,350 range: only a move above 1,350 would count as an upswing; a drop back below 1,200 would mean it is still range-bound. Do you think retail traders’ shorts this time will be fuel—or a sign they saw it coming?

“The global debt crisis has never been this severe”? Debt ratios haven’t hit new highs—the cost of interest has

【First, a correction to a widely shared chart】
On October 20, a post with 80,000 views claimed that “the global debt crisis has never been this severe,” and included debt ratios for nine countries. Each of the nine figures matches the IMF’s forecast for 2026 in its April edition this year. They are not “current debt ratios,” and the IMF has already released a new edition on October 14. The new 2026 figures are: Japan 199.4% (204.4 in the previous edition), China 104.4% (106.9), Canada 115.6% (110.7, the biggest upward revision), the United States 126.8%, France 118.5%, Italy 137.9%, the United Kingdom 102.0%, and Germany 64.6%. Using an outdated edition and presenting forecasts as current figures isn’t fabrication, but it gets the ranking wrong: Canada was undercounted by the most.

【“Never been this severe”—let’s look at the two halves separately】
- By debt ratio, the claim doesn’t hold. Global public debt peaked at 97.5% in 2020, compared with 94.6% in 2026; advanced economies were at 119.3% in 2020 and 107.8% in 2026.
- By absolute amount, it does. According to the IIF, total global debt (government, household, corporate, and financial) reached a record $365.5 trillion in the first half of 2026.
- The gap between the two is inflation from 2021 to 2024: it pushed up nominal GDP and brought the ratio down. The IIF itself says, “Inflation helped contain debt ratios, masking vulnerabilities.”

【What’s really gotten worse is interest rates】
10-year government bond yields in the G7, December 2021 versus September 2026 (monthly averages):
- France: 0.05% → 4.48%, up 4.4 percentage points
- United Kingdom: 0.84% → 5.24%, up 4.4 percentage points, the highest in the G7
- Germany: −0.38% → 3.47%, up 3.9 percentage points
- United States: 1.47% → 4.99%, up 3.5 percentage points
- Italy: 1.05% → 3.99%, up 2.9 percentage points
- Japan: 0.07% → 2.94% (October 19), up 2.9 percentage points
- Canada: 1.45% → 3.88%, up 2.4 percentage points
In a nutshell: in 2020, there was more debt, but it cost almost nothing. In 2026, there’s slightly less debt, but every dollar costs 2.4 to 4.4 percentage points more. It will take years to replace old low-interest debt, so the increase in interest burdens is far from over. U.S. federal interest outlays were running at an annualized rate of about $1.28 trillion in the second quarter of 2026, around 3.9% of GDP.

【A country-by-country look at who’s most at risk (estimates)】
Using “long-term interest rate minus nominal growth” as a rough measure:
- “France” is most at risk: the interest rate minus growth is about +1.7, the highest of the eight countries; the deficit is 5.4%; and it has no central bank of its own. The 10-year yield of 4.48% is already higher than Italy’s 3.99%—markets have reversed the risk ranking of the two countries.
- The “United States” is accumulating debt the fastest: the IMF forecasts an 18-percentage-point rise from 2026 to 2031, to 145%; its 8.5% deficit is the highest among advanced economies. The interest rate minus growth is still negative, but the deficit is so large that growth alone can’t contain it.
- “Japan” is having its debt diluted by inflation: its debt ratio has fallen 29 percentage points in six years, but its 30-year yield has reached 4.03% and its 40-year yield 4.04%. The interest rate minus growth has just turned positive, and the old strategy of “carrying high debt with low interest rates” is on its way out.
- “China” is moving in the opposite direction: its debt ratio has risen 34 percentage points in six years, the fastest among the eight countries. But its 10-year yield is only 1.69%. The problem isn’t high rates; it’s rates so low that they reflect a scarcity of investment opportunities.

【What the IMF itself says】
According to the IMF’s October 14 Fiscal Monitor, as recounted in an interview by Xinhua: global public debt will exceed 100% of GDP before 2030; the real interest rate on government debt is higher than it was before the pandemic; and when debt is already high, rate hikes will worsen deficits and debt by a greater margin. Meanwhile, in 2026, the Federal Reserve, the European Central Bank, and the Bank of Japan are all raising rates.

【A crisis won’t unfold everywhere in the same way (assessment)】
Countries with their own central banks can turn to inflation and currency depreciation, rather than default. France and Italy, which don’t have their own national currencies or central banks, are the ones more likely to face a traditional spread crisis. So a “global debt crisis” looks more like several different illnesses, with the first outbreak most likely to be in the eurozone.

Debt ratios are falling, interest rates are rising: which do you think will hit trouble first—the French bond spread or the U.S. dollar? If you hold $BTC , which side are you on?

SpaceX earnings: $14.6 billion in revenue in one quarter, AI revenue tops Starlink for the first time, first profitable quarter since going public—but it spent $19.6 billion in the same quarter

【First, the players】
SpaceX, Nasdaq ticker $SPCX , listed on June 12, with a market cap of about $2 trillion. Its reporting is split into three segments: Launch (Space), Starlink (Connectivity), and Superintelligence (SI)—formerly called the AI division, housing Grok, Cursor (acquired in August at a $60 billion valuation), and cloud services that rent out its own computing capacity to third parties.

【Q3 figures (8-K attachment, unaudited)】
- Revenue: $14.575 billion, up 161% year over year (from $5.58 billion in the same quarter last year) and up 87% quarter over quarter
- By segment: Launch $1.77 billion (+6%); Starlink $5.25 billion (+69%); SI $7.56 billion (+830%, +195% quarter over quarter). SI topped Starlink for the first time. Of that, $7.02 billion came from “solutions and infrastructure”; the company said this was mainly from newly activated cloud service agreements, the consolidation of Cursor (which contributed $608 million), and Grok usage
- Profit: Operating profit of $2.79 billion (versus a loss of $270 million in the same quarter last year); net profit of $4.40 billion (versus a loss of $1.05 billion last year). Its first profitable quarter since going public. The $1.6 billion by which net profit exceeded operating profit came from $870 million in interest income, $1.03 billion in other gains (including $560 million from related parties), and a $390 million income tax benefit—not from its core business
- SI segment operating profit was $430 million, turning positive for the first time, even after accounting for $1.2 billion in stock-based compensation from the Cursor acquisition
- Starlink consumer subscribers: 13.6 million, up 1.6 million in the quarter; $69 per household per month
- Rated computing capacity: 1.7 GW, versus 0.6 GW in the same quarter last year

【How fast the money is going】
- Q3 capital expenditures: $19.59 billion, $5 billion more than revenue; the SI segment accounted for $16.44 billion, spent on computing centers such as Colossus II
- Capital expenditures for the first three quarters: $48.07 billion; operating cash flow: $13.09 billion; a $35 billion shortfall, covered by the $85.68 billion in net IPO proceeds
- Cash at the end of September: $51.29 billion, plus $38.56 billion in marketable securities, for a total of about $89.8 billion; debt was about $38.8 billion
- CFO’s Q4 capital expenditure guidance on the earnings call: $35 billion to $40 billion. At this rate, the $89.8 billion on the balance sheet will last a little over two quarters
- Compare that with the October 18 piece on CoreWeave: they borrow to build computing capacity, with interest consuming one-fifth of revenue; SpaceX is using IPO proceeds, so for now it doesn’t have to answer to creditors. But the pace of spending is the same

【Guidance and the market’s price tag】
- CFO: Q1 2027 revenue of $33 billion to $37 billion, 2.3 to 2.5 times this quarter’s revenue
- Musk: annualized recurring revenue of between $500 billion and $1 trillion by the end of 2027
- A $2 trillion market cap is 34 times revenue based on an annualized Q3 figure of $58.3 billion; based on annualized Q1 guidance of $132 billion to $148 billion, it’s 14 to 15 times. The market price already treats the guidance as fact
- Merger: Musk said Tesla and SpaceX have each formed independent special committees of the board to study a merger. He also said it’s “not a done deal”

【Binance perpetuals】
SPCXUSDT rose from 148 to 159.9 (+7.6%) after the earnings report. Trading volume on October 20 was $1.76 billion, more than double the previous day; TSLAUSDT was up 3.9%. At the same time, $BTC was around 83,000, with no follow-through.

AI revenue doubled in a quarter; capital expenditures hit $19.6 billion in a quarter: when you buy SpaceX, are you buying rockets and Starlink—or Musk’s computing bill?

Oct. 20 market overview: South Korean stocks fell another 3.7% overnight, pushing BTC down to 81,200; shorts were massively liquidated during the U.S. session, and BTC closed at 83,028, ending a six-day streak of closes below the bottom of the range

【Overall market】
- 「$BTC 83,028」+1.33%; high 83,439 (the 22:00 hour Beijing time), low 81,200 (the 09:00 hour Beijing time). The close was 465 above the range floor of 82,563, ending the six consecutive closes below it that began on Oct. 14; so far in October, there have been 8 closes below it
- Price action: South Korean stocks opened lower at 09:00 Beijing time. Binance’s 3x South Korean stock ETF contract and SK Hynix contract fell first at 08:40; ETH and BTC followed after 09:00, with a low of 81,200 at 09:14. Long liquidations totaled $25 million during that hour. Around 22:00, Binance’s MSTR and COIN contracts moved first, and BTC took off at 22:01, surging from 82,149 to 83,439 during that hour; short liquidations totaled $63.5 million ($26.1 million in BTC). At 23:00, BTC fell back to 82,211, with $23.3 million in long liquidations; afterward it traded sideways between 82,400 and 83,100. No news was found to explain the 22:00 move, and Strategy had no new 8-K filing that day, so no cause is attributed
- Liquidations for the day: $92.9 million shorts / $76.1 million longs. Second day since Oct. 19 with more short liquidations; BTC shorts $29.9 million / longs $18.6 million
- Perpetual open interest: 94,819 coins, down 217 in a day. Price up, open interest down; this move was mainly driven by short covering. Funding rates have been positive for 15 consecutive periods; the settlement at 08:00 on the 21st was +0.0065%
- Fear & Greed Index: 63, previous reading 55

【Equities】
- U.S. stocks: S&P 500 +0.28%, Nasdaq −0.02%, Dow +1.02%; Philadelphia Semiconductor Index −0.29%; VIX 17.31, previously 18.44. Micron −1.0%, Nvidia −0.3%, CoreWeave +1.7%, COIN +1.4%, IBIT +0.9%; 「MSTR +4.9%」. Strategy’s Oct. 19 8-K: acquired 267 BTC from Oct. 12–18 (average price 82,900; total holdings 848,517); increased its preferred stock repurchase authorization from $2 billion to $3 billion
- After hours: SpaceX’s third-quarter revenue was $14.6 billion, up 161% year over year; AI revenue exceeded Starlink revenue for the first time, and the company was profitable for the first time since listing. Shares rose from 147.79 to around 159.7 after hours. According to the WSJ and Reuters, Tesla and SpaceX have each formed committees to discuss a merger
- Asian stocks: 「KOSPI −3.73%」 to 5,552.74, back to the level seen on Oct. 15, the circuit-breaker day; Samsung −3.0%, SK Hynix −4.2%; Nikkei −1.30%, Taiwan stocks −1.13%, Hang Seng −1.32% (compared with Oct. 16; markets were closed on the 19th), Shanghai Composite +0.12%; MiniMax −5.8% (171.5, 4% above its IPO price of 165), Zhipu −6.4%
- European stocks: Euro Stoxx 50 −0.38%, DAX −0.06%

【U.S. Treasuries and macro】
- U.S. Treasuries (Treasury Department figures, compared with Oct. 19): 2-year 4.79% (+2bp), 5-year 4.94% (unchanged), 10-year 5.12% (−3bp), 30-year 5.42% (−6bp). Yields at the long end fell, flattening the curve
- Kalshi: 14% chance of an October rate hike (13% for 25bp + 1% for a larger hike); gold futures 4,132.9 (−0.65%); Brent 109.33; U.S. Dollar Index 101.79

【Altcoins】
- Among 69 USDT pairs with trading volume above $5 million: median +2.6%, 61 up and 8 down, with 20 gaining more than 5%
- 「$ETH 2,449.2」+1.7%, SOL +1.4%, XRP +2.4%, BNB +1.3%, DOGE +2.7%, HYPE +1.5%, 「ZEC +9.1%」 (1,257.2), ENA +3.5%, SUI +3.1%, LINK +2.3%, UNI +1.8%, AVAX +1.5%, STRK −0.8% (0.0572, down for eight consecutive days)
- Top gainers: IOST +41.3%, COMP +19.7%, ERA +15.3% (was among the biggest losers the previous day, −6.6%), WLD +10.2%, PEPE +9.6%. Top losers: 「AVA −18.8%」, PHA −6.6%, 「GTC −6.3%」. AVA and GTC gave back their sharp gains from the previous two days. No causes were verified, so none are attributed

【Seven bull-market confirmation signals】2/7, not triggered (readings through Oct. 19)

【What to watch】Can BTC hold above 82,563? Can South Korean stocks stabilize? Meta and Tesla earnings on Oct. 21; Anthropic’s public S-1 filing (the Wall Street Journal reported on the draft contents on Oct. 20); the Oct. 28 FOMC meeting and Strategy’s special shareholder meeting

Breaking: Anthropic lost $80.7 billion in six months? And it has a $600 billion compute bill!

【First, let’s break down the losses】
According to a confidential draft prospectus reported by The Wall Street Journal on October 20:
- Revenue in the first half of 2026 was $16.3 billion, up from $1.1 billion in the same period last year—about 15 times higher. Net loss was $80.7 billion, compared with a $13 billion loss in the same period last year.
- About $72 billion of that $80.7 billion was an accounting item: financing instruments Anthropic had previously issued, including convertible notes, have to be revalued based on the company’s valuation. Its valuation rose from $380 billion in February to $965 billion in May, pushing up the value of the notes. The difference is recorded as a “loss” on the financial statements, but no cash changed hands.
- Excluding that item, the operating loss for the half-year was about $8.7 billion. Compute and infrastructure costs over the same period were about $9 billion—roughly the same amount.
- The accounting treatment is consistent with last year’s: of the $42 billion net loss for all of 2025, $34 billion came from the same type of accounting item. The more sharply the valuation rises, the larger the paper loss—a pattern that will recur.

【The numbers that really matter】
- Cloud-computing agreements total about $450 billion over 7 to 10 years, with most of the commitments non-cancellable. Annualizing the third-quarter revenue of $16.8 billion reported by the FT, that’s equivalent to 6.7 years of revenue.
- Planned compute spending exceeds $600 billion, rising to $750 billion if ongoing negotiations are included. A previous draft reported by Reuters in September put the figure at $518 billion—nearly $100 billion was added in a month.
- We discussed CoreWeave on October 18: AI’s money is borrowed. Here’s the other half of the story: model companies are backing their upstream suppliers by signing long-term contracts they can’t get out of. From neocloud debt to long positions in $BTC , the entire chain is betting on the same assumption: demand will keep doubling. If that assumption holds, $450 billion is scale; if it doesn’t, it’s a fixed bill.

【A draft isn’t the same as a filing】
- This is a confidentially submitted draft. As of October 21, there were still zero public Anthropic S-1 filings on EDGAR, and the company declined to comment.
- Only the figures in the public filing will be official. When it comes out, check the revenue, how the $72 billion is reported, and the level of customer concentration.

【How did Binance perpetuals react?】
ANTHROPICUSDT ($ANTHROPIC ) hit a low of 1,860 after the report came out (around 04:00 Beijing time on October 21), then recovered to 1,899 by 07:00. The October 20 candle was down 2.1%, with $8.5 million in volume—about normal. At that price, the implied valuation is still $1.9 trillion. Did the market simply shrug off the $80.7 billion loss?

On one side, an $8.7 billion operating loss in six months; on the other, a non-cancellable $450 billion bill. Which one is more likely to keep investors up at night?

South Korean stocks rebound for two days, then fall back to circuit-breaker-day levels: this time, it wasn’t foreign investors selling, but domestic institutions

On October 20, the KOSPI closed at 5,552.74, down 3.73%, after hitting an intraday low of 5,493 (−4.8%). On October 15, the day trading was halted by a circuit breaker, it closed at 5,554.22. It rebounded for two days—up 2.05% on the 16th and 1.76% on the 19th—then gave it all back in one day.

【Storage stocks took the biggest hit again】
- SK Hynix fell 4.23% to close at 1,223,000 won, reportedly its lowest closing price since April; Samsung Electronics fell 2.98%; SK Square fell 5.0%
- 771 stocks fell and 117 rose; trading volume was 16 trillion won
- The pattern: the market opened down just 0.6%, touched 5,493 around 09:15 Beijing time, rebounded to 5,606 at 09:30, then fell again in the afternoon to close at 5,553

【Who was selling this time?】
According to exchange data cited by The Korea Times, institutions were net sellers of about 1 trillion won, foreign investors net sellers of about 535.4 billion won, and retail investors net buyers of about 900 billion won.
On October 13, when foreign investors were net sellers of more than 4.5 trillion won, and on the 15th, the circuit-breaker day, when they sold about 3.6 trillion won, foreign investors were the main sellers. This time, domestic institutions took their place, while retail investors bought. Why institutions were selling—pension-fund rebalancing, algorithmic trading, or redemptions—was not mentioned in the report, and I couldn’t verify it.

【Not one positive catalyst could hold the market up】
- U.S. stocks the previous night: Micron +3.0%, Philadelphia Semiconductor Index +1.6%, Nasdaq +1.1%
- That day: Nomura reportedly maintained its Buy ratings on Samsung and Hynix, forecasting that the memory shortage would last until 2028; the won closed at 1,328, its strongest level in more than two years, which is positive for foreign capital
- None of these three positives supported the market. Media reports offered two explanations, both attributed to reports: first, an analyst at Daishin Securities said that “credit risks and earnings concerns surrounding AI investment have not been fully dispelled” (context: CoreWeave’s 2031 and 2032 bonds reportedly fell below 80 cents; even as U.S. stocks rebounded on the 19th, it was still down 0.4%); second, unrest broke out in several parts of France over the weekend, Paris imposed a curfew, and the prime minister said a state of emergency could be declared if necessary, while European stocks fell for a second straight day. The link between France and South Korean stocks is only what the media have suggested
- Takeaway: selling pressure came from valuations and liquidity, not from any one piece of bad news

【Across markets】
- Same day: Nikkei −1.30%, Taiwan stocks −1.13%, Hang Seng −1.32%, Shanghai Composite +0.12%. South Korea, with the largest weighting in storage, fell the most
- Hong Kong stocks resumed trading, and two large-model companies fell alongside storage stocks: MiniMax −5.8%, closing at 171.5, just 4% above its IPO price of 165; Zhipu −6.4%, closing at 484.8, below its October 15 low
- In crypto: at 08:05 Beijing time, Binance’s 3x leveraged Korean-stock contract fell first; at 09:15 it hit a low of −10.1%, while the Hynix contract was down 6.3%. ETH followed at 08:47 and $BTC 09:08, while BTC hit a low of 81,200 at 09:14. About $25 million in long positions across the market were liquidated during that hour

【Where things stand】
Down 39.1% from the June 22 high; down 20.7% since October 2; still up 31.8% since the end of 2025.

Two days of rebound, then one day to give it all back. This time, domestic institutions were selling and retail investors were buying; gains in Micron, Nomura’s Buy call, and a stronger won couldn’t prop up the market. The issue isn’t the news—it’s valuations and liquidity.

Do you think retail investors buying these 900 billion won are picking the bottom or catching a falling knife?

What is the VIX? The “fear index” for U.S. stocks rose 35% in a week—what exactly does it measure?

On October 16, the day AI-linked U.S. stocks plunged, the VIX jumped from 16.46 to 19.99. A week earlier, it was still at 14.84. That’s a 35% rise, and plenty of people are asking what it is and whether they should be worried.

【What does it measure?】
- The VIX is an index compiled by the Chicago Board Options Exchange, based on the prices of S&P 500 options. The more expensive the options, the more buyers are willing to pay for insurance, which means the market expects greater volatility over the next 30 days. The VIX expresses that expectation as an annualized percentage.
- How to read it: A VIX of 18.44 means the options market expects the S&P 500 to fluctuate by about ±18% over the next year, or about ±5.3% over a month and ±1.2% in a day. It indicates the size of the move, not its direction.
- The scale: Below 15 is considered calm; around 20 is near the long-term average; above 30 signals panic, generally corresponding to a 2%–3% one-day drop in the S&P 500. The highest-ever closing value was 82.69, on March 16, 2020.

【Its relationship with the index: they move in opposite directions】
- Over the past two years, when the S&P 500 fell, the VIX spiked; when the S&P 500 hit new highs, the VIX hovered around 15. On April 8, 2025, the day of the tariff scare, the VIX closed at 52.33 and the S&P 500 closed at 4,983—the lowest point of the past two years. The S&P 500 then climbed all the way to 7,700, while the VIX stayed between 14 and 20 most of the time.
- In the past two years, the VIX closed above 30 on only 14 of 502 trading days: during those two weeks in April 2025, and for two days at the end of March 2026. Panic is short-lived; calm is the norm.
- That’s why a VIX spike is often seen as a “bottom signal”: panic peaks, and everyone who was going to sell has already sold. But when it hit 52, it peaked before the market bottomed; when it hit 30, that didn’t happen. There’s no set formula.

【Where are we now?】
- On October 19, it was at 18.44, having just pulled back from 20—within the “a little tense, but not panicking” range. Its median over the past two years is 17.1.
- For comparison, $BTC : the equivalent index for BTC is Deribit’s DVOL, which stood at 37.95 on October 20. That means a normal daily move in crypto is about ±2.4%, twice that of U.S. stocks. A VIX of 38 is a big deal for U.S. stocks; for BTC, 38 is just another day.

【How do you trade it?】
- You can’t buy the VIX itself. You can buy VIX futures and options, as well as ETFs linked to futures, such as UVXY, which offers 1.5x long exposure. Binance also has a tokenized stock version of UVXY, called UVXYB, which launched on October 14.
- The crucial catch: VIX futures are usually more expensive further out, so long ETFs pay a cost every day as they roll over their positions. Hold them long term and much of your principal can get eaten away. They’re only suitable for short-term trades when you expect something to happen in the next few days—not for buying and holding.
- Products that short the VIX work the other way around: they make a little money every day in normal conditions, then can lose it all in a single day when panic hits. That’s how XIV went to zero overnight in February 2018.

【In a nutshell】
The VIX doesn’t predict direction; it puts a price on “how scared the market is.” 18 means a little scared, 30 means really scared, and 50 means terrified.

When was the last time you saw the VIX go above 30?

October 19 Market Overview: U.S. AI stocks rebound, KOSPI extends gains, but BTC spikes to 83,158 then retreats, closing below the bottom of its range for the sixth straight day

【Markets】
- “BTC 81,938” −0.22%; high 83,158 (the hour beginning 15:00 Beijing time), low 81,749 (the hour beginning 00:00 on October 20); closed 625 below the range bottom at 82,563. It has closed below that level for six straight days since October 14, the eighth time this month; traded above it for most hours from 10:00–20:00, but failed to hold
- Price action: Short covering from Sunday night continued into the early hours (short liquidations totaled $14.9 million in the hour beginning 09:00); peaked at 83,158 at 15:00. At 21:00, fell from 82,600 to 82,126; after the U.S. market opened at 21:30, slid from 22:00–00:00 to 81,749 (both longs and shorts were liquidated in the hour beginning 22:00: $7.9 million long / $5.2 million short), then moved sideways between 81,800–82,200
- Divergence from U.S. stocks: Nasdaq +1.11%, Philadelphia Semiconductor Index +1.61%, while BTC fell about 1.5% from its high; COIN −2.7%, MSTR −0.6%. Rising Treasury yields may have played a role, but no specific news was found, so no attribution is made
- Liquidations for the day: $54.7 million short / $30.0 million long, with more short liquidations; BTC shorts $9.8 million / longs $4.5 million
- Perpetual open interest: 95,036 coins, up 1,905 in a day; price flat as volume rose. Funding rates were positive for 12 consecutive periods; the settlement at 08:00 on the 20th was +0.0092%
- Fear & Greed Index: 55, down from 59

【Equities】
- U.S. stocks: S&P 500 +0.41%, Nasdaq +1.11%, Dow −0.07%; Philadelphia Semiconductor Index +1.61%, ending a four-day losing streak; VIX 18.44, versus 19.99 previously. Micron +3.0%, Nvidia +1.4%, Oracle +1.1%; CoreWeave −0.4%, failing to join the rally
- Asian stocks: KOSPI +1.76%, up for a second straight day; the catch-up decline feared over the weekend did not materialize. Shanghai Composite +0.39%, Taiwan stocks −0.61%, Nikkei −0.62%; Samsung +1.2%, SK Hynix +0.4%
- European stocks: EURO STOXX 50 −0.96%, DAX −1.18%

【U.S. Treasuries & Macro】
- U.S. Treasuries (Treasury Department figures, compared with October 16): 2-year 4.77% (+3bp), 5-year 4.94% (+5bp), 10-year 5.15% (+6bp), 30-year 5.48% (+5bp). Longer-dated yields rose more, steepening the curve
- Kalshi: 13% chance of an October rate hike; gold futures 4,151 (−0.7%); U.S. Dollar Index 101.98. Crude oil futures rolled over, so no percentage change is reported

【Altcoins】
- Among 70 USDT pairs with more than $5 million in trading volume: median +1.0%, 46 gainers and 24 decliners
- ETH 2,408.0 (−0.1%), SOL +0.9%, XRP −0.2%, BNB +1.2%, DOGE +0.4%, HYPE +0.4%, UNI +1.7%, “ZEC +1.9%,” LINK +0.2%, QNT +1.3%, AVAX −1.7%, NEAR −2.2%, “STRK −4.6%” (0.0577, down for seven straight days)
- Top gainers: “$AVA +32.4%,” “$GTC +29.9%,” WLD +9.1%; top decliners: ERA −6.6%, STRK, TIA −3.2%
- GTC: Upbit added KRW and USDT markets on October 12. On the 19th, GTC trading volume on Upbit was about KRW 74.1 billion (about $55 million), 2.8 times Binance’s. Since October 17, the founder has posted daily about a “restart,” while also saying they were “concerned that hype was getting ahead of fundamentals.” No official announcement has been made. AVA: No official announcements about buybacks, burns, or partnerships; X has only posts soliciting people to join groups. The reason for the move could not be verified. Both tokens carry Binance monitoring tags

【Seven Bull-Market Confirmation Signals】2/7, not triggered (readings through October 16)

【What to Watch】Can BTC end its six-day streak of closes below the range and hold 79,732? Can the AI-stock rebound continue? Meta and Tesla earnings on October 21; Anthropic is reportedly planning to make its prospectus public around October 26; FOMC meeting and Strategy special shareholder meeting on October 28

Oct 18 Market Overview: BTC rallied to 82,295 but failed to clear the range floor, closing below it for a fifth consecutive day; the weekly golden cross remains intact, but the histogram has contracted for two weeks

【Market】
- “BTC 82,116” +0.77%, high 82,295 (the 17:00 hour Beijing time), low 80,684 (the 04:00 hour on Oct 19); the close was 447 below the 82,563 range floor, marking a fifth consecutive daily close below it since Oct 14; perpetuals bottomed at 80,622, holding above the Oct 15 low of 79,732
- Three phases through the day: at 17:00, BTC rallied to 82,295 but failed to clear the range floor; two selloffs followed, at 21:00 and 03:00 on Oct 19. In the 21:00 move, Korean stocks and memory contracts in Binance’s stock futures fell first, with crypto following 10–25 minutes later; in the 03:00 move, crypto fell first (SOL at 03:25 and ETH at 03:30 broke lower first). Long liquidations totaled $45 million in that hour; from 06:00–08:00, BTC rebounded from 81,222 to 82,116, liquidating $19.6 million in shorts. No specific news catalyst could be verified for either move
- Liquidations for the full day: $95.5 million longs / $47.5 million shorts; BTC: $21.6 million longs / $10.6 million shorts
- Perpetual open interest stood at 93,131 BTC, slightly lower; funding rates remained positive for ten consecutive intervals, with the Oct 19 08:00 settlement at +0.0050%
- Fear & Greed Index: 59, Greed, up from 53

【Weekly chart (week of Oct 12, now closed)】
- Opened at 83,773 → closed at 82,116, −2.0%; high 84,352, low 79,800
- MACD: The golden cross from the week of Jul 20 remains intact, and the DEA has just risen above the zero line; however, the histogram has contracted for a second consecutive week: 2,916 → 2,680 → 2,290
- Moving averages: The close was below the 5-week MA (83,614) and above the 10-week MA (79,357); the 5/10-week golden cross remains intact. The 20-week MA at 71,578 remains below the 50-week MA at 76,584; the death cross from January has not reversed

【Altcoins】
- Among 62 USDT pairs with trading volume above $5 million (small weekend sample): median +0.4%, 41 up and 20 down
- ETH 2,410.3 (+0.3%), SOL +0.6%, XRP +1.0%, BNB +0.7%, DOGE +0.1%, HYPE +1.1%, UNI +0.7%, ZEC +0.3%, LINK +0.5%, SUI +2.1%, “$NEAR +3.1%”, ENA +0.1%, AVAX +0.1%, QNT −1.8%, “$STRK −6.3%” (0.0605, down for a fifth consecutive day)
- Top gainers: GTC +11.3%, ERA +9.6%; top losers: AVA −15.5%, STRK and PYTH −6.3%. No first-hand cause could be verified, so no attribution is made
- Binance stock perpetuals broadly rebounded on Sunday: 3x long Korean stocks +3.4%, Micron +2.1%, Nasdaq 100 ETF +1.1%, Samsung +0.9%, SK Hynix +0.8%; MiniMax −0.5%, Zhipu −0.4%. Weekend futures liquidity is thin; treat these figures only as a reference ahead of Monday’s open

【Macro】No new data over the weekend; Kalshi puts the probability of an October rate hike at 14%. No CPI, PPI, or jobs-report-level data this week, and the Fed is in its blackout period; Meta and Tesla report earnings on Oct 21, followed by the Oct 28 FOMC meeting and Strategy’s special shareholder meeting

【Seven bull-market confirmations】2/7, not triggered (readings through Oct 16)

【What to watch】Asian equities at Monday’s open, especially the KOSPI and memory stocks; whether BTC can reclaim 82,563 and hold 79,732; whether the weekly MACD histogram can stop contracting

MiniMax is down 85%, Zhipu down 78%: Has the bubble in Chinese AI model stocks been squeezed out? A look at Zhipu through the lenses of MiniMax and Anthropic

The short answer: By the valuation measures below, much of the froth has already come out of MiniMax. Whether Zhipu’s valuation can hold depends on how long the revenue growth disclosed by the company can continue.

【First, a look at the two stocks】
- MiniMax (HKEX 00100; Binance perpetual contract $MINIMAX ): IPO price of HK$165; its highest closing price was HK$1,238 on March 18. It closed at HK$182.1 on October 16, down 85.3% from its peak but still about 10% above its IPO price.
- Zhipu Z.AI (HKEX 02513; $ZHIPU ): IPO price of HK$116.2; its highest closing price was HK$2,410 on June 22. It closed at HK$518 on October 16, down 78.5% from its peak but still 4.5 times its IPO price.
- In the week from October 9 to 16, the two stocks fell 14.5% and 19.3%, respectively. From October 17 to 18, both perpetual contracts fell another roughly 3%, potentially putting the shares under pressure at the open on October 19.
- Since July 20, the two stocks’ betas to the Hang Seng Tech ETF have been 2.7 and 1.9, and their correlation with each other has been 0.67. Their correlation with CoreWeave on the previous day was just 0.07. This looks more like trading in “Hong Kong-listed AI stocks” than following US AI stocks.

【First, how many times annualized revenue are the valuations?】
- Anthropic: Implied valuation from Binance perpetuals of about $1.88 trillion, versus $965 billion in its May funding round; annualized revenue of about $67 billion (third-quarter revenue of $16.8 billion multiplied by four; unaudited). That works out to about 28x and 14x, respectively.
- OpenAI: Implied valuation from perpetuals of about $1.49 trillion, versus $852 billion in its funding round; annualized revenue of about $50 billion (according to the FT). That works out to about 30x and 17x.
- Zhipu: Market cap of about $32 billion; company-disclosed annualized revenue of $1.6 billion; about 20x.
- MiniMax: Market cap of about $8.2 billion; company-disclosed annualized revenue of more than $800 million; about 10x.
- But the annualized revenue figures for the two Hong Kong-listed companies use figures given at earnings calls (the latest month multiplied by 12; unaudited, and not included in the interim reports). Using first-half revenue multiplied by two, the multiples are 35x and 120x, respectively. That’s a big gap, but it alone doesn’t mean the companies have overstated revenue: according to Zhipu’s interim report, token call volume has grown more than 40-fold since the start of the year, and the average API price has risen about 101%. Based on volume multiplied by price, monthly API revenue in August is estimated at about RMB 900 million, or roughly $1.5 billion annualized—within the same ballpark as $1.6 billion. That reflects the August run rate; subsequent disclosures will show how long it can be maintained.

【What would Zhipu be worth at MiniMax’s valuation? (Scenario estimates)】
- Zhipu has about HK$48 billion in net cash, equivalent to roughly HK$99 per share. This does not mean the share price has a floor at that level.
- The two companies had similar first-half revenue (about $130 million vs. $120 million), yet Zhipu’s enterprise value (market cap minus net cash) is 4.4 times MiniMax’s.
- Using HK$518 as the reference price: applying MiniMax’s enterprise value-to-revenue multiple implies HK$207 (60% lower); its enterprise value-to-gross-profit multiple implies HK$258 (50% lower); and its enterprise value-to-annualized-revenue multiple implies HK$288 (44% lower). Applying Anthropic’s 14x funding-round multiple implies HK$467 (10% lower); applying OpenAI’s 17x funding-round multiple implies HK$535 (3% higher).
- Valuing Zhipu using MiniMax’s measures puts it 44%–60% below the current price; using the US companies’ funding-round valuations puts it near the current price. The key to this wide gap is whether Zhipu can sustain its revenue going forward.

【How fast does growth need to be to support the current price?】
Assuming a 10% annual return and a 25x price-to-earnings ratio in five years: based on first-half reported revenue, Zhipu would need revenue to grow 84%–90% annually for the next five years, versus 43%–49% for MiniMax. Starting from annualized revenue of $1.6 billion would bring Zhipu’s required growth rate down to 39%; if it reaches Goldman Sachs’ year-end forecast of $3.2 billion, the rate falls to 21%.

【Another hurdle: the lock-up expiry around January 2027 (estimate)】
About 178 million Zhipu H shares will come out of lock-up, 2.1 times the roughly 85 million H shares currently in free float. That’s equivalent to 38 days of trading volume, based on the average over nearly the past 20 trading days. If 10% of those shares are sold, that would amount to about HK$9.2 billion, or four days of trading value. MiniMax already went through a lock-up expiry in July: about half its share capital was unlocked, and the stock fell 26% over two days.
According to company announcements, Zhipu raised about HK$70.5 billion after listing. Its placement prices of HK$1,588 and HK$714, and conversion price of HK$892.5, are all above the current share price. Its cash can support the business for about ten years. MiniMax’s placement price of HK$268 and conversion price of HK$335 are also above the current price. Funding pressure is low, but investors holding shares at higher prices could create selling pressure during a rebound.

【What to watch next】
- Whether Southbound Stock Connect funds sell for two consecutive weeks.
- The 2026 annual reports: if the companies maintain their disclosed annualized run rates, Zhipu’s full-year revenue should be at least about 6.3 billion yuan, and MiniMax’s at least about $500 million.
- Anthropic’s reported IPO pricing in mid-November: a successful listing could prompt the market to compare valuation multiples across the two markets again; a weak debut could weigh on valuations for similar companies.
- Whether Zhipu’s major shareholders agree to extend their lock-ups before January 2027; and the listing progress of Kimi and DeepSeek.

These are valuation estimates under different assumptions, not share-price forecasts.

If you could hold only one: would you choose MiniMax, which has already fallen sharply and is cheap but growing slowly, or Zhipu, which is still 30% more expensive but was growing at a blistering pace in August? Or would you short AI model stocks, following the plunge in Korean equities, semiconductors, and memory stocks? Leave your pick in the comments~

AI’s money is borrowed: CoreWeave ($CRWV ) fell 25% in a week, and its bond yields reportedly approached 15%. Why does a tremor there shake the entire AI chain?

【First, who are they?】
CoreWeave, ticker CRWV on Nasdaq, started out as an Ethereum mining company and pivoted to GPU cloud computing in 2019: it buys Nvidia GPUs, builds data centers, and rents them out under contracts to customers like OpenAI, Microsoft, and Meta. It doesn’t build models or sell chips; it’s the link in the AI supply chain that “borrows money to stock up on GPUs, then rents them out.” Companies like this are called neoclouds. It went public at $40 on March 28, 2025, peaked at a closing price of $183.58 on June 20, and closed at $61.39 on October 16.

【The books (SEC 10-Q, as of June 30, 2026)】
- Revenue: $2.575 billion in Q2, up 112% year over year; $103.7 billion in contracted backlog; 98% of revenue came from signed contracts; the three largest customers accounted for 36%, 26%, and 10%
- Losses: Q2 net loss of $626 million; first-half loss of $1.37 billion
- Debt: $35.55 billion on the books, up from $8.7 billion in March 2025—more than quadrupling in just over a year
- Interest: $558 million in Q2, or $2.2 billion annualized—more than a fifth of revenue
- First-half capital expenditure was $14.1 billion, while it issued $16.7 billion in debt over the same period: virtually all the money it spent was borrowed
- Total liabilities: $72 billion; shareholders’ equity: $5 billion; $6.2 billion comes due within a year
- The cost of new debt: an August $2.6 billion loan at SOFR plus 5.5 percentage points; then $3 billion in convertible bonds issued in September

【What happened this week?】
- On October 14, the stock fell 15.5% in a single day. Yields on bonds maturing in 2031 and 2032 reportedly approached 15%. In an internal letter to employees that day, the CEO expressed confidence in funding, sales, and orders, according to Bloomberg, saying that “the weaknesses of competitors have seeped into the narrative.”
- It closed at $61.39 on October 16, down 25% for the week and 67% from its peak
- High-yield bond spreads: 311 bp on October 9 → 325 bp on October 14 → 323 bp on October 15

【Why does a tremor there shake the entire chain?】
In this AI capacity buildout, the giants rely on their own cash flows, while neoclouds and memory manufacturers rely on borrowing. As interest rates rise—with the 10-year Treasury at 5.1%—the first link to buckle is the one with the heaviest debt and thinnest cash flow. CoreWeave’s bond yields are the market’s price quote for “the cost of financing AI capacity.” When that quote jumps, positions across the board—from triple-leveraged semiconductor plays to Hynix to $BTC long positions—get repriced overnight.

【Don’t overstate either side】
- It’s not a shell company: $103.7 billion in contracts, revenue doubled, and customers include OpenAI, Microsoft, and Meta. The catch is that it has to spend the money to build capacity before it can fulfill those contracts—and that money is borrowed at double-digit interest rates.
- It’s not safe either: debt has quadrupled, interest eats up a fifth of revenue, and $6.2 billion comes due within a year. Its ability to refinance depends entirely on the mood of credit markets.

【What to watch】
- Its bond prices, which speak before the stock price does
- Whether high-yield bond spreads can fall back below 310 bp
- Meta’s capital expenditure in its October 21 earnings report: if customers ramp up spending, CoreWeave’s contracts become more valuable; if customers pull back, it’s still the first link in the chain to take a hit

Is a 15% bond yield a bargain created by excessive selling—or a warning sign?

What to watch next week (Oct 19–25): Fed blackout, a macro data vacuum; markets take their cues from earnings and breaking news

【Scheduled events, Beijing time】
- Mon, 10/19, 10:00: China’s Q1–Q3 GDP (in past years, it has been released on the 19th or 20th; check the National Bureau of Statistics); 20:30: U.S. September housing starts
- Tue, 10/20, 09:00: China’s LPR; Netflix earnings after the close (early morning on the 21st in Beijing)
- Wed, 10/21: AT&T before the open; Meta, Tesla, Texas Instruments, IBM, and Lam Research (semiconductor equipment) after the close—all at 04:00–05:00 on the 22nd in Beijing. This is the biggest night of the week
- Thu, 10/22, 20:30: U.S. initial jobless claims; Procter & Gamble, Honeywell, Lockheed Martin, and Blackstone before the open; Western Digital (storage) after the close
- Fri, 10/23, 16:00: Deribit weekly options expiry; preliminary October PMIs for the U.S., euro area, and Japan
- No CPI, PPI, or payrolls-level data all week; the Fed enters its pre-meeting blackout period on 10/17, with no official remarks through 10/29

【Coming after next week】
- FOMC meeting, 10/27–28; statement at 02:00 on 10/29. ECB meeting, 10/28–29; Bank of Japan meeting, 10/29–30
- GOOGL and MSFT earnings on 10/28; AMZN, COIN, and MSTR on 10/29; Apple on 11/2
- Strategy special shareholder meeting on 10/28; monthly options mega-expiry on 10/30; MSCI’s final list of non-operating companies on 11/11
- U.S. midterm elections on 11/3

【Three things to watch in earnings】
- Meta’s capex guidance: the AI chain sold off this week on fears of “borrowing money to build capacity.” Whether the giants are willing to keep ramping up investment will provide the first answer
- What Lam Research and Western Digital say about storage demand: storage stocks were among the hardest hit this week. Earnings could either stop the bleeding or trigger another leg down
- Tesla will make headlines, but it won’t have much impact on AI-chain pricing

【Unscheduled powder kegs】
- Asian stocks at Monday’s open: U.S. AI stocks fell on Friday, and storage-stock and Korean-stock contracts on Binance plunged 7–9% early Saturday morning. The KOSPI, Samsung, and SK Hynix have yet to reflect this
- Strait of Hormuz: reports say a tanker hit a mine over the weekend, and Iran says the strait is closed. Brent is at 109.69; a move above 110 would bring the inflation narrative back
- AI credit: CoreWeave is down 51% from its high, and high-yield bond spreads widened 14 bp in three days. If another neocloud or storage company runs into financing trouble, this week’s sell-off could happen all over again
- U.S.–China: export controls and the ChangXin DRAM story could produce news at any time

【Crypto’s own position】
$BTC has closed below the bottom of its range at 82,563 for four straight days. The October 15 low of 79,732 has not been breached. Funding rates have been positive for seven consecutive periods, with longs paying to wait for a rebound. In a macro data vacuum week, keep an eye on Korean stocks during the day and Meta’s earnings that night.

Which one worries you most next week?

October 17 market overview: Nothing moved during the day on Saturday; in the early hours, storage stocks in Binance stock contracts plunged 7–9% first. Crypto followed lower and then recovered, while BTC closed below the bottom of its range for the fourth consecutive day

【Market】
- “BTC 81,491,” −0.15%; high 81,770 (15:00 Beijing time), low 80,820 (the hour starting at 05:00 on October 18). The close was 1,072 below the range floor at 82,563; it has closed below that level for four straight days since October 14. The perpetual low was 80,751, above the October 15 low of 79,732
- Full-day liquidations: $117 million long / $41.8 million short; BTC accounted for just $10.6 million long / $1.7 million short. Most liquidations were in altcoins and stock contracts
- Perpetual open interest: 94,267 coins, unchanged; funding rates were positive for seven consecutive periods, settling at +0.0064% at 08:00 on the 18th
- Fear & Greed Index: 53, neutral; previous reading 55

【The early-hours drop】
- At 05:05 Beijing time on October 18, the 3x leveraged Korea-equity contract KORU fell first; the 3x leveraged semiconductor ETF SOXL followed at 05:15. At 05:20, SK Hynix, Samsung, Micron, and SanDisk all dropped, each falling 7–9% around 05:30. Nvidia fell just 1.2%, and the Nasdaq 100 ETF never fell as much as 1%
- ETH followed lower after 05:25, reaching a low of −2.7% at 05:39; BTC’s low was −1.1%. Storage and Korea-equity contracts led crypto by 5–20 minutes
- From 05:00 to 06:00, long liquidations across the market totaled $87.7 million: ETH $16 million, BTC $8.1 million, SanDisk $6.3 million, SOXL $4.6 million, and SK Hynix $4.4 million
- By the UTC close, most had recovered: Hynix −2.2%, Samsung −1.6%, Micron −1.6%
- No cause identified: around 05:30, X had only after-the-fact complaints, with no new information. This week’s backdrop includes two concerns about storage stocks (ChangXin’s new DRAM architecture and expectations of excess capacity in 2028–29), but these are context, not confirmed triggers. Spot stock markets are closed over the weekend, and these 24-hour contracts are thinly traded, making large moves more likely

【Altcoins】
- Among 61 USDT pairs with more than $5 million in trading volume (small weekend sample): median −2.0%; 13 up, 47 down, and 11 down more than 5%
- ETH 2,402.8 (−0.9%), SOL −0.9%, XRP −1.3%, BNB −0.5%, DOGE −2.0%, HYPE −2.8%, UNI −2.5%, ZEC −1.9%, LINK −2.9%, SUI −4.4%, “ENA −5.3%,” NEAR −4.0%, AVAX −1.9%, QNT −2.6%, “$STRK −13.0%” (0.0645, down 16% in three days)
- The only top gainer was $EPIC +15.9% (delisting on October 28); the biggest losers were STRK, BAT −8.9%, ZRO −8.7%, PYTH −8.4%, and FIL −7.8%. No primary source confirming the cause, so no attribution

【Macro】No new weekend data; Kalshi puts the odds of an October rate hike at 14%. As of October 11, MSTR held 848,250 coins; it has been a net buyer for four consecutive weeks, but the purchase size is shrinking. MSCI’s final list is due November 11

【Seven bull-market confirmations】2/7, not triggered (readings through October 16)

【What to watch】Whether KOSPI and storage stocks catch up on Monday when Asian equities open: Friday’s US AI stocks fell, and contracts plunged in the early hours of Saturday, but Asian equities have yet to reflect those moves. Will BTC test 79,732?

October 16 Market Recap: Rate-Hike Odds Plunge from 30% to 14%, U.S. AI Stocks Keep Falling, BTC Closes Below the Range Bottom for a Third Straight Day

【Markets】
- “BTC 81,616” −0.77%; high 82,941 (19:13 Beijing time), low 80,809 (01:00 on October 17). Closed 947 below the range bottom of 82,563, marking its fifth close below that level in October and its third straight day since October 14
- During the 19:00 hour, BTC spiked to 82,941, liquidating $25.3 million in short positions. The dollar, Nasdaq futures, and gold were all unchanged over the same period, suggesting a crypto-specific short squeeze. After U.S. stocks opened at 23:00, BTC fell below 81,400 and tested 80,809 at 01:00, liquidating $23.7 million in long positions; it then moved sideways between 81,500 and 81,800
- Liquidations for the day: $94.4 million long / $59.8 million short; BTC: $18.3 million long / $16.2 million short
- Perpetual open interest: 94,276 BTC, little changed; funding rate settled at +0.0100% at 08:00 on the 17th, positive and rising for five consecutive periods
- Fear & Greed Index: 55, down from 63
- This week: −1.2% (82,636 → 81,616), weekly range 79,800–84,168

【Federal Reserve & Rates】
- Warsh reportedly gave no signal of a rate hike at the IMF meeting in Bangkok on October 16; Kalshi’s October rate-hike odds fell from 30% to 14%
- U.S. Treasuries (Treasury figures): 2-year 4.74% (−3 bps), 10-year 5.09% (−1 bp), 30-year 5.43% (unchanged)

【Equities】
- U.S. stocks: S&P 500 −1.09%, Nasdaq −2.14%, Dow −0.34%; Philadelphia Semiconductor Index −3.83%, down for four straight days and −11.4% this week; VIX 19.99, up from 16.46. Tech reportedly fell more than 3%; energy rose 2.5%
- MSTR −9.56% (MSCI rule change; see the October 17 morning post), COIN −5.79%
- Asian stocks: KOSPI +2.05%, its first gain after seven consecutive losses; Nikkei +1.39%, Hang Seng +0.52%, Taiwan stocks +0.30%, Shanghai Composite +0.24%. Asian markets closed before this latest U.S. sell-off; watch on Monday to see whether they catch up on the downside

【Commodities & Dollar】
- Brent 109.69 (+3.6%), WTI 99.04 (+1.1%); crude oil rose for a seventh straight day, with attacks on tankers by Iran reportedly continuing
- Gold futures 4,186.7 (+0.4%); Dollar Index 101.78, down from 102.00
- U.S. spot BTC ETFs (Farside): −$261 million on October 14; cumulative −$537 million from October 1–14

【Altcoins】
- Across nearly 70 USDT pairs with over $5 million in trading volume: median −1.4%, 18 gainers and 50 losers, with 12 down more than 5%
- ETH 2,423.5 (−0.8%), SOL −0.9%, XRP −1.8%, BNB −1.1%, DOGE −0.2%, HYPE −1.8%, UNI −1.3%, “ZEC −5.9%” (1,149), NEAR −3.7%, ENA −3.0%, LINK −0.9%, AVAX −0.8%, “$STRK −14.5%” (0.0741, giving back its two-day gains), QNT −1.8%
- Top gainers: $TIA +16.6%, SEI +4.2%, DOT +3.0%; top losers: WAL −12.2%, BAT −8.8%, PYTH −6.8%. Tokenized U.S. stocks MSTRB −8.7% and TQQQB −6.7%, moving in line with U.S. equities. No primary source for the cause was verified, so no attribution is made

【Seven Bull-Market Confirmation Signals】2/7, not triggered (readings through October 14)

【Weekend Watchlist】Can BTC hold the October 15 low of 79,732 (perpetuals)? Further impact of the MSCI rule change on treasury stocks; can Brent hold above 110? Will Asian stocks catch up on Monday?

MSCI makes its move: MSTR plunges 9.6% in a day, Bitcoin treasury stocks tumble across the board, while Bitcoin falls just 0.6%

【First, two names to know】

“MSCI”: One of the world’s largest index providers. The MSCI Global Indexes are tracked by a large number of index funds. When a company is included in an index, those funds have to buy its shares in proportion; when it’s removed, they have to sell. Inclusion in an index directly determines whether a stock gets passive buying support.

“MSTR”: Strategy (formerly MicroStrategy), Michael Saylor’s company. It used to make software, but since 2020 it has put all the money raised by issuing shares and borrowing through bonds into buying Bitcoin. It now holds 848,250 coins—the most of any publicly listed company—which is why it’s called a “Bitcoin treasury stock.” Its share price has long traded above the value of the coins it holds. That premium has been sustained by continuously issuing shares and buying more Bitcoin.

【What happened】

- After the U.S. market closed on October 15, MSCI announced: “Non-operating companies” will no longer be eligible for the MSCI Global Investable Market Indexes. It will first check whether a company has substantial operating assets. Companies that fail that test will then be assessed against five financial ratios; those that meet four or more of the criteria will be excluded.
- Existing index constituents will be removed only if they fail to qualify for two consecutive annual reporting periods. “The list will be announced on November 11 and take effect after the close on November 30.”
- MSCI will also end its previous temporary freeze on companies whose “digital assets account for ≥50% of total assets.” These companies will be assessed under the general rules going forward.

【The reaction on October 16】

- “MSTR closed at 136.83 (−9.56%),” with 31.77 million shares traded—1.9 times the average volume over the previous five days. Other treasury stocks: NAKA −11.9%, SBET −7.5%, BMNR −7.3%; $COIN fell 5.8%.
- “$BTC spot closed at 81,745 (−0.61%),” while spot ETF IBIT fell 0.9%. Bitcoin barely moved; it was the premium of treasury stocks relative to their Bitcoin holdings that fell.
- Metaplanet in Tokyo, by contrast, rose 4.5%; the Japanese market has yet to follow suit.

【Why it matters】

- Removal means passive funds tracking MSCI have to sell. Analysts have reportedly warned that the resulting passive selling could run into billions of dollars.
- MSTR’s flywheel is “issue shares / preferred stock / convertible bonds → buy Bitcoin.” Passive demand from its index membership is one of the factors supporting its ability to issue shares at a premium. Without that support, “the premium could narrow, reducing the efficiency of issuing shares to buy Bitcoin.”
- From October 6 to 11, it bought only 250 coins; that same week, it repurchased $192 million worth of its STRC preferred stock.

【What’s still undecided】

- We won’t know until November 11 whether MSTR will make the removal list; the announcement didn’t say whether it has failed to qualify for both annual reporting periods.
- Saylor’s response on October 16 was a 21-second AI video: wearing a suit, he walks into a Tokyo alley, knocks over two bears, and walks out. Not a word spoken.

Do you think MSCI will really kick it out?

I took advantage of Binance delisting $EPIC to manually open a small short position. I didn't dare use much leverage, so I only made a tiny profit 😃 Now I’ve got food money for the next few days 😂

October 15 Market Recap: Korean stocks hit a circuit breaker, PPI runs hot at 6.3% year over year; BTC dipped to 79,800 intraday, then made a V-shaped recovery, but closed below the bottom of its range for a second straight day

【Market Overview】
- “BTC 82,246” +0.69%; high 82,634 (October 16, 01:00 Beijing time), low 79,800 (during the 21:00 hour on October 15; perpetuals at 79,732, the lowest since September 18)
- Closed 317 below the 82,563 range floor: the fourth October close below it (October 8, 12, 14, and 15), and the first time it has closed below it two days in a row. It recovered above the floor the day after each of the previous three instances. During the 06:00 hour on October 16, it briefly reached 82,588, then fell back in the final hour
- Four phases of the day: ① 10:22–10:28, ETH led a cascade of long liquidations in the early Asian session; BTC wicked down to 80,226 and recovered within half an hour. Long liquidations across the market totaled $285 million that hour (ETH $122 million, BTC $65 million) ② At 20:30, PPI was released, sending BTC down to 80,520 in five minutes ③ During 21:00–22:00, BTC tested the lows at 79,800; another $33.1 million in longs were liquidated that hour ④ From 22:00–00:00, Treasury yields fell and U.S. stocks turned higher; BTC climbed back above 82,000 in two hours, liquidating $33.4 million in shorts
- Binance liquidations for the full day: $383 million in longs / $73.6 million in shorts; BTC longs $88.4 million / shorts $19.9 million
- Perpetual open interest: 94,329 BTC at 07:00 on the 16th (93,117 at 08:00 on the 15th); funding rate turned positive, settling at +0.0054% at 08:00 on the 16th
- Fear & Greed Index: 63, Greed; previous reading 53

【September PPI, 20:30 Beijing time】
- Final demand rose 1.3% month over month (August: +0.5%) and 6.3% year over year (August: +5.6%); core (excluding food and energy) rose 0.7% month over month and 5.0% year over year
- Energy rose 10.8% month over month and 35.5% year over year; oil was still the main driver
- Kalshi priced in a 30¢ chance of a rate hike in October, about the same as on the morning of the 15th. The hotter-than-expected data did not push rate-hike pricing higher

【Stocks】
- U.S. stocks: S&P 500 +0.33% (7,744.1), Nasdaq +0.08% (26,713.0), Dow +1.08% (52,177.7); Philadelphia Semiconductor Index −0.88% (11,581.2), down for a third straight day and −7.9% from its October 9 close; VIX 16.46. Money shifted out of semiconductors and into Dow components
- MSTR +5.6% (151.3), COIN +6.3% (183.8)
- Asia: KOSPI −7.57% (5,554), triggering a circuit breaker after falling 9.05% intraday; down 20.7% over seven consecutive sessions and 39% below its June high. Samsung −9.5%, SK Hynix −11.1%; Nikkei −4.9%, Taiwan stocks −3.4%, Shanghai Composite −1.6%, Hang Seng −1.0%

【Treasuries and Macro】
- 10-year yield 5.09% (previously 5.19%), 30-year 5.43% (previously 5.53%), 5-year 4.90% (previously 4.99%); 2-year 4.77% (Treasury Department figure, previously 4.82%). Despite hot inflation data, Treasuries rallied sharply. Reports attributed this to safe-haven buying amid concerns about AI-related credit risks; this has not been verified firsthand, so no attribution is made
- Gold futures 4,178 (+1.7%); WTI 96.2 (+1.1%), Brent 105.7 (−0.3%); U.S. Dollar Index 102.05 (previously 102.55)
- U.S. spot BTC ETFs (Farside): after October 14 figures were updated, outflows were $261 million; cumulative outflows for October through the 14th were about $537 million. IBIT data for the 15th is still missing

【Altcoins】
- Nearly 100 USDT pairs with trading volume above $5 million: median gain +0.9%, 70% rose, and around 20 gained more than 5%
- ETH 2,442.1 (+0.8%), SOL 105.44 (+1.1%), XRP +0.2%, BNB −0.1%, DOGE −0.9%, HYPE +1.1%, UNI +3.3% (7.43), “ZEC +6.4%” (1,221), “$NEAR +7.8%” (5.26), ENA +4.7%, LINK +0.2%, AVAX +0.2%, “$STRK +12.6%” (0.0867), QNT −0.3%
- Top gainers: BAT +25.3%, INJ +11.4%, ZRO +10.3%, PYTH +9.5%; top losers: EPIC −17.6% (to be delisted on October 28), WAL −13.1%. The reasons have not been verified firsthand, so no attribution is made

【Seven Bull-Market Confirmations】2/7, not triggered (readings through October 14)

【What to Watch】BTC has closed below 82,563 for two straight days; first, watch whether 79,732 holds. Can the Philadelphia Semiconductor Index stop falling? Will foreign investors in the KOSPI stop selling?