Wall Street called it a rotation. Corning called it a cost.

📌 The one-line takeaway

On Tuesday, July 28, the US chip complex fell through the last price floor it had, while the Dow Jones Industrial Average gained 537 points and the market's fear gauge went down. The consensus word for this is "rotation." One earnings report suggests a harder word.

Today I read 🟠 ALERT — attention rising, not crisis — and my internal alert stays at LEVEL 3, the highest, for a 9th session. It only steps down after a four-part recovery checklist is met, and I score 0 out of 4.

Two numbers in this piece are missing on purpose, and I say so where they belong: the exact price of oil, and the market-implied odds on today's Federal Reserve decision. Neither cleared my three-source rule, so neither gets published.

📰 What happened

The session was made in Seoul. On Tuesday in Korea, SK Hynix closed 14.65% lower and Samsung Electronics fell more than 13%. That selling rolled through Europe and into New York, and it landed almost entirely on one industry.

SOXX — the main US semiconductor ETF, a basket of the big chip stocks and the benchmark I track every day — fell 4.80% to 491.46, its fourth straight decline. Trading was heavy: 14.9 million shares, 1.42 times the average of the last 50 sessions.

The number that matters is not the percentage. It is the geometry. Since July 17 the lowest price SOXX had ever traded in this decline was 498.54. On Tuesday its best price of the day was 498.48. The entire session, from first trade to last, took place below a floor that had held for eight sessions. It was not tested and broken. It was gapped through and never revisited.

Underneath, the damage was concentrated with unusual precision. Memory stocks fell 9.50% as a group: SanDisk −14.25%, SK Hynix's US listing −8.98%, Micron −8.85%, Seagate −8.53%, Western Digital −6.91%. Equipment makers fell 6.03%. Optical and networking names fell 5.90%.

And the flagship sat it out. Nvidia rose 0.25% to 197.01 — the first session in this decline where SOXX lost nearly five percent while Nvidia closed green. So did the cloud giants that pay for all of this: Alphabet +2.19%, Microsoft +1.09%, Apple +0.94%. Of the 9 stocks that rose across my 52-name watchlist, four were cloud giants.

The broad market, meanwhile, had a good day. The Dow rose 1.03% to 52,747.32. The S&P 500 added 0.21% to 7,428.78. The equal-weighted version of the S&P — which gives the same weight to the smallest company as to the largest, and so measures how many stocks are participating — rose 1.17%, beating the regular index for a fourth straight session.

Rotation, or a de-rating? One earnings report casts the vote

The word going around the market on Tuesday was rotation: money leaving a crowded trade and going somewhere else. It is a comforting word. It implies the money comes back.

I think Tuesday supplied evidence for a harder word, and it came from an unlikely place.

Corning — the glass and optical-fiber company, a supplier to the AI build-out rather than a chipmaker — reported a strong quarter. Core earnings of $0.78 a share, up 30% from a year earlier; core sales of $4.74 billion, up 17%. Both beat what analysts expected. Then it guided the current quarter below consensus and named the reason: memory prices as a headwind to margins. The stock fell 12.10% to 126.01, and it took the optical group with it — Coherent −10.31%, Astera Labs −7.89%, Marvell −7.77%, Credo −7.62%.

Read that sequence again, because it is the whole argument. For two years the price of memory chips has been a revenue line — the reason Micron, SK Hynix and SanDisk were among the best-performing stocks in the world. On Tuesday, for the first time in this cycle, a company downstream of them put those same prices in the cost column and cut its own guidance because of them.

A rotation moves money between sectors. A de-rating changes what a business is worth, because it changes what the business earns. A cost that travels down a supply chain is the second thing, not the first.

Two more facts point the same way. Memory stocks are now roughly 30% below their highs — a bear market, by the standard definition of a 20% fall — reached in about five weeks. And the pressure on them is not read as purely cyclical: several broker notes now put the peak in memory prices as early as 2027, and reporting I have flagged all week — a single-source exclusive by The Information, which I treat as unconfirmed — describes Chinese progress on the lithography machines that print circuits onto silicon, alongside a Chinese competitor in memory. Rotations do not usually arrive with a competitive thesis attached.

The honest counter-argument, which I hold in view: on the same day, Celestica — the company that assembles AI server racks — rose 10.04% after reporting revenue up 62% to $4.70 billion and raising its guidance for the year. Final demand for AI infrastructure showed no crack at all. Tuesday was not a demand event. It was a price-and-margin event. Which is precisely why "rotation" is the wrong word for it.

🚦 The signal board — my 9 gauges in 6 families

I run nine gauges, grouped into six families, and read them together rather than one at a time. Green is calm, amber is watch, orange is alert, red is broken.

Putting it together. Family A is compromised (one red), Families B and C are under tension, D and E are calm. That combination confirms ALERT. I do not read it as CRITICAL, because the deepest breakdown my framework asks for in the price-trend family has not happened: SOXX closing below its 200-day average, or a "death cross." SOXX is 22.95% above that average.

I flag that as a judgment call, and I flag something else with it: that cushion has fallen from 40.64% to 22.95% in four sessions. The single reading holding this at ALERT rather than CRITICAL is eroding faster than anything else on the board.

📏 The levels that matter now

• SOXX closed at 491.46, which is 25.08% below its 52-week peak of 655.95, set on June 22.
• The 30% mark sits at 459.17 — now 6.57% away. On Monday it was 11.05% away. In the last three sessions SOXX has moved −4.40%, −2.05% and −4.80%. The distance to that level is now smaller than a single session of the kind I have already seen this week.
• The new floor to watch is 480.50, Tuesday's low. A close below it, especially on genuinely capitulatory volume, would turn my volume gauge red.
• SOXX's 50-day average is 568.20, 13.51% above the price. Its 200-day average is 399.72, 22.95% below.
• Nvidia closed at 197.01, 16.71% below its May peak of 236.54, and 2.08% above its own 200-day average at 193.00. A close below that line would compromise the price-trend family structurally, and would feed straight through to my regime reading.
• The S&P 500 is 2.38% below its record close of 7,609.77 from June 2.

One ratio is worth stating plainly. SOXX is down 25.08% from its peak; the S&P 500 is down 2.38%. The chip complex is falling 10.5 times harder than the market it sits inside. That is not a market-wide problem yet. It is a very concentrated one.

🏭 The sector, group by group

Average move across my 52-name watchlist: −4.26%. Nine stocks rose.

• Memory −9.50% — SanDisk −14.25% (worst), Western Digital −6.91% (best of a bad group)
• Equipment −6.03% — Applied Materials −7.82%, Teradyne −4.22%
• Networking −5.90% — Corning −12.10%, Cisco +0.88%
• Energy and data centers −4.77% — Vertiv −6.27%, Eaton −3.11%
• Chip design −4.48% — AMD −8.15%, Nvidia +0.25%
• Speculative periphery −4.53% — Nebius −9.68%, Applied Digital +0.93%
• Servers and hardware −2.16% — Dell −8.15%, Celestica +10.04%
• EDA (chip design software) +0.24% — Cadence +1.80%, Synopsys −1.33%
• Cloud giants +0.48% — Alphabet +2.19%, Tesla −0.58%

The shape is the story. The two groups that rose are the ones that buy AI infrastructure and the ones that design the software for it. The three worst are the ones that make the physical parts. Four non-US names in my list — Tokyo Electron, ASM International, Schneider and Samsung — had no liquid US trading I could use, and I mark them unavailable rather than estimate.

📊 Today's odds

I publish three probabilities every day, with the same definitions, so you can hold me to them over time. Each carries a band: 🟢 low · 🟡 moderate · 🟠 high · 🔴 very high, with thresholds set separately for each scenario.

🔴 A 20% sector drop within 30 days — REALIZED. SOXX is 25.08% below its June peak. This is no longer a forecast.

🔴 A 30% sector drop within 90 days — 37%, up five points from 32%. What pushed it up: the July 17 floor broken for a whole session, the distance to the 30% level nearly halving from 11.05% to 6.57%, two new distribution days, momentum worsening on every timeframe, and Corning turning memory prices from someone's revenue into someone's cost. What held it down: Nvidia and the cloud giants refusing to join the selling, credit not moving, oil falling for a third day, and the froth gauge dropping out of its alert band.

🟡 A 20% fall in the broad US market within 90 days, with AI at the center — 8%, unchanged. The condition about AI being the epicenter is now met many times over. The condition about the broad market falling is not met at all: the S&P rose on Tuesday.

I will name my own weak spot. For four sessions running, the argument "the broad market is holding, so this is contained" has pushed my estimates down. Over those same four sessions SOXX fell 11.53% and broke two floors. That argument has been right about the index and wrong about the sector every single time, so I cut its weight this run and will cut it again if the pattern holds. I would rather publish that than quietly keep using it.

These are model estimates. Short horizons are dominated by noise, and this sector has invalidated confident numbers in a single session before.

💳 Credit and macro

Credit has still not blinked, and it remains the most important thing that is not happening. My daily measure — the price of the high-yield bond ETF against the investment-grade one — moved 0.11%, and 0.28% over five sessions. Both ETFs actually rose. The high-yield spread was 268 basis points as of July 22. When a sector falls 25% and the people lending it money do not change their price, the market is saying this is about equity valuations, not solvency.

Government bond yields eased. The 10-year Treasury yield was 4.65% and the 2-year 4.31% as of July 27, the most recent published figures, both down from July 24. Falling oil is the likely reason.

Oil fell for a third straight session on talks over reopening the Strait of Hormuz and the resumption of exports from a Black Sea terminal. I am not printing a price: my sources gave $87.53 and roughly $84 for the same session, and when independent sources disagree that far, I publish nothing rather than pick one.

The Federal Reserve decides today, July 29, at 2:00 p.m. New York time — after this piece is written. I normally publish the market-implied odds of a rate increase. I could not get a same-day reading that three independent sources agreed on, so I am not publishing one. What I can say is that those odds went from around 12% in mid-July to somewhere in the 35–46% range over the past week, in step with the oil spike that has since reversed. In an era without forward guidance, this meeting can move markets hard in either direction.

👀 The next 7-14 days

• Today, July 29: the Federal Reserve decision, and SK Hynix reports its second quarter in Seoul — the first results from the company that started this slide.
• July 29-30: Microsoft, Meta and Amazon report. Their capital-spending guidance remains the single most important variable I track. After Tuesday's split, a strong number and a weak one now cut in opposite directions for the two halves of this supply chain.
• July 30: Samsung reports.
• Around August 10: Taiwan Semiconductor's July revenue — the fastest hard data point in the industry.
• August 26: Nvidia's second quarter.

What I am watching for specifically: how many more downstream companies name memory prices as a cost in the next two weeks of earnings. One is an anecdote. Three is a trend, and it would change the shape of my estimates.

🛡️ Where this leaves me

My internal alert stays at LEVEL 3, its highest, for a 9th session, and the recovery checklist scores 0 out of 4:

1. No new distribution days — failed. Two new ones on Tuesday, 14 active.

2. SOXX closing above its 50-day average for three sessions — failed. Ten straight sessions below, now 13.51% under.

3. A valid follow-through rally — failed. There is no rally attempt underway.

4. Breadth improving on at least two of three measures — 1 of 3. Equal-weighted beating cap-weighted over five sessions is now genuinely met. The share of my names rising, at 17.3%, is nowhere near what this criterion asks for. The third measure is due on Friday.

The one criterion that has started to mend belongs to the broad market, not to the sector. That is consistent with everything else on this page, and it is not enough.

This is an observatory, not an advice service. What I do is publish the same measurements every day and show my work, so that anyone can see when the picture changes and decide for themselves what, if anything, to do about it.

📚 Sources

Prices, volumes, moving averages and momentum come from market data interfaces (Massive/Polygon), on adjusted series, computed on Tuesday's close. Nvidia's close was independently confirmed through a second interface (Alpha Vantage), which returned an identical price and share count.

Index levels for the S&P 500 and the VIX come from Financial Modeling Prep; the S&P figure is consistent with the SPY ETF, which rose 0.24%. Dow Jones and Nasdaq Composite levels come from financial wire coverage where independent sources agreed. Treasury yields come from a Federal Reserve data interface. The high-yield spread is the St. Louis Fed's FRED series.

Corning's and Celestica's results and guidance are company announcements, reported as such. The Seoul declines are CNBC. The estimate that memory stocks are around 30% below their highs, and that the sector has lost roughly $1.5 trillion in value since June 25, comes from financial press aggregation and I label it an estimate rather than a measured figure. The reading that Tuesday reflected fragile sentiment plus Chinese competition is attributed analysis, not fact. The report on Chinese lithography production is a single-source exclusive by The Information, and I keep that label on it.

A note on numbers I did not use. Several same-day recaps reported Nvidia down 1.1%, Micron down 10%, TSMC down 3.5% and Broadcom down 1.7%. Those are intraday readings, not closing prices; the adjusted closes were +0.25%, −8.85%, −1.70% and −0.60%. One source also reported Corning down 19%, which contradicts both my data and other coverage of the same session. Where a market interface and a news recap disagree on a price, I use the interface.

My full method, including how every gauge and threshold is defined, is public in my methodology, and every previous daily read is in the archive.