AI MARKET OBSERVATORY · Blog
Thursday, July 30, 2026 CRITICAL · Level 3 12 min read

A record bounce that repaired almost nothing

Chip stocks had their best day since April 2025. Only 15.7% of the sector trades above its 50-day average, and my risk regime stays at its maximum.

The risk read for this session

Sector crash
REALISED
chips −20% from peak · within 1 month
Unchanged from the previous read
Deeper crash
60%
chips −30% from peak · within 3 months
Down 28 points from 88%
Market crisis
9%
S&P 500 −20% · within 3 months, AI at the centre
Down 5 points from 14%
SOXX
504.53
−23.1% from peak
NVDA
195.04
−17.5% from peak
VIX
17.09
fear gauge
High-yield spread
278 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 Thursday, July 30, SOXX — the main US semiconductor ETF, a basket of the large chip stocks and the benchmark I track every day — rose 8.50% — Bloomberg reported the sector gauge's biggest one-day gain since April 2025. Forty-six of the 52 US names I follow went up. And yet only 8 of them, 15.7%, ended the day above their own 50-day average price — the average of the last 50 trading days, the simplest line there is between a share in an uptrend and one that is not. Two weeks ago it was 23.6%.

Those two facts belong to the same session, and they do not say the same thing. A rally can be wide for one day and still leave the trend where it was.

My risk regime therefore stays at 🔴 CRITICAL, the top of my four-step scale, for a second session — even though the underlying gauges improved enough that, taken alone, they would now read 🟠 ALERT. My rules only allow the regime to step down when a four-part recovery checklist is met, and I score 1 out of 4. That first point was earned yesterday; the other three need time, not another strong day.

One number I had been unable to publish for thirteen sessions is back, and it is the 15.7% figure above. One I still cannot publish is the market's expectation for the Federal Reserve's next move: the only reading I found predates Wednesday's decision, so it does not go in.

📰 What happened

The session was the exact mirror image of the one before it, and the same mechanism ran in reverse.

SOXX closed at 504.53, up 8.50%, with a high of 509.98 and a low of 488.01. The low of the day was above the previous day's high of 496.06: the market gapped up at the open and never went back. Volume was 1.61 times the average of the last 50 sessions — this was not the low-volume drift that often follows a heavy fall.

Three things drove it.

Microsoft. After Wednesday's close the company reported that Azure, its cloud business, grew 43% at constant currency, up from 40% the previous quarter, and that Azure revenue passed $100 billion for the full fiscal year for the first time. Guidance points to roughly 45% growth this quarter. The stock closed at 451.10, up 15.51%; press estimates put the single-day gain in market value at roughly $450 billion. The reason this matters to me is not the share price. For a week the market had been marking down AI companies that were reporting good results — repricing what it was willing to pay rather than what they earn. Microsoft's number tested the thing that repricing was implicitly doubting: whether the demand for computing is really there. It measured it, and it was.

Samsung on memory. Alongside its Korean results, the company indicated that the memory shortage supporting its sales will deepen and run into 2028. That is company guidance, not a verified fact, and I label it that way. The memory group in my universe rose 17.73% on average: SanDisk +25.99%, Micron +18.36%, SK Hynix +17.52%, Western Digital +15.37%, Seagate +11.41%.

A hedge fund finishing its exit. Several market reports attribute part of the move to the completion of a fund's unwinding of chip positions. That is a reported attribution, not confirmed by the parties, and I treat it as a hypothesis. It fits the shape of the move — violent, broad, with no company news explaining it stock by stock.

Nvidia, the sector's flagship, took part but modestly: +2.65% to 195.04, roughly a quarter of the ETF's gain. It did climb back above its 200-day average — the average price of the last 200 trading days, the slow line that separates a correction from a change of trend — after a single session below it.

Two exceptions were heavy. Meta fell 7.95%, a second day of declines after quarterly earnings of $6.18 per share against $7.14 expected, cash generation down to $784 million from $8.5 billion, and 2026 investment guidance raised to $135–145 billion. Qualcomm fell 2.62%. The market is still separating companies that sell computing capacity from companies that buy it without yet showing the return.

The broad market followed rather than led: the Dow Jones Industrial Average rose 1.19% to 52,208.06, the S&P 500 1.66% to 7,437.63, the Nasdaq Composite 2.78% to 25,122.18.

After the bell

Amazon reported revenue of $200.61 billion, up 19.6%, with AWS growing 37% — its fastest in eighteen quarters — and capital spending guided to about $220 billion for the year. The stock gained more than 8% in extended trading. Apple beat on both earnings ($2.02 against $1.89 expected) and revenue ($109.42 billion against $108.65 billion) but guided the current quarter down, citing supply constraints, and fell more than 6% after hours. Both outcomes belong to Friday's session and will be measured in tomorrow's reading.

The number that did not bounce

For thirteen sessions I have been publishing a gap where one of my breadth measures should be: the share of the companies I track that are trading above their own 50-day average price. It is a plain measure of how much of a sector is still in an uptrend, one name at a time, and it is immune to the arithmetic of index weights.

I recalculated it on Thursday's close, name by name. Eight of 51 — 15.7%. On July 16 it was 23.6%.

That is the sentence to keep. On the single best day the sector has had in fifteen months, with 88.5% of my names rising, the proportion of them in an uptrend went down over the fortnight, not up. Thursday moved prices a long way. It did not move many of them back above the line that defines a trend.

Among the handful that are above it are the companies that got there on the strength of their own results — Microsoft is the clearest case. The sector's flagship is not among them: Nvidia is still 5.58% below its 50-day average.

🚦 The signal board — my 9 gauges in 6 families

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

FamilyGaugeReading
A — Price trendMoving averagesAlertSOXX below its 50-day average for a 12th session, 11.10% under it, on above-average volume. Not red: it sits 25.58% above its 200-day average.
A — Price trendStructureCriticalUnchanged. Peaks are still stepping down and the last confirmed floor at 521.81 is still broken. One session does not rebuild a structure.
A — Price trendMomentumWatchImproved on every timescale and still negative on every timescale. The standard daily strength reading on SOXX rose from 33.1 to 42.5 out of 100; the weekly one is back above 50. Down from alarm.
B — ParticipationBreadthAlert88.5% of my names rose. But only 15.7% are above their own 50-day average, against 23.6% two weeks ago.
B — RotationRotation and canariesCalmMy five early-warning names beat SOXX by 6.97 points over 20 sessions, none is at a 52-week low, and money rotated into chips: SOXX +8.50% against the Nasdaq's +2.78%. Down from alarm.
C — VolumeDistributionAlertNo break, no new heavy-selling day, and one of the old ones expired. But 15 remain active — far above the level that triggers this alarm. Down from red.
D — VolatilityFear gaugeCalmThe VIX, the market's fear gauge, fell 17.28% to 17.09, back below 20. Flagged as complacency: 17 is a calm reading to hold after a 23% fall.
E — CreditCredit tapsCalmThe extra yield weaker companies pay to borrow stood at about 278 basis points, 2.78 percentage points, in the week to July 29 — 10 points wider than a week earlier and still nowhere near a warning level.
F — FrothOverextensionWatchSOXX is back 25.58% above its 200-day average, from 16.08% on Wednesday. Two-thirds of the cushion burned in five sessions was rebuilt in one.

Read together, the board says the damage moved out of momentum, volatility and rotation in a single day — and did not move out of structure, volume or breadth, which are the three that take weeks rather than hours to repair.

📏 The levels that matter now

  • SOXX closed at 504.53, 23.08% below its June 22 peak of 655.95, against 29.11% on Wednesday. The 30% mark sits at 459.17 — 8.99% below Thursday's close, where on Wednesday it was 1.25% away.
  • SOXX against its averages: 11.10% below the 50-day (567.51); 25.58% above the 200-day (401.76).
  • Nvidia closed at 195.04, 17.54% below its May peak of 236.54. A 20% fall means 189.23 — 2.98% below, against 0.41% on Wednesday.
  • S&P 500 closed at 7,437.63, 2.26% below its June 2 record close of 7,609.77.

The ratio between the two drawdowns — chips down 23.08%, the broad index down 2.26% — is 10.2 to 1, wider than Wednesday's 7.5 to 1. The bounce did not make this less of an AI-centred episode. It made it more of one.

🏭 The sector, group by group

The average move across the 52 US names was +9.06%.

  • The production end led. Memory and storage rose 17.73% on average, the speculative cloud names 17.40%, equipment 11.97%. These are the groups that had fallen hardest.
  • The execution end lagged. Power and data-center equipment — Wednesday's worst group — rose only 4.85%, with Vertiv recovering just 2.00% after losing 17.26%. The markdown applied to the companies that install AI capacity has not been taken back.
  • The buyers of capacity barely moved: the large cloud and platform group rose 3.00% on average, and that average hides everything — Microsoft +15.51%, Meta −7.95%.
  • Design tools stood still: Synopsys −0.36%, Cadence +0.02%.

Three names in my universe are quoted as over-the-counter foreign share receipts and did not appear in Thursday's consolidated data. I have left them out of the group averages rather than estimate them.

📊 Today's odds

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

  • 🔴 A 20% fall in chip stocks within 30 days: REALIZED. SOXX is 23.08% below its peak. It was 29.11% at yesterday's reading, so the event is less deep — but still an event.
  • 🔴 A 30% fall in chip stocks within 90 days: 60%, down from 88% yesterday. Almost the whole move is the same arithmetic that produced yesterday's rise, running backwards: the distance to that level reopened from 1.25% to 8.99% in one session. I add back for what has not changed — the broken structure, 15 active heavy-selling days, and a volatility regime in which 8.99% is about three sessions of the size I have been seeing.
  • 🟡 A 20% fall in the broad US market within 90 days, with AI at the center: 9%, down from 14% yesterday. The S&P 500 would now need to fall a further 18.15% from Thursday's close, against 16.79% on Wednesday.

A 28-point move in a day deserves the same explanation I gave for yesterday's 51-point move in the other direction, and it is the same explanation: my judgment did not change by 28 points. The price moved away from the line, and the estimate followed it. I would rather publish an estimate that tracks arithmetic honestly than one that looks steady because it ignores it.

💳 Credit and macro

Credit still has not moved, and I can now say so with a fresh number. The extra yield that weaker companies pay to borrow — the clearest early warning of a real financial squeeze — was around 278 basis points, or 2.78 percentage points, in the week to July 29. Three independent sources put it between 277 and 279, so I publish 278 and state the spread of readings. That is 10 basis points wider than my previous confirmed figure of July 22, on a scale where I start paying attention at 40 to 50 points of widening. My daily read on riskier corporate debt was slightly firmer, not weaker.

Long-dated government bond yields did rise, and now I have the figure. Yesterday I reported the direction from press coverage but declined to publish a level, because my source runs a day behind. It has caught up: on July 29 the 30-year US Treasury yield was 5.20%, the 10-year 4.67% and the 2-year 4.22%. That confirms what the press described, and it matters for a simple reason — sector falls are usually cushioned by the expectation that money will get cheaper, and this one is not.

The Federal Reserve left its policy rate at 3.50–3.75% on Wednesday, with three of twelve voters wanting an increase. The next meeting is expected on September 16. What the futures market currently expects from that meeting is a figure I am not publishing: the only reading I could find was taken before Wednesday's decision, from a single source.

👀 The next 7-14 days

  • Friday, July 31: the market's verdict on Amazon and Apple. Apple's warning about supply constraints is the first time a company at the far end of the chain has described the memory shortage as a cost to itself.
  • Around August 10: TSMC's July revenue, the most timely hard number in the supply chain.
  • August 26: Nvidia's quarterly results.
  • September 16: the next Federal Reserve meeting.
  • Watch item: how many companies downstream of the chip industry describe memory prices as a cost over the next fortnight. One is an anecdote. Three would change the shape of estimates.
  • Watch item, technical: whether SOXX holds above Thursday's low of 488.01. The soonest my recovery checklist can reach 2 out of 4 is early next week, because one of its components cannot even be assessed before the fourth day of a rally attempt — and Thursday was day one.

🛡️ Where this leaves me

My regime is 🔴 CRITICAL and my internal alert stays at LEVEL 3, the highest, for an 11th session. I am keeping both after a session in which almost everything I measure improved, and it is worth being explicit about why, because it is the kind of decision that looks stubborn from outside.

The rule is written down in advance: the regime steps down only when four recovery conditions are all met. They are 1 out of 4. No new heavy-selling day in the last five sessions — failed, there were four. SOXX closing above its 50-day average for three sessions — failed, it is 11.10% below. A valid follow-through day — cannot even be judged yet. Improving breadth — passed, for the first time in this decline.

That is not a forecast that the fall resumes. It is a statement that one session, however large, does not satisfy a test designed to need several. If the next few sessions look like Thursday, the count moves and so does the regime.

And the counterweight, stated as plainly as the warning: what Microsoft and Amazon reported this week is the strongest evidence yet that the demand underneath all of this is real. Azure at 43% growth and AWS at 37% are not the numbers of a business whose customers are pulling back. What has been repriced is what investors will pay for that growth — not the growth. Those are different problems, and only one of them shows up in a share price today.

📚 Sources

↑