Weekly Report: The AI capex consensus was wrong last week — the scorecard points to one crowding risk

Three consensus calls went into last week. Two of them were wrong, and the way they were wrong is more useful than the loss itself: the Philadelphia Semiconductor Index fell 12.15%, Micron dropped 17.46% after beating estimates, and the Dow lost only 0.94% against the Nasdaq's 3.37%. The scorecard says the mistake was not about AI demand. It was about how crowded the expression of that demand had become.

Four misses out of five is not a forecasting failure — it is what happens when every call is a different version of the same call. All four were expressions of "AI capex is real." That statement was true. The problem was that the market had already paid for it.

Micron's quarter was not the issue. The company beat, and the FY27 EPS consensus near $160 was left intact. The stock fell 17.46% because the shares had run far enough that a 2028 factory delay — a plant that makes no chips until 2028 — was enough to reset the multiple. That is a positioning event, and it is only visible in the correlation data.

The 60-day correlation between SOX and the S&P 500 is 0.729, and between the Nasdaq and the S&P 500 it is 0.954. Those are not independent exposures. When the AI trade unwinds, the index-level move understates the sector-level move by a factor of four: the S&P fell 2.68% while semis fell 12.15%. The scorecard's real lesson is that the "AI capex" consensus was being expressed through a single crowded channel, and the channel broke before the thesis did.

The Dow's -0.94% is the counter-evidence that matters. If this were a growth scare, the Dow would have fallen with the Nasdaq. It did not, because the money leaving semis went into banks, industrials and defensives — a rotation, not an exit.

The rotation is visible in the sector tape, not just the index. Banks and industrials were the buyers' destination: NAB fell 6.19% on its own result, but the broader financials held, and the equal-weight S&P outperformed the cap-weighted index for a second consecutive week. That is the signature of money leaving a concentrated theme rather than leaving equities. The Dow's 0.94% decline against the Nasdaq's 3.37% is a 2.4 percentage-point spread — the widest since April 2025 — and it arrived without a recession signal.

The earnings detail points the same way. Micron beat and fell 17.46%; AMD fell 17.3%; Apple beat and still slipped. Three beats, three declines, across three different links of the AI supply chain. When beats stop working, the market is not pricing the quarter — it is pricing the next multiple. That is a crowding signal, and it usually takes more than a week to clear.

Four calls are now in the price:

1. CPI core at +0.2% or lower — December hike odds fall below 70% and the Nasdaq reclaims 25,000.

2. AI capex is intact — CoreWeave's $3.54bn consensus and $35–39bn FY26 capex guide confirm it on Tuesday.

3. Semis have bottomed — Friday's +4.20% holds 9,390 as support.

4. Volatility has peaked — VIX retreats below 22 as the CPI passes.

Number three. Friday's 4.20% bounce is the only one of the four that has no data behind it yet — it is a price move after a price move. CoreWeave reports Tuesday after the close, Cisco and Applied Materials on Thursday. Until those three print, the "semis have bottomed" call is a chart pattern, not a fundamental view.

The fragility is measurable. SOX closed at 9,390.25, roughly 12.2% below its Oct 29 close of 11,518.20. A retest of Friday's low near 9,221 would put the index back below the bounce and confirm that the +4.20% was a lower high. A close above 10,000 would confirm the bottom. Everything between is noise.

Three earnings dates carry the argument. CoreWeave reports Tuesday after the close with a $3.54bn revenue consensus and a $35–39bn FY26 capex guide; Cisco reports Thursday with $9.3bn of FY26 AI networking orders behind it; Applied Materials reports the same day with the wafer-fab-equipment guide that sets the 2027 memory capex path. If two of the three cut guidance, the crowding risk becomes a demand question — and the "AI capex is intact" consensus fails on fundamentals rather than positioning.

Watch the SOX–S&P spread on Tuesday afternoon. If CPI is soft and semis still underperform the index, the rotation is deeper than one week and the "AI capex" trade is being funded, not bought. If semis outperform on a soft print, Friday's bounce was the start of the repair, and the crowding risk has moved to the next earnings cycle.

Either way, the level is 9,390. Friday's bounce either becomes the floor or it becomes the lower high that defines the November range.