The Morning 10

What

The US semiconductor selloff went global overnight: the Nikkei fell more than 4,100 points at the afternoon worst — down 6.2%, below 63,000 — and closed off 4.7% at 63,674, with memory-maker Kioxia diving as much as 16%, its market value now roughly halved from last month's peak. The Taiex lost 4.9% to 43,399 the day after TSMC's record print — though TSMC itself fell only 3.4% in Taipei, outperforming its own index. South Korea's market didn't trade — the one major chip market that sat the session out.

If

Asia's close is the floor for the US premarket, the verdict day opens with the selling already priced: September Nasdaq-100 futures are down 1.65%.

Why

The chain that led the selling in New York — memory, foundry, AI infrastructure — is headquartered in Asia. When Tokyo and Taipei confirm the move instead of fading it, the repricing is about the complex, not about one session's positioning. The Kioxia framing making the rounds in Tokyo — leverage built against AI names being unwound — is the deleveraging read of the same tape.

Then

Watch whether the US premarket holds Asia's lows once real volume arrives — and remember Seoul reopens Monday with Friday's global rout still unpriced.

What

SOXX fell 4.5% to 530.50 — through the 554 floor without a pause, all the way to the June capitulation zone, and stopped exactly on it: 530–532 is where the June 9 washout low (522.24 intraday on record 24.5 million volume, closed 562), the 539 double bottom, and the May high stack at one price. SMH did the opposite: it pierced its 566.83 triple-bottom intraday to 564.45 and closed at 568.92, back above the line. Two benchmark indices, same industry, opposite sides of their last lines.

If

The open confirms the futures — Nasdaq-100 September contracts down 1.65% put both indices below their lines at the bell — the entire session becomes one question: what do the closes reclaim. Wednesday is the template: a fifteen-point intraday break of the 554 floor, fully reclaimed by four o'clock. An opening print below a floor is not a verdict; the close is.

Why

A floor built from a capitulation low, a double bottom, and a prior breakout stacked at one price is the most heavily defended line in the structure — and a market that had genuinely chosen direction wouldn't leave its two benchmarks disagreeing: SOXX says distribution confirmed, SMH says the fourth test got bought. That makes today one of the most important sessions of the year for this complex.

Then

Below 530 there is only 522, and below that, air; SMH's line is 567. Friday's two closes are the week's entire scoreboard — Sunday scores them into the track record — and monthly options expiration adds mechanical flow to whichever side wins.

Open the detail →

What

The rout has a mechanism, not a headline: July of a midterm year is seasonally one of the equity calendar's weakest stretches; money is selling ahead of the big tech reports because the reaction function has turned punitive — a record quarter from TSMC and growth from Netflix were both sold this week; the AI-doesn't-produce-results chatter is back in circulation; and underneath it all sits the specific fear that the Mag 7 use the late-July calls to guide spending growth down.

If

Both halves of the fourth force land, the perfect storm assembles: the suppliers just raised capacity — TSMC to $60–64 billion, ASML adding 30% EUV capacity for 2027 — and the Mag 7 would be guiding spending growth down into it. Capacity expanding into decelerating demand is the double negative the bear case needs. If no hyperscaler cuts, three of the four forces expire with the calendar.

Why

Three of the forces are positioning and mood — they pass. Spending growth touches the complex's actual revenue — and here the counter-frame is already emerging: Meta has begun selling third parties access to its newest models, and the chatter has Oracle next. Hyperscaler capex that produces external revenue doesn't need faith to justify itself — that is the argument the spending lines have to survive these calls.

Then

The Mag 7 report over the next two weeks; every capex line and every external-revenue line is now the market's most-watched pair of numbers.

Open the detail →

What

The record print got its answer: TSMC raised full-year capex to $60–64 billion from $52–56 billion — citing 'structural demand including the newly emerging agentic AI market' — lifted full-year growth guidance above 40%, and added $100 billion to the Arizona build. The ADR fell 2.3% to 409.74 anyway, trading as low as 403.50 before a late bounce — and on Friday the home listing fell another 3.4% to NT$2,385, though it outperformed a Taiex down 4.9%.

If

A stock cannot rally on a record quarter, a raised guide, and a raised capex plan, the selling is about ownership and positioning, not about the business.

Why

The Print Record card published before the print said exactly this: TSM's two heaviest AI-era reactions were 4.5% drops after clean double beats. The pattern held on the biggest beat yet — the market is repricing what it will pay for confirmed AI demand, not doubting the demand.

Then

The capex raise is this cycle's hardest data point that the buildout continues — file it against every demand-doubt headline the selloff produces.

Open the detail →

What

Netflix reported growth in both revenue and profit after Thursday's close — and got sold anyway: down 5.2% in the first after-hours reaction to 70.45, below the 52-week low of 70.86, and the overnight bid kept sliding to about 67.60, roughly nine percent under the close. The release also cut disclosure: the viewership report goes annual instead of semi-annual.

If

The regular session confirms the overnight level, Netflix opens at prices last seen in its drawdown lows — after a quarter in which the business grew.

Why

Yesterday's read said prints from weakness carry the most information: a name in an accelerating downtrend that sells a beat is telling you the market wanted an excuse. It found two — the growth wasn't enough, and less disclosure reads as management seeing numbers it would rather report less often.

Then

The calendar's next print from weakness is Tesla on Wednesday — from 391, below both moving averages, 22% off its high — and it arrives with the Musk premium freshly repriced: SpaceX closed at 131.11 on Thursday, decisively below its $135 IPO price and 42% off its post-listing peak. The card published last night shows Tesla's inverted record — its biggest post-print rallies followed misses, and the last four prints went nowhere. Two Musk-priced narratives get marked in one week; the market's willingness to pay for the story, not the numbers, is what's on trial.

Open the detail →

What

Gold gave up the line it had defended through the entire 28% drawdown: GLD fell 2.0% to 364.96, putting spot in the 3,950 area, below 4,000 for the first time since the top formed. The break came on a day when the dollar barely moved (UUP +0.3%) and long rates did nothing (TLT flat) — no macro push, just supply.

If

The break holds through a few sessions, the stalled-hedge story becomes a liquidation story: positions built for fiscal anxiety being unwound because the anxiety premium is leaving the market.

Why

Yesterday's read said it in advance: gold holding 4,000 was the last line between 'dead money' and 'forced selling'. A hedge that breaks support without a macro catalyst is being sold because it's owned, not because the world changed.

Then

Bitcoin at 49% off its high is the sibling trade — watch whether it confirms the unwind or diverges for the first time in the cycle.

What

On a day the semiconductor complex fell 4–6%, the rest of the tape barely blinked: XLF tagged a new 52-week high at 56.84 and closed green, the Dow slipped 0.2%, small caps closed flat, and SPY lost just 0.5% against QQQ's 1.6%. Index volatility finally moved off its floor — VIXY rose 2.5% to 20.56, one day after printing a 52-week low.

If

Breadth keeps absorbing what semis shed, the selloff stays a sector-degree event — the money leaving the AI complex is still not leaving the market.

Why

This is the bull's counterargument to the distribution read, and it held on the worst semi day of the week: distribution that stays confined to one complex while financials break out is rotation; distribution that spreads to the index is a top. So far the tape keeps choosing rotation.

Then

But the index is running out of air: semis already trade at their June lows — the Nasdaq-100 doesn't yet. Its June support zone sits at 28,550–28,200, roughly two to three percent below Thursday's close, and futures down 1.65% open the index at the zone's doorstep. A test of that zone moves the correction's degree question up one level — from the complex to the index.

Open the detail →

What

The memory chain had its second heavy down day in a row, and the violence escalated: Seagate fell 10.0%, Western Digital 9.2%, Micron 5.6% — and then Kioxia dove as much as 16% in Tokyo overnight, its market value roughly halved from last month's peak. Four sessions ago this group was swinging both directions; now it only swings one way.

If

The chain keeps falling with this force while TSMC raises capex and confirms the buildout, the market is repricing memory margins and pricing power specifically — not AI demand broadly.

Why

Memory is the complex's highest-beta expression of the same trade, and it has become the epicenter: the names that ran hardest on the AI-scarcity thesis are giving it back fastest. Epicenters mark where positioning was heaviest, not necessarily where the fundamentals are worst.

Then

SK Hynix last traded Thursday's 11% Seoul drop — Monday's reopen prices Friday's global memory rout in a single session.

What

IBM traded down to 204.44 — a fresh 52-week low, capping a one-third drawdown from its high — and then reversed to close up 3.7% at 219.05 on 22 million shares, its heaviest volume since the post-print collapse. It was the only green name of size in the entire complex.

If

The reversal holds above the 211 area it broke from, Thursday marks the first session where the memory-squeeze warning found buyers who think it's fully priced.

Why

A high-volume reversal at a new low is how downtrends make their first bottoms — sellers exhausted into strength they didn't expect. One day proves nothing, but it's the first day of the entire re-sort where legacy caught a bid while the disruptors fell.

Then

Watch whether IBM holds the reversal on a red tape today — a bounce that survives a down day means more than one that needed a green one.

Open the detail →

What

Jesper Koll — the longtime Japan strategist writing as the 'Japan Optimist' — has argued for years that Japan's bull market rests on domestic structural change: corporate governance reform, rising capex, and wage growth, not on any single export cycle.

If

N/A

Why

Against our tape: the Nikkei's worst day in weeks was made almost entirely of chip names — Kioxia, the AI supply chain — while Koll's domestic-reform pillars weren't the story at all. His framework poses today's cleanest question: is Tokyo down 5% a Japan story or a semiconductor story wearing a Japanese index as a costume? How the non-tech two-thirds of the market closes answers it.

Then

Read his work and separate the index from the complex before concluding anything about Japan.

Jesper Koll · Japan Optimist ↗