The thesis under review: the AI build-out needs custom accelerators and optical interconnect as well as GPUs; Marvell is a main supplier of both and has just raised its long-range targets again; the revenue ramp holds the stock up into its early-December report and lets it retest the June record of $330. The practical question is narrower: at what price, and in which expiry, is it worth being paid to own Marvell with a short put? We test each link, read the chart and dealer positioning, and price both the put sale and the bullish structures.
Our view in one paragraph. Refreshed 17 October: the $250 shelf this page recommended selling against on 14 October failed the next session. Marvell closed Friday at $227.83, −17% on the week and 25% below the 6 October investor-day high, through the June swing low, the $250 put wall and the 50-day average, and the dealer book has flipped from long to short gamma. The 20 November $250 put recommended here is now about $23 in the money; what follows replaces that advice. The story has not changed; the price of it has. At its 6 October investor day Marvell said it ships custom silicon to all four large US hyperscalers (35+ sockets), raised its FY28 revenue target to about $20 billion, set FY31 at $70–90 billion, and claimed a two-year lead in 1.6T optical DSPs; eleven brokers raised targets, to as high as $450. Second-quarter revenue was $2.74 billion (+37%), data centre 79% of it, and the third-quarter guide is $3.15 billion. The weaker links are share and margin: Broadcom's AI revenue is about five times Marvell's whole data-centre business, Amazon's Trainium position is contested by Alchip, gross margin is stepping down as custom silicon grows, and the stock fell 10% on a beat-and-raise in August. At about 31× next year's earnings after a 25% fall in eight sessions the stock prices less of the plan than it did, and the fall was again the sector's (SMH −10.5% on the week, its worst since April 2025, on AI-debt and Chinese-memory fears) with no Marvell news in it. A retest of $330 by Christmas is a chance on the market's numbers and we would not pay for it. For a put seller the setup is different in kind from last week: there is no tested level under the price, dealers are short gamma below $242.6 and chase a break, and the next shelf is the 1 September swing low at $200.62, which is also the new put wall. Sell the 20 November $200 put for about $6.67 a share, paid to own Marvell at $193.33, 15% below Friday's close, under the September low and the put wall, with no earnings in the way; and do not sell it on a down day. Wait for the first close above the previous day's high, because in negative gamma a fall does not stop at a level, it stops when the selling does.
Hyperscaler capex is about $725 billion in 2026 and expected above $1 trillion in 2027. Broadcom's AI revenue was $16.7 billion last quarter (+221%) with a $21.7 billion guide and six XPU customers, the clearest evidence that custom accelerators are a large and growing share of AI spend. Optical interconnect scales with every cluster: 1.6T transceivers are in volume, and co-packaged optics are next.
Verdict: true and quantified. The question is not whether the demand exists but how much of it Marvell keeps.
In optics, yes: Marvell says its 1.6T DSP ships in high volume two years ahead of its competitor, analysts put its DSP share at 60–70%, and interconnect is guided to grow more than 70% this fiscal year. In custom silicon it is a distant second to Broadcom. It now ships to all four large US hyperscalers, signed a Google deal for TPU-attach parts on 29 July (with a warrant to Google on 59 million shares at $206.58 that vests with up to $120 billion of purchases), and Microsoft's Maia 300 is reported to be co-designed with Marvell. But Amazon's Trainium 3/4 compute work is reported to have moved to Alchip (never confirmed by either company), Google can move any programme at each generation, and Broadcom and Credo are now shipping competing 1.6T parts.
Verdict: a clear leader in optics and a credible number two in custom silicon; the Amazon socket is the risk to watch at re:Invent (30 Nov–4 Dec).
Revenue is ramping fast: $2.74 billion in Q2 (+37%), a $3.15 billion Q3 guide (consensus was $3.03 billion), FY27 about $12 billion and FY28 now about $20 billion, with about $1 billion of customer capacity prepayments. Margin is going the other way: non-GAAP gross margin is guided to 57.5–58.5% in Q3, down about 90 basis points on custom mix, and consensus sees it near 57% by FY29. Earnings growth has to come from operating leverage.
Verdict: revenue yes, margin no. The market punished the August print (−10% the next day) for exactly that mix.
Marvell is up this year against for the semiconductor ETF and for Broadcom. It trades at about 37× FY28 consensus EPS of $7.28 and 20× FY27 sales, the richest multiple in its group (Broadcom 19×, Nvidia 15× forward earnings). The consensus target is about $338 after the investor-day raises. $330 needs the early-December report to beat a $3.15 billion guide and lift the Q4 outlook, and no new AI-capex scare in the meantime: the 14 September "pace the frontier" sell-off and the 28 September OpenAI training-halt headline each knocked 5–6% off the stock in a day, with a hiking Fed and a 5.3% ten-year yield as the backdrop.
Verdict: holding support is likely; the retest needs the report. That asymmetry is why we lead with a put sale that expires before it.
What confirms it: a close above $301 (the investor-day high and the call wall) on volume, then $324; consensus FY28 EPS moving above $7.60.
What confirms it: a close below the $254 gamma flip and the $248 September low; a Q4 guide below about $3.6 billion.
| Date | Event | Why it matters for the thesis |
|---|---|---|
| 9 Oct | Ex-dividend ($0.06) | Immaterial for options. |
| 16 Oct | Monthly options expiry | Part of the $280–300 call gamma rolls off. |
| 28 Oct | FOMC | Rates: a second hike would weigh on high-multiple semis. |
| ~28 Oct–5 Nov | Hyperscaler results and 2027 capex guides | The demand signal for custom silicon and optics; watch Microsoft Maia and Amazon Trainium commentary. |
| 15–20 Nov | SC26 (supercomputing), photonics panel 18 Nov | Optics news flow. |
| 20 Nov | Monthly options expiry: the recommended put | Before the report. |
| ~mid–late Nov | Nvidia results | Sets the AI-semis multiple. |
| ~1 Dec (after close) | Marvell Q3 FY27 results (date not yet announced; window 26 Nov–4 Dec) | Revenue vs the $3.15B guide, gross margin vs 57.5–58.5%, the Q4 guide. |
| 30 Nov–4 Dec | AWS re:Invent | Trainium 4: the Amazon socket question. |
| 8–9 Dec | FOMC with projections | Rates path into year-end. |
| ~mid Dec | Broadcom Q4 results | Read-across for custom silicon. |
| 24 Dec | Target date |
Daily closes. Hover for values; the table shows the same data.
| Total return | 1 week | 1 month | 3 months | 6 months | YTD | 1 year |
|---|
Marvell is up in six months and this year while Broadcom is down: the market rotated from the largest custom-silicon supplier to the one with the most upside to its targets. That kind of relative strength persists while estimates rise, and it reverses hard when the tape turns: this week Marvell fell against for Broadcom and for SMH, with no news of its own.
Blue candles closed up, red closed down. Orange dashed: dealer-gamma levels (call wall, put wall, gamma flip); swing support and resistance are listed in the key-levels table. Hover for the crosshair: the price is tagged on the price axis and the date on the time axis. Right panel: net dealer gamma per strike on the same price axis (blue where dealers are long gamma, red where short).
Trend: the shelf failed, and the stock is 25% below the investor-day high in eight sessions. The year has four acts: a run from $71 in February to the $329.88 record on 18 June; a 40% drop to $163 by late July; a recovery from the $201 low on 1 September to the investor-day high of $301.27 on 6 October (51 million shares, nearly three times average); and the sell-off since: −3.5% (8 October), −3.4% (12 October), the 13 October reversal from a $248.45 low that this page took as a successful test, then −8.7% on 14 October to $245.28, closing through the shelf on 26 million shares, −1.6% on 15 October (low $227.31, close $241.24: a lower wick that was not followed), and −5.6% on 16 October to $227.83, the low of the day and of the week. Four of the last five bars are full-bodied red; the one reversal bar was engulfed the next day.
Levels. Overhead: the 50-day at $241 and the gamma flip at $242.6, together the level a bounce has to reclaim; $245–248 (the failed shelf); the 20-day at $263; $275–285; and $300–301 (the investor-day high and the call wall). Below: nothing tested until the 1 September swing low at $200.62 and the $200 put wall; then $190 and $180 (the next put strikes), the 200-day at $176 and the July lows at $163–178.
Volume. The decline has run on 18–26 million shares a day against a 20-day average of ; Friday's 21.4 million was average, not a capitulation. None of the eight bars has the shape of a washout (a wide range closing near its high on heavy volume): the 13 October bar had the shape and not the follow-through.
Technical verdict: a downtrend inside negative gamma, not a pullback to support. The level this page defended on 14 October failed in one session; the 50-day and the flip are now overhead; the next support is 12% lower. For a put seller that means two things: the strike goes under the next shelf rather than on it ($200, not $220), and the entry waits for a reversal bar (a close above the previous day's high) instead of anticipating one. A close below $200 opens the July lows ($163–178) and makes the $200 put a loser; a close back above $242.6 turns the dealer book positive and makes $263 the next stop.
| Level | Price | vs spot |
|---|
Support and resistance are clustered swing lows and highs from the last eight months of daily bars; the gamma levels come from CBOE open interest.
What the profile says for Marvell. Dealers are short about of gamma per 1% move, of it in the next 35 days: a change of sign since 14 October, when the book was long $80 million and one of the largest positive books we measured. The $250 puts have been absorbed and the protection re-struck lower: the largest put strikes are now $200 (−$8 million, the put wall), $220 (−$6 million), $230, $190 and $180; the call side is still $300 (+$7 million, the call wall) with $280–290 behind it. The flip is $242.6, at the 50-day average. Below the flip dealers chase the move, selling into weakness and buying into strength, which is why this week's falls extended into the close and why the reversal, when it comes, will be sharp. For the trade: the strike belongs at or below the $200 wall, not in the $200–242 zone where the hedging is one-way; and a close back above $242.6 is the first sign the regime has turned.
How to read this. Market makers who sell options hedge them with stock; how much stock they must buy or sell as the price moves is their gamma. We estimate it from CBOE open interest for every listed strike and expiry, recompute each option's gamma with Black-Scholes from its own implied volatility, and sign it by the usual convention: dealers are long the calls customers sell and short the puts customers buy. Positive net gamma means dealers sell as the stock rises and buy as it falls, dampening moves; negative means they chase the move and amplify it. The sign convention is an assumption, and open interest is end-of-day, so the profile changes after every expiry.
Interactive Brokers' 30-day implied and historical volatility series (refreshed 14 October).
Solid: implied volatility of the at-the-money option for each listed expiry, on a calendar axis. Dashed: the forward volatility the market prices for the period between one expiry and the next; the spike is the earnings report. From CBOE's delayed chain.
Last three months of closes, then the range the market prices to each expiry: today's price plus and minus the at-the-money straddle (shaded, the expected absolute move) and ± one standard deviation from at-the-money IV (dotted). Green dashed: the report's target.
Out-of-the-money puts left of spot, calls right of spot, for the first monthly expiry after a month, the expiry that carries the earnings report, and January. A smile that rises to the right means upside calls are bid: the market pays up for a squeeze.
Implied volatility is fair, about what the stock delivers. Thirty-day at-the-money IV is against realized volatility of , with an IV rank of in a one-year range of – (the top of that range is the April–May spike, when realized volatility passed 140%). Premium sellers are paid about what the stock delivers; the edge has to come from the strike (under the next shelf and the put wall) and the timing (after a reversal bar, before the report).
The term structure is flat to the report. At-the-money IV is 68% for next week's expiry, 65–67% to 6 November, 64% for the 20 November monthly, 61% for 27 November and 64% for 18 December, the first expiry after the ~1 December report. The market prices that report at about ±. Marvell's last eight reports moved the stock +23.2%, −19.8%, −5.6%, −18.6%, +7.9%, +18.4%, +3.1% and −10.3% the next day: a mean absolute move of 13.3%, four up and four down. The earnings premium in December options is thin; that argues against selling puts through the report and in favour of the 20 November expiry, which ends eleven days before it.
The smile has turned to the downside. In the 20 November expiry the $200 put trades at 66% volatility against 63% at the money and 61% for the $260 call: 25-delta puts are now 4–6 volatility points above 25-delta calls out to November, where a week ago calls were 1–3 points above puts. By December and January the smile is flat again. The market has started paying for protection in the front expiries, and the $200 put at 23 delta carries that premium: the one thing that has improved for a seller.
The cone. The one-standard-deviation move to the target date is ±; the 20 November straddle prices ±16% and the January straddle ±25%. $330 is now 1.6 standard deviations away, which is why the market gives it .
| Structure | How it is built | What you want | Risk | Best used when |
|---|---|---|---|---|
| Long call | Buy a call | A big move up, fast | The premium (100% if it expires below the strike) | You expect a move larger than the market prices; implied volatility is low |
| Bull call spread (debit) | Buy a call, sell a higher call | The stock at or above the upper strike at expiry | The net debit | You have a price target and a date; you want the highest return per dollar to that target |
| Cash-secured short put (credit) | Sell a put, hold the cash to buy the shares | The stock to stay above the strike; you keep the credit | Owning the shares at the strike in a sell-off (large, stock-like) | You would be happy to own the stock lower; volatility is high |
| Bull put spread (credit) | Sell a put, buy a lower put | The stock above the short strike at expiry; keep the credit | Spread width minus credit (defined) | You are bullish-to-neutral and want to be paid to wait with a known worst case |
| Risk reversal | Sell a put, buy a call (often for zero cost) | A large rise; behaves like stock above the call strike | Stock-like below the put strike; margin required | You want stock exposure without paying premium and accept the downside |
| Put-financed call spread | Sell a put, buy a call spread | The stock at the top of the spread; the put pays for it | Stock-like below the put strike | You want a call spread for free and accept assignment risk |
| Put spread + call spread | Sell a put spread, buy a call spread | The stock above the call spread's top strike | Both spreads' widths minus net credit (defined both ways) | You want a leveraged payout at the target and a known worst case, with part of the cost paid by the put spread |
Returns on the short-put structures below are shown on a cash-secured basis (capital = strike minus credit). Brokers typically require about 20% of that as margin; for a stock this volatile expect more.
Two briefs: a put sale (be paid to own Marvell below the price, ideally in an expiry without earnings) and the bullish thesis to $330 by 24 December. We priced every $5–10 strike call from $230 up and put from $150 to $290 in the November, December, January and February monthlies from the CBOE delayed chain at the 16 October close (IB Gateway was offline this weekend; Greeks are model values), and valued each structure on the target date under four scenarios.
Scenarios on 24 December: Bear $181 (−20%, a post-earnings drop through the July lows; implied volatility +5 points) · Flat $227 · Base $245 (+8%, back to the 50-day) · Bull $330 (+46%, the thesis; implied volatility −6 points).
| Trade | Net | Max loss | Break-even | Delta | PoP |
|---|
Net is the mid-price debit (paid) or credit (received) per share; one contract is 100 shares. Max loss is the capital at risk per share (for a short put, the strike minus the credit). PoP is the lognormal probability, at today's implied volatility, of finishing above break-even at expiry. Scenario cells show profit or loss as a percentage of the capital at risk on 24 December (20 November trades have expired by then, so their cells are the result at expiry); −100% means a total loss. Highlighted rows are the recommendation.
| Trade | Capital at risk | P&L | Return |
|---|
Pick up to four trades. Values on the target date use Black-Scholes with each contract's own implied volatility; trades expiring before then show their value at expiry.
The 14 October advice (sell the 20 November $250 put) is withdrawn: that put is about $23 in the money and the level it stood on is gone. The new brief is the same idea one shelf lower, with one extra rule: no entry on a down day.
| Strike | Bid | Ask | Mid | Break-even | IV | Delta | Theta/day | Volume |
|---|
Prices, implied-volatility history, option quotes and Greeks: Interactive Brokers via IB Gateway, refreshed (16 Oct 2026 close; this refresh and the first build on 8 Oct used Yahoo prices and CBOE quotes because Gateway's history service was down, the 14 Oct refresh used IBKR). Term structure, smile, straddles and dealer gamma: CBOE delayed chain. Fundamentals and news: Marvell Q2 FY27 results (8-K Ex. 99.1, 27 Aug 2026) and call; Marvell Investor Day transcript (6 Oct 2026); CNBC, Motley Fool and Yahoo Finance on the Q2 reaction (27–28 Aug 2026); TechTimes and Futurum on the Google deal and warrant (Aug 2026); Motley Fool (18 Jun 2026) and RBC on Trainium; Wedbush on Maia 300; DigiTimes and Futurum on 1.6T DSP competition (2026); Broadcom FQ3 2026 results (Sep 2026); CBS, Fortune and Parameter on the 14 and 28 September sell-offs; CNBC on the Fed decision (16 Sep 2026); TipRanks, Benzinga, CNBC and 24/7 Wall St on target changes (7 Oct 2026); stockanalysis.com and Yahoo Finance for consensus and statistics (7–8 Oct 2026); Marvell 10-K (Mar 2026) for concentration and China exposure; Yahoo daily closes for earnings-day moves.