AI MARKET OBSERVATORY · Blog
Wednesday, July 29, 2026 CRITICAL · Level 3 10 min read

Vertiv beat, raised guidance, and fell 17%

SK Hynix posted record profits and missed. Vertiv raised its full-year outlook and lost 17%. Good news has stopped working for AI stocks.

The risk read for this session

Sector crash
REALISED
chips −20% from peak · within 1 month
Unchanged from the previous read
Deeper crash
88%
chips −30% from peak · within 3 months
Up 51 points from 37%
Market crisis
14%
S&P 500 −20% · within 3 months, AI at the centre
Up 6 points from 8%
SOXX
465.00
−29.1% from peak
NVDA
190.01
−19.7% from peak
VIX
20.66
fear gauge
High-yield spread
268 bp
credit stress

These are model estimates, not forecasts anyone can guarantee, and not investment advice. The dot before each probability is its band: 🟢 low · 🟡 moderate · 🟠 high · 🔴 very high, with thresholds set separately for each scenario. SOXX is the main US semiconductor ETF, the basket I track; from peak is the fall from the highest close of this cycle; VIX is the market’s fear gauge; the high-yield spread is the extra interest riskier companies pay to borrow, my early-warning line for funding stress. My internal alert is at Level 3, on a scale where 3 is the highest.

📌 The one-line takeaway

On Wednesday, July 29, SK Hynix reported the largest revenue and the largest operating profit in its history, with a 76% operating margin — and the memory sector kept falling. Vertiv beat on earnings and raised its full-year guidance above what analysts expected — and lost 17.26% in a single session. Corning did something similar the day before. When a market sells companies that are doing well, it is not repricing the news. It is repricing what it is willing to pay for the news.

Today I moved my risk regime to 🔴 CRITICAL, the top of my scale, for the first time since I began publishing. My internal alert stays at LEVEL 3, the highest, for a 10th session; it only steps down after a four-part recovery checklist is met, and I score 0 out of 4.

Three numbers are deliberately missing below, and I say so where they belong: the size of Wednesday's fall on the Korean market, the closing level of long-dated US bond yields, and SK Hynix's net profit. None of them cleared my three-source rule, so none of them gets published.

📰 What happened

SOXX — the main US semiconductor ETF, a basket of the large chip stocks and the benchmark I track every day — fell 5.38% to 465.00. Volume was 1.76 times the average of the last 50 sessions, and the close came in 0.20% above the day's low: there was no attempt at a recovery into the bell.

That takes SOXX to 29.11% below its June 22 peak of 655.95. A 30% fall from that peak means a price of 459.17. Wednesday's close was 1.25% away from it. For context, the four sessions before Wednesday moved SOXX by −4.40%, −2.05%, −4.80% and −5.38%. The remaining distance is smaller than any one of them.

Nvidia, the sector's flagship, fell 3.55% to 190.01 and closed exactly on its low of the day. It also closed below its 200-day moving average — the average price of the last 200 trading days, the slow line that separates a correction from a change of trend — for the first time in this decline, by 1.54%. Nvidia is now 19.67% below its own peak; a 20% fall is 0.41% away.

The selling was not confined to chips. The Dow Jones Industrial Average fell 1,153 points, or 2.19% — its worst session since April 2025. The S&P 500 lost 1.52% and the Nasdaq Composite 1.74%. In Korea, the exchange triggered a trading halt on its main index for the second consecutive session; the size of that fall is reported inconsistently across sources, so I am not publishing a figure for it.

The tell: three companies, three different quarters, one outcome

This is the part that changed on Wednesday, and it is worth being precise about it.

SK Hynix: a record that wasn't enough

The Korean memory maker reported second-quarter revenue of 79.32 trillion won and operating profit of 60.54 trillion won — an operating margin of 76%, with revenue up 257% and operating profit up 557% from a year earlier. Both figures are records. Both came in below what analysts expected (roughly 84 trillion and 64 trillion). The stock and the sector fell anyway.

Vertiv: a beat, an upgrade, and a 17% loss

Vertiv makes the power and cooling equipment that data centers run on. It reported adjusted earnings of $1.52 per share against expectations of $1.42, and raised its full-year outlook to $6.65–6.75 in earnings per share (from $6.30–6.40) and $13.8–14.2 billion in revenue (from $13.5–14.0 billion) — both above consensus. Quarterly revenue came in at $3.27 billion, up 18% organically, against expectations of $3.37 billion; the company attributed the shortfall to temporary supply-chain congestion and to the timing of larger, multi-phase AI projects.

The stock fell 17.26% to 223.04, its worst day in more than fifteen months.

Why this matters more than the percentages

A rotation is when money leaves one sector because it prefers another. A de-rating is when the same earnings are worth less than they were last month. On Wednesday the earnings were, on balance, good — and the prices fell. That is the second explanation, not the first.

🚦 The signal board — my 9 gauges in 6 families

I track nine gauges grouped into six families. Each one is either 🟢 calm, 🟡 watch, 🟠 alarm or 🔴 confirmed breakdown.

FamilyGaugeReading
A — Price trendMoving averagesAlertSOXX below its 50-day average for an 11th session, 18.04% under it. Nvidia closed below its 200-day average for the first time.
A — Price trendStructureCriticalThe 480.50 floor from Tuesday's low broke on the close, by 3.23%. New reference: 464.08.
A — Price trendMomentumAlertWorse on every timescale. The standard daily strength reading on SOXX fell to 33.15 out of 100; the weekly one fell below 50 for the first time in this decline.
B — ParticipationBreadthAlertOnly 4 of the 52 US names I track rose. Broader-market breadth, by contrast, improved for a fifth session.
B — RotationRotation and canariesAlertChips still the epicenter, but the destination of the rotation disappeared: the broad market fell too. CoreWeave hit a 52-week low — the first of the early-warning names I watch to do so.
C — VolumeDistributionCriticalThe break happened on heavy volume, and two more distribution days were added.
D — VolatilityFear gaugeWatchThe VIX rose 13.45% to 20.66, breaking above 20 for the first time in this decline.
E — CreditCredit tapsCalmHigh-yield credit did not deteriorate. The extra yield demanded from weaker borrowers stood at 268 basis points, unchanged.
F — FrothOverextensionWatchSOXX now sits 16.08% above its 200-day average, down from 40.64% on July 22.

The board says one thing clearly: the damage is in price, volume and structure, and it is still not in credit. The market that lends money to fragile companies has not moved. That is the single most important thing still holding.

📏 The levels that matter now

  • SOXX: closed at 465.00, which is 29.11% below the June 22 peak of 655.95. The 30% mark sits at 459.17 — 1.25% below.
  • SOXX vs its averages: 18.04% below the 50-day average (567.33); 16.08% above the 200-day average (400.59). That second cushion is what still separates this from a structural break of trend, and it has shrunk from 40.64% in five sessions.
  • Nvidia: closed at 190.01, 19.67% below its May peak of 236.54. A 20% fall means 189.23 — 0.41% below.
  • S&P 500: closed at 7,316.15, 3.86% below its June 2 record close of 7,609.77.

The ratio between the two drawdowns — chips down 29.11%, the broad index down 3.86% — is 7.5 to 1. AI is unambiguously the epicenter of this decline.

One historical marker, offered as context and not as a forecast. In the dot-com collapse of 1999–2002, the Nasdaq reached a 20% fall on the 23rd trading day after its peak and a 30% fall on the 25th. SOXX reached 20% down on the 24th trading day after its June 22 peak, and on the 26th it stands 29.11% down. The pace is almost the same. That says nothing about where this ends — the 2000 decline ran for another two and a half years, and plenty of declines that started this way simply stopped. It does say the speed is not unprecedented.

🏭 The sector, group by group

The average move across the 52 US names I follow was −4.62%, and only four of them rose.

  • Buyers of AI infrastructure held up best. The large cloud and platform companies fell 1.19% on average, with Alphabet the only one higher (+0.90%).
  • Sellers of AI infrastructure were hit hardest. Power and data-center equipment fell 6.56% on average, dragged by Vertiv's 17.26% loss; the speculative cloud names fell 7.41%, with IREN −13.62%, Applied Digital −12.77% and Nebius −12.65%.
  • Memory actually stabilized (−3.58% against −9.50% the day before): Seagate rose 2.29%, Western Digital was nearly flat. The pressure moved from the price of memory to the people installing the racks.
  • Equipment and servers fell 5.61% and 5.87%, with KLA −10.80% and Super Micro −9.67%.

The shape has been the same for four sessions: whoever buys AI capacity loses a little, whoever builds and installs it loses a lot.

📊 Today's odds

These are model estimates of specific, defined events. The colored dot in front of each is my band: 🟢 low · 🟡 moderate · 🟠 high · 🔴 very high.

  • 🔴 A 20% fall in chip stocks within 30 days: REALIZED, and deeper. SOXX is 29.11% below its peak, against 25.08% at yesterday's reading.
  • 🔴 A 30% fall in chip stocks within 90 days: 88%, up from 37% yesterday. The move is almost entirely arithmetic rather than judgment: in one session the remaining distance to that level collapsed from 6.57% to 1.25%, and the price closed on its low.
  • 🟠 A 20% fall in the broad US market within 90 days, with AI at the center: 14%, up from 8% yesterday. The S&P 500 would need to fall a further 16.79% from Wednesday's close.

A 51-point move in a single day is unusual and deserves an explanation rather than a shrug. Here it is: nothing in my judgment changed by 51 points. The price moved to within one ordinary session of the line, and the estimate followed it there.

One argument left my model today. For four consecutive readings I had reduced my risk estimate partly because the broad market was absorbing the chip decline. On Wednesday it stopped absorbing it. I have removed that argument rather than quietly shrinking it, and I am recording that it was wrong for four sessions running.

💳 Credit and macro

The Federal Reserve became a second front. At 2:00 p.m. Eastern the Fed left its policy rate unchanged at 3.50–3.75% on a 9–3 vote. The three dissenters — Beth Hammack, Neel Kashkari and Lorie Logan — did not want a cut. They wanted a quarter-point increase. The statement says inflation "remains elevated relative to the Committee's 2 percent goal, in part reflecting supply shocks that have driven price increases in certain sectors, including energy," and that the Committee "will deliver price stability."

Sector declines are usually cushioned by the expectation that money will get cheaper. On Wednesday that expectation went the other way. Press reports describe long-dated bond yields reaching their highest in nearly two decades in response; my own source for those yields runs a day behind, so I am not publishing a closing figure for Wednesday.

Credit itself has still not moved. The extra yield that weaker companies pay to borrow — the clearest early warning of a real financial squeeze — stood at 268 basis points, or 2.68 percentage points, at its most recent reading, and my daily read on riskier corporate debt improved slightly rather than worsening. Nothing in the debt market is confirming what the equity market is doing.

👀 The next 7-14 days

  • After Wednesday's close, Microsoft reported revenue of $90.01 billion and adjusted earnings of $4.74 per share, both above expectations, and guided Azure to roughly 45% growth. It also extended the assumed useful life of its data center buildings and offices from 15 to 25 years starting in fiscal 2027 — worth flagging, but worth flagging accurately: chips and servers stay on a six-year schedule, and roughly two-thirds of last quarter's spending was on those. Separately, about $15 billion of the reduction in Microsoft's stated 2026 capital spending comes from reclassifying leases, not from spending less. Meta reported revenue up 28% but missed on earnings and fell more than 7% after hours.
  • Thursday, July 30: Amazon reports after the close, and Samsung reports in Seoul. Amazon completes the picture on 2026 AI capital spending.
  • Around August 10: TSMC's July revenue, the most timely hard number in the supply chain.
  • August 26: Nvidia's quarterly results.
  • Watch item: whether more companies downstream of the chip industry describe memory prices as a cost in the next two weeks. One is an anecdote. Three would change the shape of estimates.

🛡️ Where this leaves me

My regime is 🔴 CRITICAL and my internal alert is at LEVEL 3, the highest, for a 10th session. That reflects what the price, the volume and the structure are doing, and nothing about what anyone should buy or sell — I publish an observatory, not recommendations, and none of this is investment advice.

It steps down only when a four-part recovery checklist is satisfied. I score 0 out of 4, and the single component that had turned positive belongs to the broad market rather than to chips — and the broad market fell on Wednesday. In plain terms: nothing in this decline has yet stopped, and the one thing that had started to look better is the thing that broke.

The counterweight, stated as plainly: final demand still looks intact. Vertiv raised its outlook, SK Hynix earned a 76% operating margin, Microsoft beat and raised its spending because it says customers want the capacity. What is being repriced is the multiple, not the demand. Those are different problems, and only one of them is visible in a share price today.

📚 Sources

↑