Thursday's selloff in US megacap equities was not an opening gap that stuck. Six of the ten largest names opened above Wednesday's close on the view that a Democratic midterm sweep and tighter data-center permitting would slow artificial-intelligence capital spending and relieve pressure on hyperscaler cash flow. By the bell, all ten had closed red, and the heaviest damage landed on the companies that sell the silicon rather than the ones that buy it.
The Data Point
Advanced Micro Devices closed at $424.36, down $64.43 or 13.18%, its fifth consecutive red close and a decline of 25.7% over five sessions. The stock gapped down 7.41% at the open and then lost a further $28.22 during cash hours, finishing within the middle of a $403.92–$456.05 range but exactly at its 90-day low. That places AMD 35.5% below its 90-day high of $657.93 and well inside a 52-week band of $188.22–$658.52. My October 28 call that AMD would close below $538.72 confirmed at $505.44; the name has since fallen another $81.08 per share, and the de-rating is now deeper than that thesis contemplated.
| Symbol | Close | Change | Day range | 52-week range |
|---|---|---|---|---|
| AAPL | $346.59 | -2.92 (-0.84%) | $345.49–$355.38 | $243.42–$355.38 |
| MSFT | $541.88 | -13.92 (-2.50%) | $532.68–$557.61 | $349.20–$559.15 |
| NVDA | $186.63 | -13.50 (-6.75%) | $181.58–$196.23 | $164.27–$243.37 |
| GOOGL | $283.54 | -5.39 (-1.87%) | $277.14–$295.59 | $270.70–$408.61 |
| AMZN | $258.30 | -7.74 (-2.91%) | $252.35–$273.19 | $196.00–$287.20 |
| META | $602.76 | -27.25 (-4.32%) | $583.83–$634.87 | $520.26–$779.82 |
| TSLA | $349.42 | -35.77 (-9.29%) | $340.58–$374.29 | $297.38–$498.83 |
| JPM | $339.81 | -12.30 (-3.49%) | $335.97–$359.67 | $279.10–$366.50 |
| V | $430.41 | -6.16 (-1.41%) | $429.53–$442.12 | $293.89–$442.12 |
| AMD | $424.36 | -64.43 (-13.18%) | $403.92–$456.05 | $188.22–$658.52 |
The policy read is what gives the move its weight. Korean and Japanese equities fell on the same AI and chip rout even as the broader indexes in both markets held up, and CNBC noted that Wall Street's volatility gauge moved more than 11 points from intraday low to high — the largest one-day swing since April 9. Regulation that constrains data-center buildout does not cancel demand for accelerated compute; it changes the timing and the location of where that demand can be booked. For a supplier, timing risk is revenue risk.
The Numbers
Take one share of each of the ten names. The basket lost $189.38 against a combined prior close of $3,953.05, a decline of 4.79%. Of that, $166.14 — 87.7% — accrued between the opening print and the close, which is the opposite of a gap-and-stabilize session. AMD, TSLA and NVDA together supplied $113.70, or 60% of the total loss, from three positions out of ten.
Six of the ten opened above the prior close. All ten finished below it, and not one closed in the upper third of its session range.
NVIDIA at $186.63 is the cleaner instrument for reading the complex. The close marks a second straight red session, a 12.6% five-session decline, and the low end of its 90-day range — 22.9% below the 90-day high. In contrast, the stock still sits 13.6% above its 52-week low of $164.27, so the drawdown has unwound the autumn advance without yet threatening the annual structure. JPMorgan deserves a separate note: I argued in late October that a close below $330.45 would broaden the de-rating into financials, and that level never printed. What changed is that banks rallied as the AI trade funded rotation out of it. Thursday broke that pattern — JPM gapped up $6.53 and closed $18.83 below its open at $339.81, down 3.49%. When the safe haven inside the index stops working, the selling is no longer sector-specific.
The Counterargument
The bull case is not frivolous. Nothing in Thursday's session was an earnings event, a guidance withdrawal, or a cancelled order; it was a repricing of a policy probability. Slower permitting plausibly reduces the oversupply risk that has been the central bear argument against the buildout, which is why the morning narrative lifted the hyperscalers in the first place. Visa closed at $430.41, down only 1.41% and still within 0.5% of its 90-day high after a 17.8% advance over thirty sessions — evidence that the damage was concentrated in AI-levered names rather than distributed across the market. CNBC also reported institutional dip-buying into a bank position on the decline, which suggests appetite rather than capitulation.
The weakness in that case is that the relief was supposed to show up somewhere. If slower capex genuinely improves hyperscaler free cash flow, Amazon and Alphabet should have held their 2.11% and 1.74% opening gaps. Both gave them back entirely, and Meta closed at $602.76, down 21.4% over thirty sessions. A thesis that benefits no one on the day it is priced is a thesis the market does not believe.
Verdict
Fair. AMD at $424.36 has surrendered $233.57 per share from its 90-day high, and a recovery merely to the midpoint of that range, $541.15, would require a 27.5% advance. That is the arithmetic of a stock that has been de-rated rather than one that has been discounted — the premium is gone, but the valuation does not yet compensate for an order book exposed to permitting timelines. The thesis is most directly tested at AMD $403.92: a close below Thursday's low within five sessions confirms the compute complex is still unwinding, while a close above the session high of $456.05 within ten sessions would mark Thursday as an overshoot and reverse this view. Recovery in NVIDIA above $196.23 on the same horizon would do the same, and it is the signal I would weight more heavily, because the supplier with the broader customer base is the better read on whether demand or policy is doing the repricing.