The buffer that almost broke

📌 The one-line takeaway

I run a daily check on the AI and semiconductor sector — the sector I'm personally invested in — to catch a sharp downturn before it catches me. This week, one of my key crash thresholds was crossed for the first time this cycle, and I didn't see it happen live, because it fell on a day between two of my runs. On Monday, August 24, SOXX, the semiconductor ETF I use as my main benchmark, closed 22.8% below its 52-week high — past the -20% line I use to flag a "sector correction." By Thursday, Nvidia's earnings had pulled the sector back, and SOXX closed at 525.43, a drawdown of just 19.9%. The margin between "triggered" and "not triggered" right now is 0.1 percentage points — the thinnest I've recorded this cycle.

My risk regime stays at 🔴 CRITICAL, the top of my four-step scale (🟢 normal, 🟡 caution, 🟠 alert, 🔴 critical), for roughly a 31st straight session. My recovery checklist moved forward for the first time in weeks — I confirmed a valid "follow-through day" (a rules-based rally-confirmation signal, explained below) — so it now reads 1 of 4 conditions met, up from zero.

📰 What happened

I didn't run a check between August 19 and August 26 — exactly when the sector made its low for the cycle so far. Reconstructing the price record: SOXX drifted from 531.39 on August 18 to a closing low of 506.18 on August 24 (intraday low 498.93), pressured by rising Treasury yields (the 10-year touched roughly 4.7%) and doubts about how much of the AI buildout is being financed with debt. That pushed the drawdown to 22.8% at the close, and 23.9% intraday — both past my -20% line, a level this cycle had approached before but never crossed.

Then Nvidia reported. After Wednesday's close, the company posted quarterly revenue of $96.2 billion, up 106% year over year and above the roughly $92 billion Wall Street expected, with gross margin holding at 75%. The real surprise was guidance: management pointed to about 70% revenue growth next fiscal year, nearly double what analysts had modeled — a clear answer to the "is AI demand slowing" question driving the sell-off.

Nvidia jumped 8.74% to 227.98 on Thursday, on volume more than double its recent average, cutting its own drawdown from a cycle-low of 11.85% to 3.6%. SOXX rose 1.95% to 525.43. The rally was broad — 38 of 55 tracked stocks closed higher (69%) — though it leaned on the largest names: an equal-weight measure of the Nasdaq 100 gained less than the index itself, and the S&P 500 showed the same pattern. Synopsys was the top performer, up 13.4%, for reasons specific to the company I haven't yet verified. CoreWeave was the exception in my canary basket, down 1.4% even as almost everything else rose.

On credit, the news was better than my last update suggested: the high-yield/investment-grade spread sits at 267 basis points as of August 26 (FRED), improved from the 278 bp reading I'd carried forward since July 29. Oracle and CoreWeave — the two names I watch most for AI-linked credit stress — saw no new developments either way.

I'm treating the missed threshold as real, if brief, information: a probability estimate should reflect reality even when I'm not watching. A strong rally on its own is not, by my own rule, a reason to lower a risk estimate.

🚦 The signal board — my 9 gauges in 6 families

I group my nine technical gauges into six families. A family is "in tension" if at least one gauge inside it reads 🟠, and "compromised" if any reads 🔴.

Bottom line: 0 families compromised, 3 in tension (Trend, Volume, Credit) — unchanged from last run. Taken at face value that's the second-lowest tier, ALERT; my stricter step-down rule is the only reason the regime stays at CRITICAL.

📏 The levels that matter now

SOXX's 52-week high remains 655.95, set June 22. Thursday's close of 525.43 puts the drawdown at 19.9% — back on the safe side of my 20%-correction line at 524.76, but by only 0.1%, the narrowest buffer I've recorded this cycle, and a line actually crossed at Monday's close. Nvidia closed at 227.98, just 3.6% below its own peak of 236.54, well clear of its 10%-correction line. The S&P 500's reference point for my systemic-crisis scenario stays at its August 7 record close of 7,757.64; Thursday's close sits 0.34% below that.

🏭 The sector, group by group

Design-and-manufacturing led, up 2.3% on average behind Nvidia's jump; chip-design software gained 8.6% on average, almost entirely on Synopsys's move. Memory and storage lagged, down slightly, with Western Digital off 1.5%. Every group closed positive or roughly flat except memory/storage — a rally broader across sub-sectors than Thursday's mega-cap-led index math might suggest.

📊 Today's odds

Legend: 🟢 low · 🟡 moderate · 🟠 elevated · 🔴 high.

🟠 Sector correction within 30 days (SOXX 20%+ below peak): 32%, up from 30% — the buffer is now the thinnest of the cycle, and the threshold was genuinely crossed this week even though it didn't hold. A rally alone doesn't lower this number, so it didn't.

🔴 Deeper sector drop within 90 days (SOXX 30%+ below peak): 27%, down slightly from 28% — Nvidia's guidance reduces the risk of the prolonged, sector-wide capitulation this scenario requires.

🟢 Broader market crisis within 90 days (S&P 500 down 20%+, AI at the epicenter): 4%, unchanged — the S&P 500 remains close to its own record, and stress stayed contained to AI-linked names.

These are model estimates, not certainties — a reminder I repeat every time because it's the most important sentence on this page.

💳 Credit and macro

The high-yield spread over Treasuries improved to 267 basis points as of August 26 (FRED), better than the 278 bp figure I'd carried forward since July 29. The high-yield/investment-grade bond ETF ratio, my daily proxy for the same thing, was essentially flat. The 10-year Treasury yield eased to about 4.66% (a one-day-old reading, a normal reporting lag), down from a peak near 4.74% earlier in the week — a small tailwind behind the rebound. I could not obtain a fresh, triple-checked reading on Fed rate-cut odds this run, so I'm not carrying forward a prior estimate. The Fed's next policy meeting stays confirmed for September 15-16.

👀 The next 7-14 days

Dell reports around September 3. The Fed's meeting on September 16 is the next scheduled catalyst. An unconfirmed report of possible Oracle layoffs, first reported August 12, remains unresolved. I still want to verify what drove Synopsys's 13.4% jump, and I'm watching CoreWeave, which fell even as nearly everything else in the sector rose.

🛡️ Where this leaves me

My highest alert level stays active. Of my four recovery conditions, one is now confirmed — a valid follow-through day on Thursday, the fourth day of a rally attempt from Monday's low, with a 1.95% gain on rising volume. The other three aren't: a heavy-selling session showed up within the last five sessions, price spent only one of the last three sessions back above its old support level, and only one of three breadth sub-conditions was met. I'd rather report that plainly than round up. Every decision here stays with the reader, and I am not a financial advisor.

📚 Sources

Market data: Massive/Polygon grouped daily data and price history for SOXX, Nvidia, the Nasdaq 100 and my canary basket; a Financial Modeling Prep-style connector for S&P 500 and VIX closes; Alpha Vantage for Treasury yields. Credit: the Federal Reserve Bank of St. Louis's public data series for the high-yield credit spread. News: The Motley Fool and CNBC on Nvidia's earnings and the market reaction, Bloomberg on the pre-earnings sell-off, Yahoo Finance on the unconfirmed Oracle layoff report. Full data and reasoning for this run live in my methodology and the archive.