Why Didn't the Options Market Reprice the AI Semiconductor Rally? All Nine Tickers Show IV Below 60-Day Realized Volatility

Data basis (read this first)

• Options metrics and settled prices: the 2026-09-22 U.S. close. The options snapshot was collected in-house at 06:00 Beijing time on 2026-09-23, covering the 09-22 trading session across 289 tickers; nine are used here.
• The 2026-09-23 U.S. session was still trading at the time of writing. Every figure labelled "09-23 intraday" is a live quote taken at 10:31 ET / 22:31 Beijing time on 2026-09-23. These are not settled closing prices and are used only in the "First test" section, never in the core conclusions.
• Implied-volatility percentiles are computed against an in-house sample of n=98-127 trading days, not 252. They must not be read as "a one-year low."
• The ATM implied volatilities here correspond to near-term contracts with 4-8 days to expiry, not a 30-day constant-maturity measure, so they are not directly comparable to VIX-style indices.
• This article contains no position, sizing, or trading advice.

The short answer

On 2026-09-22 the AI semiconductor complex rallied across the board in the spot market (SanDisk +6.82%, Micron +5.00%, the semiconductor ETF +1.92%). Yet the options market priced all nine tickers, without exception, at an implied volatility below their own 60-day realized volatility. The market treated the calm of the past 20 sessions as the new baseline and did not reprice for a return to the past quarter's turbulence. The differentiation was not in the level of volatility — it was entirely in skew.

What happened: spot moved, volatility did not

IV = at-the-money implied volatility; HV20/HV60 = realized volatility over 20/60 trading days; skew = the relative price of downside protection (positive means puts are richer than calls). All figures annualized percentages.

Three testable facts

Fact 1: all nine tickers show IV below 60-day realized volatility

IV/HV60 ranges from 0.749 (SanDisk) to 0.961 (Broadcom), averaging 0.844 across the nine. The options market was pricing the next few days roughly 16% below the volatility these names had actually delivered over the prior quarter.

At the same time, seven of the nine show IV above HV20. Putting both together gives something precise:

• IV is bracketed between HV20 and HV60.
• Realized volatility over the past 20 sessions is lower than over the past 60 — eight of nine have HV60 > HV20 (NVIDIA is the sole exception). Volatility has been compressing.
• The options market accepted that compressed level as its baseline and added a modest premium on top, but priced nothing for a return to the quarter's turbulence.

The sharpest compression belongs to the two biggest gainers: SanDisk HV60/HV20 = 1.718, Micron 1.621.

Fact 2: the differentiation is in skew, not in volatility level

An easily missed detail: the two largest gainers carry almost identical volatility premia.

• Micron IV/HV20 = 1.281
• SanDisk IV/HV20 = 1.287

Their absolute IVs differ by more than a factor of one (63.28 vs 92.19), yet as a multiple of their own recent realized volatility they differ by just 0.006. The market applied the same multiplier to both.

The real disagreement sits entirely in skew, spanning 5.07 points:

Micron and SanDisk are both storage names, both rallied hard on the same day, and carry the same volatility multiplier — yet their skews sit on opposite sides of zero, 4.84 points apart. In plain terms: buyers of Micron picked up insurance along the way; buyers of SanDisk took pure exposure.

Fact 3: protection bought at the index, exposure taken in single names

Put/call ratios expose this layer cleanly:

• Nasdaq 100 (QQQ): 3.252
• Semiconductor ETF (SMH): 0.288

Same session, both index products, an 11.3x gap. On the single-name side: NVIDIA 0.342, Micron 0.373, Broadcom 0.390 — all below 0.4.

This is a recognizable structure: unwilling to get off, afraid of the fall. Risk is expressed in individual stocks while the hedge is parked on a broad index. The cost is specific — when the drawdown happens inside the sector rather than across the whole market, a hedge sitting on the Nasdaq 100 does not pay.

What this means

Everything above is data. What follows is interpretation — read them separately.

1. "Rallying" and "being repriced" are two different things. Spot repriced AI storage over two consecutive sessions; the options market did not move a notch. Every percentile lands in the bottom fifth of the in-house sample (SanDisk, the highest, is only 16.8). Reality is running; pricing has not followed.
2. This is not "cheap" — it is an assumption. IV below HV60 does not automatically mean options are underpriced. It means the market made an explicit assumption: the calm of the last 20 sessions is the norm, and the last quarter's turbulence was the exception. That assumption may prove right, but it is an assumption, not a neutral state.
3. NVIDIA is the most extreme cell. Its IV of 33.16 sits below its own HV20 of 44.57 (IV/HV20 = 0.744), with an IV percentile of 0.0 — no lower reading exists anywhere in the in-house sample. The largest name in the complex was priced as the least likely to move.
4. The disagreement in skew is more informative than the disagreement in price. Price tells you where the money went; skew tells you whether it brought insurance. On 09-22 money went into storage, but bought protection only on the Micron leg.

First test: the next session

⚠️ The figures below are live quotes from 10:31 ET / 22:31 Beijing time on 2026-09-23. That session was still trading — these are not closing values.

The complex gave back ground, and gave back more than the broad market — the semiconductor ETF at −2.03% against the S&P 500 at −0.56%. The decline did land inside the sector rather than across the market, which is precisely the scenario in which a Nasdaq 100 hedge fails to pay.

But this must be stated plainly: this is one intraday observation, not a verification. A single unsettled session with a sample of one cannot establish that skew predicts next-day returns. Treating one day's move as a "verdict" on the prior day's pricing is textbook hindsight attribution.

Falsification criteria (fixed in advance)

• Criterion 1: if, after the 09-23 close, SanDisk is still down more than Micron, and that ordering holds again on 09-24 and 09-25, the "low-skew leg is more fragile" reading strengthens. Any single-day reversal voids it — no intermediate interpretation will be offered.
• Criterion 2: if the median IV percentile across the nine names rises above 35 within the next three sessions, the conclusion "the options market has not repriced" expires — the market will have caught up.
• Criterion 3: if the put/call gap between the Nasdaq 100 and the semiconductor ETF narrows from 11.3x to under 3x, the structural claim "protection at the index, exposure in single names" is void.
• Criterion 4: if HV20 rises back above HV60 (i.e. volatility stops compressing), the entire "IV bracketed between the two" framework loses its premise and must be recomputed.

What this article does not support

• ❌ It does not support "options are cheap, so buy them." The contracts used here have 4-8 days to expiry; short-dated IV is heavily affected by weekend effects and the event calendar, and is not comparable to a 30-day constant-maturity measure.
• ❌ It cannot be read as "the lowest volatility in a year." The sample is n=98-127 trading days, not 252.
• ❌ It says nothing about who is buying or selling. Investor-type attribution requires exchange-level data broken out by investor category, which is not an input here.
• ❌ The 09-23 intraday figures must not be treated as settled conclusions.

Frequently asked questions

Q: If implied volatility is below realized volatility, does that mean options are underpriced?
A: Not necessarily. It means the options market expects less future volatility than the asset recently delivered. That can reflect pricing lag, or a correct anticipation of volatility convergence. All nine names here sit in that state (IV/HV60 averaging 0.844); the uniformity is itself notable, but uniformity is not the same as error.

Q: Micron and SanDisk both rallied hard the same day — why was their options pricing different?
A: At the level of volatility it was not different: IV/HV20 was 1.281 and 1.287 respectively, a gap of just 0.006. What differed was skew — Micron +3.72 (downside protection richer) versus SanDisk −1.12 (downside protection cheaper), 4.84 points apart. In the same rally, one leg was hedged and the other was pure exposure.

Q: What does a Nasdaq 100 put/call ratio of 3.252 signify?
A: Put volume on those contracts ran about 3.25 times call volume — clear hedging behaviour. Against the semiconductor ETF's 0.288, it indicates hedging demand concentrated on the broad index rather than on the semiconductor sector itself. When the drawdown occurs inside the sector, that protection does not cover it.

Q: What do positive and negative skew mean?
A: Positive skew means that, at comparable moneyness, puts are more expensive than calls — the market is paying a premium for downside. Negative skew is the reverse: calls are richer, and the market fears missing the upside more than it fears a decline.

Q: Why emphasize that the sample is 98-127 trading days rather than 252?
A: Because an IV percentile means nothing without its reference window. The common industry convention is 252 trading days (about one year). This sample is shorter, so "percentile 10.4" can only mean "low within an in-house sample of roughly six months" — not "a one-year low." Mixing the two conventions systematically overstates the conclusion.

Q: Does this article imply prices are about to fall?
A: No. It characterizes pricing structure, not direction. Low implied volatility predicts neither a rise nor a fall; it only shows that the market is paying very little for the possibility of a large move. That move could resolve in either direction.

Sources: options metrics from an in-house snapshot (2026-09-22 U.S. close, collected 06:00 Beijing time on 2026-09-23, covering 289 tickers); live quotes taken at 22:31 Beijing time on 2026-09-23. Not investment advice.