Google wired itself into Marvell’s roadmap. Then the market read the vesting schedule.
Revenue grew 37% to a record $2.739bn last quarter and management raised the fiscal-2028 outlook to $18bn — yet the stock fell 10.3% the next morning and sits 36% below its June high. Marvell sells the connective tissue that attaches to somebody else’s compute; the debate is whether that attach point is a moat or a queue position. Six analyst lenses, three scenarios, four horizons.
Gray line = Marvell’s actual price into today: $127.48 all-time high (Jan 23, 2025) → $47 on the April 2025 tariff shock → $61.44 52-week low (Sep 4, 2025) → $329.88 52-week high (Jun 18, 2026) → a $163.40 trough on Jul 29 → $210.39 now. Colored paths are synthesized scenario midpoints, probability-weighted bear 30% · base 45% · bull 25%. Log price axis — the stock has traded a 5.4× range in twelve months, so a linear axis would flatten the history into a line. The last four months are compressed at the right of the history panel because the time axis is continuous; that near-vertical spike and plunge is real, not an artifact. Wall Street’s 12-month consensus is $284.80 (range $143–$400, “Strong Buy” from 39 of 44 analysts, S&P Global, Sep 1, 2026).
Re-weight the scenarios
Those probabilities are a judgment call — so make them yours. We start bear-tilted at 30 / 45 / 25 because the multiple, not the business, is doing most of the work here. Drag to set how likely the bear and bull cases are (base takes the remainder); the blended target below, the dotted line on the chart, and the prob-weighted row of every scenario card update live.
Six desks look at the same quarter and reach a $175 spread
Each lens was reasoned independently before reading the others, so the disagreement is real rather than staged. Targets are 12-month and each states the multiple behind it. The mean of the six is $217.50 — meaningfully below the Street’s $284.80, which is the honest summary of this desk’s view.
Estimates are still going up
A company that has raised its forward outlook twice in two quarters, while growth accelerates, does not get a cyclical multiple mid-inflection.
- FY27 revenue outlook lifted to $12bn (~+45%); FY28 raised $1.5bn to $18bn (~+50%).
- Data-center growth guidance raised from ~50% to 60% in FY27 and above 60% in FY28.
- Custom silicon to more than double in FY28 and exceed $10bn in FY29; interconnect +70% this year.
- Non-GAAP operating margin enters a 38–40% band in Q4 FY27, from roughly 37% today.
FY29E EPS $9.50 × 33×
The moat is the plumbing, not the processor
Marvell’s differentiator was never winning the ASIC bid; it is owning every metre of the link around it. The question is whether the plumbing keeps you in the room when the compute contract is re-bid.
- Full-distance coverage: scale-up (NPO/CPO), scale-out (1.6T PAM DSP, 51.2T Ethernet), scale-across (DCI).
- Celestial AI ($3.25bn, closed Feb 2, 2026) adds Photonic Fabric — $500M run-rate targeted by late FY28, $1bn by FY29.
- But Credo holds ~88% of the AEC market and Astera Labs owns retimers — the “full-distance” claim gets picked apart lane by lane.
- Google’s warrant covers TPU-attach silicon. Marvell is paid to be adjacent, not central.
FY29E EPS $8.90 × 28.5×
A levered call on one crowded line item
Strip away the product detail and this is a geared bet on 2027 hyperscaler capex — the single most consensual assumption in the market right now.
- Beta 2.25, no meaningful yield (0.11%), sub-1% FCF yield: maximum duration risk when bond yields back up, as they did on Sep 1.
- Fabless and leading-edge dependent — Taiwan concentration and export-control regimes are live variables.
- Marvell and Broadcom share the same vulnerability; execution quality does not protect either from a capex pause.
- The Oct 6, 2026 investor day is the swing factor: it either re-anchors FY2029+ or confirms the back-loading fear.
FY29E EPS $8.90 × 25× cycle-average
The re-rate already happened — through earnings
The counter-intuitive finding: on its own multiple history MRVL is close to fair. What has broken is the momentum factor, and the beta means you are not paid for that.
- Forward P/E 38.6× on next-twelve-month EPS vs a ~31.8× average of the last five fiscal year-end readings — a ~21% premium, not a bubble.
- But price-to-sales is 19.5× against 8.2× at the FY26 close: the sales multiple is stretched.
- Realised vol is extreme: four sessions of ±9% or worse since Jul 1; the Aug 28 gap-down came on 49.0M shares, 2.3× average.
- Three successive rallies (Jul 23, Aug 17, Aug 26) each failed lower against the June high — that is distribution, not accumulation.
NTM EPS $5.45 × 36× partial mean-reversion
The cash is not scaling with the story
Revenue is up 30.6% over twelve months. Free cash flow margin went down. At 107× trailing free cash flow you are underwriting a conversion improvement that has not yet appeared.
- TTM FCF $1.73bn on $9.45bn revenue = 18.3% margin, versus 24.2% in FY2025. FCF yield 0.94%.
- GAAP operating margin 16.8% against roughly 37% non-GAAP — the gap is stock comp and acquisition amortization, both real costs.
- FY26 GAAP EPS of $3.07 exceeded non-GAAP EPS of $2.84 — an inversion caused by a $1.8bn pre-tax divestiture gain, not by operations.
- $200M of buyback per quarter against a $184bn market cap is a rounding error; net debt is $1.35bn.
FY28E FCF ~$4.0bn × 40× P/FCF
A seven-year IOU priced like a signed backlog
The $120bn Google headline is a vesting schedule, not an order book. Strip the story out and you own a 79%-data-center business with a handful of customers at 50× forward earnings.
- Only ~1.4M of the 58,970,907 warrant shares vest automatically in year one; the rest at ~240,000 shares per $500M of purchases through FY2033.
- Broadcom holds ~70% of custom AI silicon share and guides above $100bn of AI revenue in FY2027 — roughly six times Marvell’s entire company. Second sources do not set price.
- Mix is dilutive by management’s own guidance: non-GAAP gross margin steps down to 57.5–58.5% because custom is accelerating.
- Concentration has bitten before: $127.48 in January 2025 to the high $40s by April 2025, on fears about one customer program.
- Insiders sold roughly $632M in three months; the CEO sold 7,500 shares at ~$236 on Aug 17.
FY29E EPS $6.40 × 21×
Forty-four analysts, zero sell ratings, and a $257 spread
Every target below was published on Aug 28, 2026, the morning after the Q2 print, except Goldman Sachs (mid-August, Neutral). The striking feature is not the average — it is that after a 10.3% one-day drop, almost nobody cut. The Street reads the sell-off as timing, not thesis. That unanimity is itself a risk factor.
Individual targets via TipRanks / S&P Global as of Sep 1, 2026. Ratings distribution: 31 strong buy, 8 buy, 5 hold, 0 sell. Mean $284.80, median $280, low $143, high $400. Goldman’s $195 Neutral predates the Q2 report; on Sep 1 the firm reiterated that Marvell “demands caution” while calling Broadcom a buy. Note that twelve of these sixteen sit above this desk’s own base case of $265.
What each world is actually worth
Every price below is an EPS path multiplied by an exit multiple — nothing more. The clay row is the probability-weighted blend and is wired to the sliders above: change the weights and it moves. Bars are normalised within each card, so bull is always full width. These are illustrative frameworks, not forecasts.
1 year
Sep ’272 years
Sep ’283 years
Sep ’295 years
Sep ’31▸ Show the assumptions & math
Marvell’s fiscal year ends in late January, so FY2027 is the year ending Jan 31, 2027. A 12-month target discounts the fiscal year ending roughly sixteen months out — a September 2027 price reflects FY2029 earnings. Anchors are hard: FY2026 non-GAAP EPS of $2.84 is reported; FY2027E $4.20 and FY2028E $6.70 are S&P Global consensus (33 analysts). Everything from FY2029 is this desk’s modelling.
Base case. Management’s own $18bn FY2028 target lands, custom silicon clears $10bn in FY2029, and the non-GAAP operating margin holds in the 38–40% band. Revenue compounds toward roughly $23–24bn in FY2029 at a ~40% operating margin and an 11% non-GAAP tax rate, on a diluted share count drifting toward ~930M as stock compensation and early Google warrant tranches vest. The multiple compresses from 50× forward today to 24× by 2031 — still a premium to Broadcom’s ~20×, justified by a faster growth rate off a smaller base.
Bear case. Hyperscaler capex digests through calendar 2027–28. Google keeps Broadcom as primary and Marvell’s tranche revenue arrives late and thin; custom mix drags blended gross margin below 55%; FY2028 lands near $15bn rather than $18bn. The multiple resets to the low-20s and then the high-teens. Note the shape: the bear path troughs in year two and recovers — a digestion cycle, not a terminal decline. Even so, the two-year drawdown is 44%.
Bull case. The Google relationship compounds faster than the vesting cadence implies, Celestial AI’s Photonic Fabric becomes a genuine third leg past FY2029, and operating margin pushes above 42% as custom scales on fixed engineering. This is the world in which Marvell stops being priced as a second source.
Why 30 / 45 / 25. The default tilts bearish because valuation, not the business, carries the risk here: 50× FY2027 earnings and 19.5× trailing sales against a five-year sales multiple closer to 8×. The business is executing; the price already assumes it keeps executing. Move the sliders if you disagree — that is what they are for.
Revenue is climbing a wall. Cash is walking up a ramp.
This is the single most useful chart in the report, because the bull and bear cases both live in the gap between the blue bar and the green one. Revenue roughly doubles from FY2024 to FY2027E. Free cash flow does not keep pace — the FCF margin has fallen from 24.2% in FY2025 to 18.3% on a trailing basis. Debt is drifting up, not down. Scale is arriving; conversion is the open question.
Reported figures from Marvell fiscal-year results and S&P Global (through the quarter ended Aug 1, 2026). FY2027E revenue is management’s own $12bn outlook given Aug 27, 2026; FY2027E free cash flow $2.70bn is consensus; capex is this desk’s estimate rolled forward from $470M trailing; debt is the Aug 1, 2026 balance. Marvell remains only modestly levered — management put net debt to adjusted EBITDA at 0.27× — so the balance sheet is not the risk. The conversion line is.
Every price target in this report is this chart times a multiple
Non-GAAP diluted EPS. Gray bars are reported; green bars are forward. FY2027 and FY2028 are S&P Global consensus; FY2029 is this desk’s model, drawn with a dashed edge to say so. The ladder is steep — earnings roughly 5.9× from FY2024 to FY2029E — which is precisely why a 50× forward multiple can look either reckless or reasonable depending on how much of that ladder you believe.
Reported figures from Marvell’s fiscal-year results: FY2024 $1.51, FY2025 $1.57, FY2026 $2.84 (up 81% year on year). FY2027 quarterly run-rate so far: Q1 $0.80, Q2 $0.94, Q3 guided to $1.10 ± $0.05 — which implies roughly $1.36 in Q4 to reach the $4.20 consensus, consistent with management entering its 38–40% operating-margin band in that quarter. Watch the gray-to-green step. The whole investment case is that FY2026’s $2.84 becomes FY2028’s $6.70 without a stumble.
Nothing in the business is decelerating
Sorted slowest to fastest. Olive bars are the reported core — what Marvell actually delivered in the quarter ended Aug 1, 2026. Clay bars are the frontier lines: management’s guided growth for the businesses that are supposed to carry the next two years. If growth is intact while the stock is 36% off its high, that disconnect is the bull case in one picture — and the bear’s reply is that the June high was the mistake, not the September price.
Reported growth is Q2 FY2027 versus the year-ago quarter (Marvell earnings release, Aug 27, 2026). Guided growth is management’s own framing on the same call: FY2027 revenue ~$12bn (~+45%), FY2028 ~$18bn (~+50%), data center +60% in FY2027, interconnect above +70%, custom silicon more than doubling in FY2028 and exceeding $10bn in FY2029. Communications and other — the old carrier and enterprise business — is now just 21% of revenue and grew 10%; it no longer moves the stock either way.
Bull versus bear
Each side stated as its best advocate would put it, with the evidence attached. Both cases are built from the same August 27 earnings release — which is what makes this stock genuinely contested rather than merely volatile.
The bull case
- Guidance has been raised twice in two quarters. FY2027 went to $12bn (~+45%) and FY2028 to $18bn (~+50%), a $1.5bn raise. Companies heading into trouble do not raise the out-year.
- Growth is accelerating, not fading. Q2 revenue of $2.739bn was up 37% year on year and 13% sequentially, $39M above the guidance midpoint, with data center up 46% and non-GAAP EPS up 40%.
- Google’s warrant is structural, not promotional. 58,970,907 shares at $206.58 that vest only as Google buys silicon. Google ends up owning ~7% of Marvell if and only if Marvell delivers — incentives fully aligned.
- Custom silicon inflects next year. Management guides custom revenue to more than double in FY2028 and clear $10bn in FY2029, on top of an existing Amazon and Microsoft base.
- Connectivity is a second engine, not a legacy tail. Interconnect above +70% this year, 1.6T optics ramping, 51.2T switching more than doubling toward a ~$1bn annualised run rate, and NIC and CXL wins pathing above $2bn by FY2029.
- Operating leverage is arriving on schedule. FY2026 non-GAAP EPS rose 81%; management expects to enter a 38–40% non-GAAP operating margin band in Q4 FY2027.
- The Street did not blink. After a 10.3% drop, ratings stood at 31 strong buy, 8 buy, 5 hold and zero sells, with most post-print revisions upward. The Oct 6 investor day is a defined catalyst.
The bear case
- You are paying roughly 50× FY2027 earnings and 19.5× trailing sales. Broadcom — larger, faster-growing in AI, and guiding above $100bn of AI revenue in FY2027 — trades near 20× forward. That is roughly six times Marvell’s whole company at 40% of the multiple.
- The Google headline is a schedule, not a backlog. Only ~1.4M of 58.97M warrant shares vest automatically in year one; the rest come at ~240,000 shares per $500M of qualifying purchases, and the revenue is not expected to matter until fiscal 2029. The 10.3% drop on Aug 28 was the market reading the fine print.
- Cash conversion is moving the wrong way. Trailing FCF margin of 18.3% versus 24.2% in FY2025; 107× trailing price-to-free-cash-flow; a free cash flow yield below 1%.
- The headline P/E is flattered. Trailing net income still contains a $1.8bn pre-tax gain on the August 2025 sale of the automotive ethernet business. Strip it out and the trailing multiple is not 71× but closer to 170×.
- Mix is margin-dilutive by management’s own guidance. Non-GAAP gross margin steps down to 57.5–58.5% in Q3 from 58.9% — explicitly because custom silicon is accelerating. Growth here costs margin.
- Concentration is extreme and has bitten before. Data center is 79% of revenue across a handful of programs. The stock fell from $127.48 in January 2025 to the high $40s by April 2025 on fears about a single customer.
- Insiders are selling into it. Roughly $632M of insider sales over three months, including the CEO’s 7,500 shares at about $236 on Aug 17 under a pre-arranged plan.
Where the risks actually sit
Not a list of everything that could go wrong — a map of where each risk lives. The hot corner is deliberately occupied by multiple compression rather than anything operational, because that is the honest reading: Marvell’s business risks are mostly possible, while its valuation risk is likely. One genuine tail risk sits at the bottom right.
- Custom mix margin dilution
- Google warrant back-loading
- Multiple compression
- Celestial AI integration slip
- Broadcom price competition
- Interconnect share loss
- Rate / duration shock
- Hyperscaler capex digestion
- Customer / socket loss
- Taiwan & export controls
Multiple compression
What breaks: at ~50× FY2027 earnings, a de-rate toward Broadcom’s ~20× forward costs more than two full years of earnings growth. Nothing operational has to go wrong for this one to hurt.
Hyperscaler capex digestion
What breaks: the $18bn FY2028 guide assumes no pause. A single quarter of digestion at two or three customers resets the entire estimate ladder and the multiple at the same time.
Customer / socket loss
What breaks: data center is 79% of revenue across a handful of programs. Losing one XPU generation removes roughly a year of growth — exactly what the January-to-April 2025 collapse from $127 to the high $40s priced.
Taiwan & export controls
What breaks: Marvell is fabless and leading-edge dependent. A Taiwan supply interruption or a new control regime on advanced AI silicon severs the roadmap regardless of how many design wins are booked. Low odds, total consequence.
Custom mix margin dilution
What breaks: non-GAAP gross margin is already guided down to 57.5–58.5% from 58.9% because custom is accelerating. If custom outgrows optics faster than modelled, the 38–40% operating margin target slips right.
Google warrant back-loading
What breaks: tranches vest at ~240,000 shares per $500M of qualifying purchases through FY2033. If the purchase cadence lags, the $120bn headline simply never converts into the years investors are underwriting.
Broadcom price competition
What breaks: Broadcom holds ~70% of custom AI silicon share. A second source has limited pricing power when the primary decides to defend a socket.
Interconnect share loss
What breaks: Credo holds roughly 88% of the AEC market and Astera Labs owns retimers and CXL. Marvell’s full-distance claim gets contested lane by lane by focused specialists.
Rate / duration shock
What breaks: beta 2.25, 0.11% dividend yield, sub-1% free cash flow yield. A long-duration equity with no carry is the first thing sold when bond yields back up, as on Sep 1, 2026.
Celestial AI integration slip
What breaks: $3.25bn paid, with revenue not expected until the second half of FY2028. A slip pushes the $1bn FY2029 photonics contribution out and removes the third leg from the bull case.
The jargon, decoded
Hover the dotted terms in the metrics strip above, or scan the desk’s working definitions here.
- Custom silicon / XPU
- A chip designed for one customer’s exact workload rather than sold off the shelf. Cheaper and more power-efficient at scale, which is why hyperscalers want their own. Marvell designs them; the customer owns the deployment.
- Attach
- Silicon that sits next to the main processor rather than being it — network controllers, storage controllers, memory interfaces. Google’s warrant covers TPU-attach products. Attach is a large, real market; it is not the same as owning the compute.
- Scale-up / out / across
- Three distances data must travel: inside one server rack (up), across a data-center floor (out), and between data centers (across). Marvell sells into all three; that breadth is the moat claim.
- PAM4 DSP
- The signal-processing chip inside an optical module that squeezes more bits down a fibre. Marvell’s historical stronghold; 800G is shipping in volume and 1.6T is ramping.
- GAAP vs non-GAAP
- GAAP is the audited number. Non-GAAP strips out stock compensation and acquisition amortization. Marvell’s Q2 gap was wide: $0.33 GAAP against $0.94 non-GAAP. Both are true; neither is complete.
- Warrant & tranche vesting
- A right to buy shares at a fixed price, earned in instalments. Google’s vests at about 240,000 shares for every $500M of chips it buys — so the equity only transfers as the revenue actually arrives.
- FCF yield
- Free cash flow divided by market value. Marvell’s is 0.94% trailing: the business currently throws off under a dollar of cash a year per $100 of stock. Growth has to close that gap.
- EV / EBITDA
- Enterprise value (market cap plus net debt) over earnings before interest, tax, depreciation and amortization. About 49× here on trailing adjusted EBITDA — a growth-software multiple on a semiconductor company.
- Forward P/E
- Price divided by expected earnings for the year ahead. 50.1× on FY2027 consensus of $4.20, falling to 31.4× on FY2028’s $6.70 — the multiple gets cheap fast if the estimates hold.
- Exit multiple
- The P/E assumed at the end of a forecast. Multiply it by projected EPS to get a target price. It is the single most subjective input in this report, and it drives most of the spread between the scenarios.
- Beta
- How much a stock moves relative to the market. Marvell’s is 2.25 — a 1% market move has historically implied roughly a 2.25% move here, in both directions.
- Prob-weighted
- Each scenario’s price multiplied by its probability, summed into one expected value. It is an average of worlds, not a prediction of any one of them — the bear and bull outcomes are what actually happen.