"Two Shocks, One Map"
The week threw everything at the two-country thesis: a Tokyo leverage unwind, a Beijing model that designs chips and an options expiry. The map held, the three stocks to carry it ended the week intact
This week’s edition of Closelook@Global Stock Markets, dated July 18, 2026.
Last week we traced the hyperscaler dollar to Asia and concluded the ex-US rally is a concentrated bet on the two countries that collect it. This week the market tested that bet with everything it had: the largest Tokyo unwind of the year, a Chinese AI model that claims to design chips in 48 hours, and a triple-witching Friday to amplify both. Taiwan’s ETF fell six percent on the week, Korea’s four and a half — with both now sitting well below their recent peaks.
And yet — walk the wreckage and the three stocks the whole map rests on ended the week structurally intact. This letter is about the difference between a thesis being sold and a thesis being broken.
1 · This Week’s Action
The cross-asset backdrop. The board reads risk-off with one loud exception. Oil extended its run (USO +10.8% on the week — at +79% year-to-date it is the macro question nobody can defer much longer), the mega-cap concentrates led the decline for a change (QQQ −3.1%, QTOP −3.0%, against SPY −1.2%), and the safety shelf went green: both Treasury tenors caught a modest bid (TLT +0.4%, IEF +0.4%) and bitcoin continued its repair (IBIT +2.8%).
Gold slipped 1.2% on the week but told the better story intraday: it broke the 4,000 belief line and reclaimed it the same Friday it was supposed to confirm the breakdown. The stress stayed inside equities — and inside equities, it stayed inside one sector.
The global sectors. Energy led again (IXC +3.3%) — two weeks running — followed by the defensive shelf: global REITs +2.3%, staples +1.5%, financials and healthcare flat. At the bottom, alone, sits global tech (IXN −3.8%). But the detail that matters most this week is inside the tech complex: semiconductors lost seven to nine percent (SOXX −7.4%, XSD −8.9%, SMH −6.9%) while software closed the week green — cloud +1.4%, cybersecurity +0.5%, expanded software +0.3%.
A ten-point spread between the semis and the software next door, in one week. Maximum violence, minimum contagion: the selloff refused to leave its lane.
The regions. A complete inversion of the year’s pattern. The leaders sold off — Taiwan −6.0%, Korea −4.6%, China A-shares −2.8%, Japan −2.8% — while virtually the entire periphery finished green: Indonesia +3.4%, Hong Kong +2.5%, Malaysia +2.0%, China’s H-shares +1.6%, New Zealand +1.5%, the UK +1.1%.
Last week we called the periphery’s strength a hand-off; this week it was a shelter. Note what did not happen: the money leaving Korea and Taiwan did not leave the region — it rotated inside it.
2 · The State
Shock one: the unwind. The week’s violence began in Tokyo, not in the AI trade’s fundamentals. Kioxia — the memory name that had become the poster child of the leverage crowd — fell 16% in a session and has roughly halved from its peak; the Nikkei dropped 4.7% on Friday with an intraday low near −6%; margin positioning built over months came out in days.
Seoul, closed for the holiday on Friday, never got to trade the worst of it — which sets up Monday, below. This is the correction’s mechanical half: positioning, not thesis.
Shock two: the model. On Thursday, Moonshot AI released Kimi K3 in Shanghai with a claim attached: a 48-hour autonomous run that designed a working inference chip using only open-source tools. On Friday, the two American EDA software majors — Cadence and Synopsys — lost 9.5% and 8% respectively, roughly fifteen billion dollars of combined market value.
We published a full three-part analysis of what the demonstration is, what it is not, and what it does to the silicon value chain — Part I · Part II · Part III — and the one-paragraph version for this book is this:
The design shock does not touch the Asian slice of the dollar. Recall the flow map: of every $100 the hyperscalers spend, $30–40 lands in Asia — and it lands in fabrication, memory, packaging and assembly, not in design software. K3’s claim, taken at face value, makes chip design abundant and cheap. Abundant design means more tape-outs, more wafer starts, more HBM stacks, more packaging — all flowing through the same few qualified Asian manufacturing ecosystems. The barbell result from Part II applies directly: value drains from seat-based design labor and concentrates at the physical layer. The physical layer is precisely what Korea and Taiwan sell.
The three majors, name by name. TSMC: a beneficiary twice over — more designs seeking fabrication, and its decades of process data becoming the training asset for AI-in-manufacturing; its Friday decline (−2.8% in New York, −3.4% at home against a −4.9% Taiex) was relative strength on both boards. Samsung and SK Hynix: memory takes $38 of every AI-rack $100, and while custom inference chips may carry less HBM each, there will be more of them — volume diversity is not demand destruction. None of the three is on the disrupted side of the K3 barbell. The week sold their ETFs; it did not touch their position in the flow.
The fabless check. If Friday had really been about “AI kills chip designers,” the pure design-side ETF should have been hit hardest. It wasn’t: SMHX — the fabless basket — closed Friday at −1.6% against SMH’s −2.2%. The fabless ETF outperformed the full-chain ETF on the very day the design layer was supposedly shot. The reason is in the holdings: SMHX owns no EDA software (Cadence and Synopsys are software companies, not fabless semiconductor makers) — it owns the architects, who are two-sided in this story, and ARM, which rose two percent on Friday as the market priced more design starts as more royalty streams. Our verdict for the book: watch, don’t warn. The market sold two software tickers, not the design trade.
SK Hynix and the only line it has. Brutal week for the debut — the ADR fell from its highs to $154, dragged by Kioxia’s collapse next door, and on Friday it traded through the $149 offer price to an intraday low of $145.57. Then it did what the whole tape did on Friday: reclaimed the break, closing at $154.03, back above the line. Last week we wrote that below $149 the demand narrative flips from proof to fade. Friday put that exact question to the market intraday — and the market answered by buying it back. A pierced-and-reclaimed line is not a broken line. It is, however, now a tested one.
Monday is the session that counts. Seoul reopens with Friday’s entire global round-trip unpriced. The US market has already voted on how that goes: EWY — Korea’s US-listed ETF, trading while Seoul slept — opened Friday down 4%, expecting a catch-down, and spent the day rallying to close at −0.5%. That intraday path is the market front-running a gap-up reopen, the same pattern as SK Hynix’s V off its low. If Seoul instead gaps down Monday, the recovery thesis has a problem. Watch the open in Seoul before believing anything Friday said.
The structural read. The honest note first: last week’s test failed. We wrote that the channel read stays constructive “as long as pullbacks hold the 50-day shelf” — VEU closed Friday at 81.57, about 1.8% below its 50-day (83.07), the first close below that shelf since the spring trend began.
The long view puts that break in its place. On the weekly chart from the September 2022 low, the advance counts as a large third wave, and inside it five smaller waves are on the board — the fifth being the run from April’s washout to the 85.74 high. Which sets up exactly one question with two answers. It is technically possible that the large third wave is complete: five waves up, done, and this week’s break is the opening act of the bigger consolidation — a fourth wave of the same degree that has been rising since 2022, which would be a correction measured in months, not weeks. Our lean, held loosely, is the other answer: that final fifth still looks small against its siblings, and extensions of this cycle have run further — one more leg up before the larger rest.
We hold that lean with open hands, because the end of a fifth wave is precisely the place where leaning costs the most. What separates the two paths in real time: a reclaim of the 50-day and then 85.74 extends the count; a clean loss of the 78–80 support band — where the horizontal shelves and the trendline through the April 2025 and April 2026 lows converge, and which we expect to hold — confirms that the larger consolidation has begun. Either way — and this is the part that matters for the book — the structure would be corrective within an intact multi-year advance, not a broken trend.
3 · The Outlook
The four indices. The week’s damage report sorts exactly by physical exposure. Rubin — the build-out index, the most semiconductor-physical of the four — took the full hit: −9.3% on the week, still +92% year-to-date. The Agentic Ecosystem index gave back −3.3% (+42% YTD), HALO −1.9% (+1% YTD) — and AW40, the agentic operators, finished the washout week green at +0.3% (−20% YTD).
Read that ladder carefully: in a week allegedly about AI disruption, the index of AI operators outperformed the index of AI hardware by nearly ten points. The market did not sell the AI economy this week. It sold the leveraged expression of its supply chain — and the OpEx footnote below applies to all of it.
4 · What May Lie Ahead
Levels. VEU: the 50-day (83.07) is now overhead resistance to reclaim, and below the market sits a well-defined support band between 80 and 78 — the 80.73 shelf first (the May–July congestion floor), then 78.77 (the spring breakout area). The band carries a second, independent argument: the rising trendline connecting the April 2025 and April 2026 lows arrives in the same zone — horizontal congestion and the channel’s lower rail stacked on top of each other. Confluence like that is where we expect the pullback to hold; it is also the cleanest tripwire we could ask for — losing 78 would break the shelf and the two-year trendline in one move, and hand the count to the bigger-consolidation branch with no ambiguity.
SK Hynix: $149 — tested, held, and now the single most-watched line in the global book. Korea and Taiwan both closed well under their 50-day averages after living above them since spring; leaders proving a pullback is rotation, not distribution, remains the test it was last week — now at higher stakes.
Monday, Seoul. The reopen is the confirmation session for the whole week — for the unwind (does the gap-up template print?), for the memory complex (does SKHY’s reclaim hold with its home market open?), and for the verdict-day reversal generally. The OpEx asterisk hangs over everything until then: Friday was a triple-witching expiry, and both the violence and the reclaim were partly mechanical. Monday is the first clean tape.
July 27 — the verification event. Kimi K3’s open weights and technical report are promised for July 27. That is the date the design-shock narrative either becomes reproducible fact or stays a company demonstration — and either outcome moves the EDA names far more than it moves this book. What this book watches instead: whether mature-node design starts, shuttle runs and mask orders begin inflecting over the coming quarters. That would be the design explosion arriving at the physical layer — Asia’s layer.
Earnings. The season accelerates: Tesla prints Wednesday (our pre-earnings card #4 is live), the first European industrial and semiconductor reports follow, and TSMC’s monthly sales land mid-week as the recurring pulse-check on the foundry leg. After this week, every Asia-linked print is a referendum on whether the capex flow is intact. TSMC already answered once — its Thursday report last week raised capex guidance to $60–64 billion. The demand side keeps saying yes.
The regime backdrop. Money Temperature composite at 44 — down ten points on the week, out of the risk-on band and into neutral. That is the regime gauge registering the unwind in real time: cooler, not cold. Rotations continue to get funded at 44; euphoria does not.
5 · The ETF Portfolio — Global ETFs
What we did this week. Nothing — the first zero-transaction week since the Friday rebuild, and the right kind of nothing. The thirteen-position book went through the stress week untouched and finished at a market value of $270,129, still +2.4% unrealized on cost. The damage report inside it reads exactly like the letter above: Korea is the scar (EWY −12.0% unrealized — opened near the top, now the book’s stress-test position), Taiwan barely marked (EWT −2.4%), and the quiet hero is the fabless sleeve — SMHX at +35.8% unrealized, the book’s best line — on the very week the design layer was supposedly disrupted. The core did its job: VEU sits at +2.6% unrealized across the whole episode, and the mega-cap sleeves (QTOP +24.8%, TOPT +27.6%) kept compounding. Gold’s −9.2% on cost is the price of the hedge, not a verdict on it.
The stance. The book is the map this letter keeps drawing: one broad ex-US core (VEU), the two legs where the capex dollar lands (EWY, EWT), the fabless expression (SMHX), the mean-reversion laggard (India), index-level US tech, momentum sleeve, gold and bitcoin as the hedges. A week like this is what the construction was built for: the legs took the full unwind, the core took a third of it, the hedges took none of it. Nothing in the week’s two shocks argues against where the dollar lands — the K3 analysis, if anything, reinforces the physical-layer thesis the legs express. We do not trade launch events, and we do not trade OpEx Fridays.
What we plan to do. Nothing before Monday’s Seoul reopen answers the gap question, and nothing on the K3 news — the trade there, per Part III, is still in front of us and it is not an ETF trade yet. Watch-items carry over: the periphery’s month (India and Southeast Asia as the broadening evidence), Korea’s digestion turning orderly or not, and VEU’s date with its 50-day from below.
6 · What May Go Wrong
Three ways this letter misleads us. One: Monday gaps down. The gap-up template is a US-market inference; Seoul gets the actual vote. A catch-down Monday through SKHY’s $149 with the home market open would flip the debut story to fade and the unwind story to continuation.
Two: the fifth wave was the last wave. Our lean is one more leg up before the bigger consolidation; the count allows the opposite — that the advance from 2022 completed at 85.74 and the months-long fourth-wave rest has already started. If next week fails to reclaim the 50-day from below, the lean was wrong and the consolidation scenario takes the baton.
Three: oil finishes the job the unwind started. USO +10.8% in a week with +79% on the year is a rate-path problem wearing a commodity costume; a hot print into this makes the defensive rotation permanent rather than protective.
7 · Knowledge Corner
The barbell — why cheaper design can be bullish for the factories. This week’s paradox deserves its own explainer: an AI model designs a chip, and the right response may be to worry less about the companies that manufacture chips. The mechanism is the design/manufacturing barbell. When the cost of designing silicon collapses — from tens of engineer-years toward a two-day compute bill — the number of chip designs explodes: sovereign chips, robot chips, vehicle chips, single-purpose accelerators.
But every one of those designs must still be fabricated, packaged, tested and supplied with memory, and the capacity to do that is finite, capital-intensive and concentrated — overwhelmingly in Taiwan and Korea. Design becomes abundant; production stays scarce. Value drains from the middle (design labor, seat-priced software) toward the two ends: the models that generate designs, and the physical plants that realize them. That is why Friday’s selloff hit design-software tickers hardest while the fabless basket outperformed — and why the flow framework this letter is built on treats the design shock as confirmation, not threat.
The full argument, layer by layer: The DeepSeek Moment Moves Into Chip Design, The Rotation, Not the Apocalypse, and The Index Was Built for This.
8 · Final Words
A thesis you cannot stress-test is a story; this week the market ran the test for us. A leverage unwind sold the two legs double-digits, a design shock questioned the value chain they sit in, and an expiry amplified everything — and at Friday’s close the map looked like this: the three companies that collect the Asian slice of the AI dollar ended the week as beneficiaries of the very shock that was supposed to threaten them, the fabless basket outperformed, the debut held its offer line on a close, and the money that fled the leaders hid in the same region’s periphery.
What broke this week was positioning. What holds is the flow. Monday’s Seoul open tells us which one the market believes — probability, not prophecy.