The Receding Forest

First, the scale. The KOSPI rose +111.5% from January 2 to June 22. Most of that climb came from the two companies: +236.0%, versus +41.2% for the rest of the market and +2.4% for the KOSDAQ. It then fell −37.1% after June 22.

Two-thirds of the decline came from the two companies. Multiply their weight on 6/22 (55.67%) by their return: −24.5 percentage points. The rest contributed −12.4 percentage points. The two companies accounted for about 66% of the combined −36.9 percentage points (estimated using starting weights).

But the two declines had different shapes. The first was a collapse among the leaders. The two companies fell nearly twice as much as the rest. The second pulled everyone down. The two companies and the rest fell by almost the same amount, and the KOSDAQ fell further. What began among the leaders spread across the market.

So the weakest part now is the rest of the market, rather than the leaders. The two-company index stood at 56.1 on 10/30, above its July low (51.9). The rest stood at 71.9, below its July low (73.8). It fell as far as 70.9 on 10/26.

Debt has already fallen substantially. In July and mid-October, forced selling from margin accounts amplified the declines. That selling is smaller now. But cash waiting to enter the market (investor deposits) has fallen by one-third. There are fewer forced sellers, and less money available to buy.

The same parts fell overseas. In the US, from the October 6 peak, the S&P 500 fell −3.05% and the Nasdaq −7.84%, but the semiconductor index fell −19.1%★. The Russell 2000 barely moved, at −0.65%. In Japan, the Nikkei fell −12.5% from its June 25 peak, while TOPIX fell −1.3%. High-priced semiconductor equipment stocks carry substantial weight in the Nikkei. From October 6, Taiwan's TAIEX fell −9.25% and TSMC −8.3%.

Foreign investors had net selling of about KRW 203 trillion on the KOSPI this year (10/15, Hyundai Motor Securities compilation). The two companies accounted for 91.76% of net selling from the start of the year through 10/8 (Sidae report). Money leaving Korean stocks ultimately left from the same place.

With prices down this much, what happened to profits? We separate memory makers from the equipment makers whose machines build their factories.

The two memory companies' profits grew throughout the sell-off. SK hynix's quarterly operating profit rose from KRW 60.54 trillion in Q2 to KRW 83.33 trillion in Q3. Samsung Electronics' semiconductor division rose from KRW 89.2 trillion to KRW 106.5 trillion. Their combined quarterly profit rose from KRW 18.4 trillion to KRW 189.8 trillion in a year.

The market's estimates of future profits did not fall either. Average broker estimates (consensus estimates) for 2027 operating profit rose from 6/18 to 10/14: SK hynix from KRW 377.3 trillion to KRW 404.0 trillion (+7.1%), and Samsung Electronics (the whole company) from KRW 486.6 trillion to KRW 561.0 trillion (+15.3%). Over that period, SK hynix shares fell −51.4% from their 6/22 close, and Samsung Electronics about −38%★.

Estimates did not move in just one direction. From August to early October, as the won strengthened, estimates for the near-term quarter edged down (for example, Samsung Electronics Q4: KRW 122.6 trillion → KRW 119.1 trillion; SK hynix: KRW 89.0 trillion → KRW 88.1 trillion). Some brokers, such as Kiwoom Securities, sharply cut 2027 estimates (Samsung Electronics: KRW 414 trillion → KRW 352 trillion, 8/11). Allow for differences among data providers: 6/18 is an EPIC AI compilation; the 10/14 SK hynix report says only “market consensus”; the rest use FnGuide.

So what fell was the multiple, rather than profits. SK hynix's 12-month forward P/E fell from 6.6–6.8x in June to below 3x in late October. The market effectively halved the price it would pay for the same profits. Two explanations sit side by side.

Boboo from Sky does not choose between them. We record where evidence for a judgment is likely to appear first. Memory makers' profits respond with a lag because of contract prices. Shipments, inventories and outlooks at memory buyers move first (Section 04).

Equipment makers show the same pattern. All seven companies that reported earnings in October had higher revenue than a year earlier. Most also raised their next-quarter or full-year outlook. ASML raised its annual revenue outlook three times, from EUR 34–39 billion in January to EUR 44–45 billion in October. Lam Research's memory-equipment revenue rose +120% year on year. Samsung Electronics spent KRW 29 trillion on capital expenditure in Q3 alone.

But most share prices fell on earnings day. Lam Research fell −11.8% (10/22), BESI −10.7%★ and DISCO −3.8%★. The misses against expectations were small. And the declines began before earnings season. Many large down days were unrelated to company results (9/14 comments on slowing AI development; 10/16 US 10-year yield at 5.10% and an eight-day losing streak for the semiconductor index).

Industry forecasts still point upward. SEMI (7/14) forecast a 2026 wafer-fab equipment market of USD 143.9 billion (+23.1%), with +21.8% growth in 2027. On 10/21, Lam Research said the 2027 market would exceed USD 200 billion. Equipment follows memory capital expenditure (capex), so it is the second area to shake if memory demand weakens. For now, that signal appears only in share prices, rather than orders.

The diagnosis in one sentence. This sell-off did not result from shrinking profits. It priced in the view that profits were at their peak, just when earnings were strongest. Whether that belief is right will show up first in the numbers from memory buyers, rather than memory makers. Section 04 examines those numbers.

One pattern appears when we compare countries on October 30. Intel fell −28.58% in the US, its worst day since 1974, and the semiconductor index fell −7.20%★. AI components fell by similar amounts across countries. The national indices diverged sharply.

The depth depended on the components' weight in the index, rather than the size of their fall. The two companies make up just over half of the KOSPI, so the index fell by about half as much as the components. In Europe and Japan (TOPIX), components fell around −6% but the indices lost less than −1%. The same wind reached different depths in each forest.

Taiwan has yet to absorb the full impact. TSMC fell −1.86% in Taiwan. Intel's sharp decline unfolded later in the US session, where TSMC's American depositary receipts fell −5.95%. The first check is whether Taiwan's next trading day (11/2) closes that gap.

Currencies did not panic. On a day when foreign investors sold more than KRW 1.6 trillion of KOSPI stocks, the won actually strengthened (KRW 1,334.3). The Taiwan dollar strengthened too. The AI shock arrived through equities, rather than an exit of money. Brazil's political shock that same week came through currencies and rates instead (Section 06).

Across the week (10/23→10/30), the spread was narrow. The KOSPI rose +3.25%, the Nikkei +0.95%, Germany's DAX +2.5% and the Russell 2000 +1.29%. The Nasdaq (−2.00%) and TAIEX (−1.8%★) fell. Hong Kong's Hang Seng (+1.46%) and India's Nifty (+0.51%) also rose on 10/30.

Intel got through Q3 well. Revenue was USD 16.51 billion (+21% year on year), and data-center revenue grew +78%. The Q4 outlook broke down: revenue of USD 15–16 billion, below market expectations of about USD 17.1 billion. The company said PC demand would remain weak until memory supply and prices normalized.

When memory prices rise, someone pays. In July, Amazon raised its annual capital expenditure from USD 200 billion to USD 220 billion, citing memory prices. Global Q3 PC shipments were 62.7 million units, down −20.1% year on year. Smartphone shipments were 255.3 million units, down −22.0% (IDC; the largest smartphone decline on record). The same memory prices appear as an 80% operating margin at SK hynix, fewer products sold by PC and smartphone makers, and higher capital expenditure at cloud companies.

Prices have begun to diverge by product. NAND spot prices turned down first, and contract prices stopped rising. The NAND 128Gb MLC contract price was USD 30.61, flat in October after 21 consecutive months of increases. DRAM is still rising. The DDR4 8Gb PC contract price rose +5.77% in a month to USD 27.50 (reports described DRAM contract prices as record highs). TrendForce's 10/28 report said PC makers had cut next year's purchases as inventories grew, while smartphone makers had stopped restocking.

AI still presents a different picture. SK hynix said long-term agreements with 14 customers cover more than 50% of its 2027–2028 capacity, and expects demand to exceed supply through 2030. Samsung Electronics said 2027 HBM negotiations were complete and capacity was sold out. This is where consumer memory (PCs, smartphones and NAND) and AI memory (HBM and server DRAM) begin to separate.

Boboo from Sky tracks this weekly through a new question card (Q15, “Whose cost is the price of memory?”). Next checks: Korea's October exports on 11/1, Apple's earnings on 11/2, TrendForce weekly spot prices (whether DDR4 records its first weekly decline), and late-November contract prices.

“Cheap” and “expensive” do not predict whether prices will rise or fall. They describe where today's price stands relative to that asset's profits, book value and history. The gap between prices and profits widens in a market downturn, making this position easier to see.

Korea: the two rulers point in opposite directions. After October's decline, the KOSPI's forward P/E was about 4x. One compilation places it at the lowest level among 18 major countries since 2006 (The Seoul Economic Daily★). Relative to book value, it is nearly twice its average. This is what happens when profits are at record highs but prices assume those profits will not last. It is the same diagnosis as Section 02. Observers also note that memory P/E has repeatedly looked lowest at the earnings peak (Section 02).

US: ordinary relative to its own history, expensive relative to bonds. S& P 500 forward P/E is 18.7x, near its 10-year average (19.1x). But the 10-year inflation-linked government bond real yield is 2.63%, at the 98th percentile over 20 years (20-year average: 0.83%). Subtract that yield from the equity earnings yield and the gap is about 2.7 percentage points, thinner than the average (roughly 4.4 percentage points). Investment-grade corporate bond spreads are 89bp, at the 5th percentile; high-yield spreads are 326bp, at the 17th percentile. Compensation for risk is thin.

Currencies and bonds. The yen's real value (BIS real effective exchange rate, REER) is at the 3rd percentile over 20 years; the won is at the 20th percentile (September). Brazil's inflation-linked government bond (NTN-B 2032) real yield rose from 6.72% on 10/23 to 7.94% on 10/30, according to official ANBIMA data. Brazil's 5-year credit default swap (CDS) was 151.7bp, above earlier this month (below 108bp), but moved less than stocks, rates and the currency.

Places that look cheap come with reasons. For Korean memory, it is the earnings cycle; for Brazil, fiscal policy; for Argentina, 2027 repayments and elections. This map writes those reasons alongside the prices rather than erasing them.

A country repriced in a day. President Lula was reelected with 51.79% in the 10/25 runoff (Flávio Bolsonaro: 48.21%). Markets had priced in the opposite result. The Ibovespa fell −18.1% in the week, its worst since March 2020. The dollar rose +9.15% in a week to BRL 5.446, its largest gain since October 2008.

Closer to 2015 than 2002. The 2002 crisis was a foreign-exchange crisis. Today, reserves stand at USD 362.8 billion★, and direct investment covers 147% of the current-account deficit. The problem is domestic. Government interest costs are 8.86% of GDP, and the nominal fiscal deficit is 9.48% (12 months through August; central bank). Excluding interest, the deficit is a small 0.62%. The cost of money is the problem.

Congress will not be easy. In the lower house (513 seats) taking office in February 2027, the Bolsonaro-aligned PL is the largest party, with 121 seats. Lula's PT alliance has 88. Fiscal reform depends on deals with centrist parties. Lula said he would name a finance minister by year-end. Rate markets price the policy rate at about 15% at end-2027 (12.5% before the runoff; 13.75% now).

The connection to Korea. Korean investors' custody holdings of Brazilian bonds were about USD 449 million (about KRW 596 billion)★, according to a 10/28 report. Combining the week's rise in long-term government bond yields and the weaker real gives a double-digit loss in dollar terms (press estimate: about 18%★). Next checks: Brazil's monetary policy meeting on 11/4 (morning of 11/5 in Korea), the Focus survey (first post-election compilation), and the resumption of government bond auctions.

Argentina's risk did not begin in Brazil. Since July, domestic funding through dollar bond sales had closed off (issuance halted on 8/14), the economy weakened, and US rates rose. Lula's reelection came on top of that.

Bonds took the strain, rather than the peso. The wholesale exchange rate is ARS 1,538, about 21% below the band's ceiling (1,947.79). The band, the IMF and the US Treasury swap support the currency. Meanwhile, people pay 7.5% more than the official exchange rate for the dollar stablecoin USDT. The concern is 2027. January government bond interest is about USD 4.3 billion (reported). Foreign-currency maturities in 2027 total USD 25.4 billion including international institutions, of which USD 17.1 billion remains unfunded (LCG estimate). Average polls for the October 2027 presidential election show Milei at 42.2% versus Kicillof at 38.9%.

The other side of the ledger. Thanks to Vaca Muerta shale, January–September energy exports were USD 12.45 billion and the energy surplus was USD 9.43 billion. POSCO Holdings completed the upstream process of its second Salta lithium plant on 10/6, bringing annual capacity to 48 thousand tonnes. We record rising risk premiums separately from what comes out of the ground.

This is a different direction from July. When the KOSPI tumbled in July, Bitcoin turned up from its yearly low (USD 58,558 on 7/1). In this AI shock, it fell in the same direction. At USD 77,358, Bitcoin is −37.9% below its all-time high (USD 124,659 on 10/6, 2025).

Look at cash before prices. US spot Bitcoin ETFs had outflows of −USD 640 million on 10/30 and −USD 810 million over the latest 20 trading days (still +USD 4.94 billion over 60 trading days). Ethereum ETFs had outflows of −USD 430 million over the latest 5 trading days. The kimchi premium (Korean exchange price gap) was low at +0.85%, while US Coinbase was at −0.21%, showing slightly more selling pressure.

We do not call the bottom. We carry questions and write down dates. These are the Boboo from Sky question cards with new observations this week.

When the water recedes, the bottom appears. Three things were exposed this time: record memory profits, prices that assume those profits will not last, and the first cracks among memory buyers. Boboo does not say when the water will return. Boboo records the shapes of the stones on the bottom, and returns to the same place when the water recedes again.