---
ticker: "MRVL"
company: "Marvell Technology, Inc."
date: "2026-10-22"
view: "Avoid"
fair_value: 188.06
price: 223.47
price_source: "Robinhood last regular-session trade at 15:59:59 ET; the official close was not yet published. Previous close $244.76 (2026-10-21)"
price_time: "2026-10-22T19:59:59Z"
price_vs_fair_value: "+18.8%"
bear: 32.77
bull: 441.3
exit_price: 337.59
p_price_justified: "20-50%"
restated: "2026-10-26"
method: "case-spread-2026-10-24"
original_fair_value: 188
original_view: "Avoid"
page: "https://alpha.dming.si/research/2026-10-22-mrvl"
---

# MRVL · Marvell Technology, Inc. · Avoid

Research report as of 2026-10-22, partly unblinded. Page: https://alpha.dming.si/research/2026-10-22-mrvl

We value MRVL at $188 a share. The report's $223.47 price (Oct 22, 2026) was 19% above fair value: Avoid.

_Restated Oct 26, 2026 (method case-spread-2026-10-24); the report's original value was $188, Avoid._

| Lens | Range | Marks |
| --- | --- | --- |
| Our report | $32.77 to $441 | fair $188; original $188 |
| Our simulation | $81.99 to $374 | median $175 |
| Street targets (40) | $190 to $450 | average $328, median $345 |
| Price | | $223.47 |

Bands run from our bear to bull (the hollow mark is the original $188), our simulation's 10th to 90th percentile and the Street's low to high. Street targets forecast future prices, not value today. Market data, captured Oct 22, 2026.

| Figure | Value |
| --- | --- |
| Fair value | $188 |
| Price | $223.47 (Robinhood, Oct 22, 12:59 PM PT) |
| Five-year return | 9.2% (a year, at the report's price) |
| Bear / bull | $32.77 / $441 |
| Exit price | $338 (on Jan 31, 2032) |
| Confidence | 25% |
| Price against our scenarios | The price is justified in 20-50% of our scenarios (4,000 simulated outcomes, each drawing the uncertain drivers from ranges we chose: our judgement, not a measured chance) |

**What the price needs.** Any one of these, on its own, would justify the price:

- FY2028 data center vs management's target: needs 1.18x, we have 1.00x
- Custom chips FY2029-32 vs our path: needs 1.39x, we have 1.00x
- Other data center growth vs our path, FY2029-32: needs 8.9%, we have 0.0%

_This text argues the original call. The current figures are at the top._

**Original call, Oct 22, 2026 (before the Oct 26, 2026 restatement)**

_As of 2026-10-22 · partly unblinded: six price-level exposures in deal filings, the earnout liability, the history file and Alpha's grades before the blind commit (assumptions.md). The blind value was $259; outside review moved it to $188 (section 9)._

- **Fair value today:** $188 per share, discounting at a 12.91% cost of equity. Weighting the pessimistic, central and optimistic cases 25/50/25 (our weights, not calibrated) gives $192.
- **Price:** $223.47, Robinhood, last regular-session trade 2026-10-22 15:59:59 ET (the official close was not yet published) → 15.8% above fair value
- **Five-year exit price:** $338 a share on 2032-01-31: the average of two estimates, one from letting Marvell's returns fade toward its cost of capital ($188) and one from what similar chip companies have traded at ($487, at 32.3x operating profit before acquisition amortization). Fading over 10 or 20 years instead gives $330 or $344.
- **Five-year return:** 9.2% a year from $223.47 to the central exit, dividends and buybacks included, against a 12.9% cost of equity
- **Bear / bull today:** $42 / $352: our pessimistic and optimistic cases, each one consistent business outcome.
- **Confidence:** About 25%: our judgment of the share of simulated values within 20% of $188 (subjective, not calibrated)
- **Moat:** Our judgment: Marvell is one of two companies (with Broadcom) that design custom AI chips for the largest cloud companies at scale, and a leading maker of the optical chips that link AI servers. It owns no factory and no software lock-in; its edge is design teams and customer programs.
- **Market expects:** More than our central: FY2028 data-center revenue 18% above management's ~$17.8B target, or the peers' multiple at about 42x instead of 32.3x, with everything else as in our central case
- **Price needs:** Any one of: FY2028 data-center revenue at 1.18x management's target; custom-chip revenue FY2029-32 at 1.39x our path ($13.8B rising to $22.0B); the rest of data center growing 8.9 points a year faster
- **Odds:** P(price justified): 20-50%, the share of our simulated scenarios in which the value is at or above the price. Eight in ten simulated values fall between $82 and $374. Subjective, not calibrated (odds.json)
- **Key driver:** The exit multiple: each 5x on the peers' multiple is about $19 a share (30x $179, 35x $198, 40x $217). Then data-center growth after FY2028: the optical and networking lines are worth -$43 / +$32 at 10 points a year slower / 8 faster; custom chips -$50 / +$51 at 0.45x / 1.6x our path.
- **Street:** Average 12-month target $328 (median $345, range $190-450): 113 targets published over the last year averaging $232, the last quarter's 40 at $319 and the last month's 20 at $361 (market data, fetched 2026-10-22). An expectations reference, not validation

**Avoid, at the edge of Hold: at $223.47 Marvell trades 16% above our $188 fair value;
1% more value would make it a Hold.** Our blind work, done before we looked at the price,
said $259 and Buy. Two outside reviewers and our own report check found two errors that
together cost about $67 a share, and we accept both (section 9):

- **The exit multiple compared unlike profits.** We valued Marvell's 2032 profits at the
  44.4x price-to-operating-profit ratio its peers have carried, but the peers' ratio is
  measured on profits after amortizing acquired intangibles, which for Broadcom and AMD is
  billions of dollars a year, while ours is before it. On a like basis the peers' ratio is
  32.3x (SEC filings, anchor_ebita.txt). Worth -$46.
- **Our margins ran ahead of what Marvell itself guides.** Marvell's finance chief said on
  2026-08-27 that the non-GAAP operating margin (before stock pay and amortization) reaches
  the upper end of 38-40% during fiscal 2028; our blind model had 41.9% for the year and
  44% by 2032. We now hold fiscal 2028 at 39.5% and let the margin rise with scale toward
  management's 44-46% only as far as our smaller revenue path reaches. Worth -$21.

Our revenue view is unchanged. We take management's target of about $20B in fiscal 2028
(the year to January 2028; Marvell's year ends in late January), because Marvell has beaten
the midpoint of its own revenue guidance in each of the last 11 quarters. We do not take its
$70-90B plan for fiscal 2031: our central reaches $35.6B, because our sector view holds the
big four cloud buyers' capital spending flat in 2027-2029.

On the owner's question, whether semiconductors are oversold: Marvell fell 8.7% on 2026-10-22
and 26% from its high in the week of 2026-10-05, and the sector fund SOXX is 2% above its
late-July low. On our method Marvell is still worth less than its price, though only just
below our Hold line. The price needs either management's long-range plan or a peer
multiple near the one we corrected away.

Terms. The **central case** is our best estimate; the **pessimistic** and **optimistic**
cases are coherent alternative outcomes we weight 25% each. **Custom chips** are processors
(XPUs) and supporting chips (XPU-attach) that Marvell designs to one cloud company's
specification. The **rest of data center** is Marvell's own catalogue: optical signal
processors for the fibre links between AI servers, plus network switches and storage
controllers. The **fade** values the business after 2032 by letting its returns fall toward
the cost of capital; the **outside view** values it at what comparable companies traded at.

## 1. What Marvell sells and what limits it

Data center was 74% of fiscal 2026 revenue ($6.1B of $8.2B) and 79% of Q2 fiscal 2027
($2.17B of $2.74B); the rest is chips for carrier and enterprise networks (10-K FY2026;
release 2026-08-27). Three things limit it:

- **Few buyers.** The top ten customers were 82% of fiscal 2026 revenue; one direct
  customer was 16% and one distributor 44% of Q2 fiscal 2027 revenue (10-Q Q2 FY2027).
  Custom programs are won and lost one cloud company at a time.
- **The cloud build itself.** Four buyers (Amazon, Microsoft, Alphabet, Meta) guide $733B of
  capital spending for 2026 (sector note). Marvell's data-center revenue was 0.83% of that
  in fiscal 2026; our central has it rising to 4.0% by fiscal 2031 (output.txt).
- **Lower margins on custom.** Custom chips carry lower gross margins than Marvell's own
  products (management, Q2 FY2027 call), so the blended margin drifts down as custom
  grows; management guides 56-59% for fiscal 2031 against 58.9% in Q2 fiscal 2027.

The customers have also bought into Marvell. NVIDIA paid $2.0B for preferred stock
convertible into 21.8M shares (8-K 2026-03-31). Alphabet holds a warrant for up to 59.0M
shares at $206.58 that vests one tranche per $500M of Google custom-chip revenue through
fiscal 2033, and vests only on Google's discretionary purchases (8-K 2026-08-19); Amazon
holds warrants tied to its purchases. We count them case by case.

## 2. Recent results against the bar

Q2 fiscal 2027 (to 2026-08-01): revenue $2,739M against a $2,700M guide midpoint (+1.5%);
non-GAAP EPS $0.94 against $0.93. Marvell's record against its own revenue midpoint over the
11 quarters we read runs +0.3% to +4.6%, median +0.95%; an outside reviewer found a miss
outside that window, Q3 fiscal 2023 at -1.5% (preview.txt). The Q3 guide is $3.15B ±5%, with
data center up more than 20% from Q2 and communications down in the low-to-mid teens.

Stock pay is rising fast: $534M in the first half of fiscal 2027 against $591M in all of
fiscal 2026, after the Celestial AI purchase (10-Q Q2 FY2027). The Q3 guide is $285M, 43%
of non-GAAP operating cost; we use 43% and charge it as a cost.

## 3. Our forecast and value

The model (model.py, output.txt) forecasts three lines to fiscal 2032 and values them with
the shared method: free cash flow paid to shareholders, plus an exit price in January 2032.

| Fiscal year | 2028 | 2029 | 2030 | 2031 | 2032 |
| --- | --- | --- | --- | --- | --- |
| Custom chips ($B) | 5.5 | 9.9 | 12.4 | 14.4 | 15.8 |
| Rest of data center ($B) | 12.3 | 14.8 | 16.8 | 18.8 | 20.7 |
| Communications ($B) | 2.3 | 2.3 | 2.4 | 2.4 | 2.5 |
| Revenue ($B) | 20.1 | 27.0 | 31.6 | 35.6 | 39.0 |
| Non-GAAP operating margin | 39.5% | 40.6% | 41.0% | 41.3% | 41.6% |
| Operating profit after stock pay ($B) | 6.3 | 9.0 | 10.7 | 12.3 | 13.6 |
| Free cash flow ($B) | 3.2 | 5.4 | 7.3 | 8.7 | 9.9 |

The margin row is the lower of two rules: operating cost growing at half the rate of
revenue (management's rule for fiscal 2028), and a cap of 39.5% in fiscal 2028 rising
from 40% at $20B of revenue toward 45% at $80B, a line through management's near-term
guide and its fiscal 2031 target (Q2 FY2027 call; Investor Day 2026-10-06). The cap binds
in every central year.

The exit averages two estimates. Letting returns fade toward the 12.9% cost of capital
gives $188 a share; the peers' 32.3x EV over operating profit before acquisition
amortization (Broadcom 19.3x, NVIDIA 45.4x, AMD 66.1x, Qualcomm 15.5x, ten-year medians of
fiscal year-ends; anchor_ebita.txt) gives $487. The method averages them: $338. At the
method's default 44.4x, which divides by profits after amortization, the central would be
$234. The price needs about 42x on a like basis. The two estimates differ because the fade, at a 12.9% cost of
capital and a 20% long-run return on new capital, pays 12.3x year-six after-tax profit,
about 10.5x pre-tax, against the peers' 32.3x: the peers' history prices a growth runway
the fade does not grant.

Shares at exit: 921.2M today plus 15.5M from the Google warrant on our central path, 10.3M
from the Celestial AI earnout and 4.0M from Amazon warrants: 950.9M. Google's exercise cash
at $206.58 is counted at the exit; in the pessimistic case the exit price is below that, so
Google's warrant is left unexercised.

The pessimistic case ($42) is an AI spending pause: fiscal 2028 revenue $16.6B, then data
center flat and falling 20% in fiscal 2030, the sector note's pessimistic shape, with
custom gross margin at 45%. The optimistic case ($352) puts fiscal 2031 revenue at $57.6B,
still below management's $70B low end. A path toward management's plan ($63.4B in fiscal
2031, 43.6% margin) is worth about $367 in our model.

## 4. Market expectations and how the price reacted

Weekly closes (TradingView, regular session): $81.69 in the week of 2026-02-23; a high of
$329.88 in the week of 2026-06-15; a low of $162.90 in the week of 2026-07-27; a high of
$301.27 in the week of 2026-10-05, the week of the Investor Day; $227.83 in the week of
2026-10-12. On 2026-10-22 the regular session ended at $223.47 against $244.76 the day
before (Robinhood), down 8.7%, while SOXX fell 6.8% to $475.56 from $510.36. We did not
check the day's news for a company-specific cause; the sector fell with it.

Running our model backwards, the price is justified by any one of: fiscal 2028
data-center revenue 18% above management's target; custom revenue at 1.39x our path; the
rest of data center growing 8.9 points a year faster; or a peer multiple of about 42x on
profits before amortization (odds.json, output.txt). The first two ask for more than
management guides for fiscal 2028 or than our custom path, which is already near
management's $12B-plus for fiscal 2029 in the optimistic case.

Alpha, our momentum model, ranks Marvell 55th of about 700 on 2026-10-22 (tier Strong):
growth A, momentum A, quality B, expectations D, valuation D. It ranks on momentum and
growth; we value the business. The two disagree because the stock's rise has run ahead of
the value we find.

Timing, a quick chart read: the stock is inside a $163-330 range since June, made a lower
high in October ($301 against $330) and sits near the August-September lows of $200-215;
below those, the July low of $163 is the next level. Grade C: no trend to lean on either
way. Timing is beside the view, never an input to it.

## 5. Where we differ from the Street

The Street is more optimistic than we are. Its averages (market data, 2026-10-22) are revenue of
$19.5B, $31.4B, $48.1B and $75.0B for fiscal 2028-31 (ours $20.1B, $27.0B, $31.6B,
$35.6B), close to management's plan. Its average target of $328 discounted one year at our
cost of equity is $290; the targets span a year, and the last month's average is $361. We
agree with the Street on fiscal 2028 and differ after it: the long-range plan needs the
big buyers' capital spending to keep growing, and our sector view has it flat to 2029.

## 6. Why the other side could be right

The case for the price: management's plan. Run through our model, a path toward it is
worth about $367, and Marvell has raised its fiscal 2028 target twice this year. The
Street's revenue path sits near that plan. If the big buyers keep raising capital spending
in 2027-2029 instead of holding it flat, our central is too low, and so is our view.

The case for worse: an AI spending pause. Marvell's data-center revenue fell 8% in fiscal
2024 in the last cloud spending digestion (market data segments). A pause in 2028 with one custom
program lost is the pessimistic world in our odds: median $114, the price justified in
under 10% of its draws. One direct customer was 16% of Q2 revenue, and the Google warrant
vests only on Google's discretionary purchases, while Marvell's $8.5B of foundry
commitments are firm (10-Q Q2 FY2027). An outside reviewer's harsher stress, a 20-25x
peer multiple in the pause, gives $29-35.

## 7. Next results: Q3 fiscal 2027, expected 2026-12-01 (not confirmed by the company)

Ranges aim to hold about 80% of outcomes, set from Marvell's record against its own
guidance midpoint and widened to allow a miss like Q3 fiscal 2023's (preview.py).

| Forecast | Ours | 80% range | Why |
| --- | --- | --- | --- |
| Revenue ($B) | 3.180 | 3.103-3.230 | guide $3.15B plus the median beat (0.95%) |
| Data center revenue ($B) | 2.686 | 2.592-2.753 | revenue less communications down 13% |
| Diluted EPS ($, non-GAAP) | 1.11 | 1.06-1.15 | guide $1.10 plus the median beat |
| Diluted EPS ($, GAAP) | 0.53 | 0.38-0.65 | guide; the earnout is marked to the stock price |
| Q4 FY2027 revenue outlook midpoint ($B) | 3.629 | 3.447-3.846 | the ~$12B fiscal 2027 target less H1 and our Q3 |
| Diluted shares (M) | 921 | 915-929 | guide |

The quarter matters less than any change to the fiscal 2028 target: each $1B of fiscal
2028 data-center revenue is worth about $11 a share in our model.

## 8. Dates to watch, and what would change the view

- 2026-12-01 (unconfirmed): Q3 results and the Q4 guide.
- The four big buyers' 2027 capital spending guides (late January to early February 2027):
  growth beyond 2027 in their plans is the evidence that would move us toward
  management's path; a cut of 20% or more is our pessimistic path.
- Any update to the fiscal 2028 target, at the latest with Q4 results (about March 2027):
  a data-center outlook above about $21B (1.18x today's) would justify the price alone.
- Google custom revenue: each warrant tranche that vests (10-Qs) marks $500M of Google
  purchases; fewer than four by the end of fiscal 2028 says the program is slipping.

## 9. What changed from the blind value, and the review

The blind commit (a58a893) valued Marvell at $259. Two outside reviewers (Codex, one
testing the pessimistic case and auditing the model, one arguing the opposite of the
call) and a report check (research-eval) reviewed draft 9d545d08c. Each effect below is
the central value with that one change undone, from output.txt; undone together with the
small items the changes interact, and the three small bridge fixes and their interactions
net to -$1.73.

| Finding (who) | Answer | Central effect |
| --- | --- | --- |
| Peer multiple on profits after amortization against our profits before it (both reviewers, report check) | Fixed: 32.3x EV over operating profit before amortization, from SEC filings (anchor_ebita.py) | -$45.62 |
| Margin above Marvell's guide and history (report check) | Fixed: fiscal 2028 at 39.5%, then 40% rising with scale toward 45% at $80B. Rejected a cap at the fiscal 2023 high of 35.5%: Q2 fiscal 2027 was already 36.6% and Q4 fiscal 2027 is guided to 38-40% | -$20.97 |
| Stock pay at 40% of operating cost too low; Q2 was 53% (advocate) | Fixed at the Q3 guide's 43%; 53% shown as a sensitivity (-$9.14) | -$2.64 |
| Google exercise cash at $206.58 left out (pessimistic reviewer) | Fixed; Google's warrant unexercised in the pessimistic case | in -$1.73 |
| Kept cash earned no interest (both, report check) | Fixed: risk-free after tax, 4.50% | in -$1.73 |
| Operating leases deducted while rent is in operating cost (pessimistic reviewer) | Fixed: lease liabilities removed from the claims | in -$1.73 |
| Pessimistic case too generous at a 43x multiple (pessimistic reviewer) | Fixed by the multiple change: $52 to $42; their 20-25x stress ($29-35) disclosed | pessimistic only |
| The pessimistic case also cuts long-run growth and return (report check) | Rejected: a pause that shrinks data-center revenue also shortens the growth runway, and 12% is Marvell's own return with acquisitions | pessimistic only |
| Use Marvell's raw beta of 2.26 (advocate) | Rejected: the method uses the peers' median, adjusted toward 1; Marvell's own adjusted beta is shown (-$1.62) | none |
| Working capital 25% of the revenue change, not 20% (advocate) | Disclosed: -$1.06 by the advocate's run | none |
| Custom gross margin 48% unsupported (both) | Disclosed: the pessimistic case uses 45%; a 40% stress costs about $22 by the pessimistic reviewer's run on the blind model | none |
| Google shares for fiscal 2033 counted at the 2032 exit (pessimistic reviewer) | Disclosed: +$0.90 on the blind model if left out | none |
| Next-results ranges held in sample only (advocate) | Fixed: low ends widened to allow a miss like Q3 fiscal 2023's | forecasts only |
| Management comparison mislabelled (pessimistic reviewer) | Fixed: it is a path toward management's plan, $63.4B, not the plan | none |
| No Alpha view, unlabelled moat and inferences, targets and price moves undated (report check) | Fixed in this report | none |

Both changes rest on filings, not on the price: the amortization data are SEC filings and
the margin guide is the 2026-08-27 call, read before the blind commit. Together they moved
the central 27%, more than the 20% at which we flag a review's effect for the reader.

## 10. Sources

Filings masked before reading and pinned in data/: 10-K FY2026; 10-Qs Q1 and Q2 FY2027;
releases Q3 FY2024-Q2 FY2027; 8-Ks 2025-12-02 to 2026-09-25 (the 2026-08-19 warrant
exercise price read unmasked after the blind commit). Transcripts: Q2 FY2027 call
(2026-08-27), Investor Day (2026-10-06). Market data statements, segments and peer betas
(2026-10-22). Peers' amortization: SEC XBRL company facts (fetched 2026-10-22). Sector note
docs/research/sectors/semis-ai/note.md (2026-10-15). Market pass: Robinhood quote
2026-10-22; Street targets and estimates via scripts/research-check.py (2026-10-22);
TradingView weekly bars (data/market/).