MRVL DCF / Sum-of-Parts Valuation
Valuation date: 2026-04-29
Spot: $154.83 (close 2026-04-28, source: companies/mrvl/data/STOCK_PRICE_DATA.json)
Diluted share count: ~860M (10-K FY2026 + Celestial AI 27.2M issued Feb 2 2026 + XConn ~2.1M issued Feb 10 2026; pre-Celestial earnout)
Market cap: ~$133.1B
Cross-references: segment revenue · earnings calls · comps valuation · comparable transactions · balance sheet · credit market positioning · valuation framework
1. Methodology choice
Marvell’s value is split between (a) a fast-ramping but margin-dilutive custom AI silicon (xPU/NIC) franchise that mostly didn’t exist 24 months ago, (b) a PAM4 / coherent DSP optical-interconnect franchise (Inphi-derived, capital-efficient), (c) a switching/DPU stack (Innovium TERALYNX, OCTEON), and (d) declining legacy Storage / Carrier / Enterprise lines now consolidated into “Communications & Other.” A pure consolidated DCF blurs the dilution math — gross-margin compression from custom silicon mix is a load-bearing variable that can’t be back-solved from a single-name discount rate. A pure SOTP, on the other hand, double-counts shared OpEx (R&D pooled across xPU + DSP + switch) and ignores the cross-IP synergies that justify keeping the businesses under one roof. The model uses both methods and triangulates: a five-year segment-level DCF (FY26 actuals → FY30E) for the going concern, plus an EV/Sales SOTP at peer-anchored multiples for FY27E and FY28E. Where they diverge, we reconcile the gap to either the OpEx-allocation or the multiple-vs-DCF-implied-ROIC mismatch.
2. Segment revenue model FY26 → FY30E
2.1 Segment build (USD millions; fiscal years ending late Jan / early Feb)
| Sub-segment | FY26A | FY27E | FY28E | FY29E | FY30E | Source / basis |
|---|---|---|---|---|---|---|
| Datacenter | 6,100 | 8,540 | 12,810 | 16,012 | 18,895 | Mgt: FY27 +40% YoY; FY28 +50% YoY; ✓ Q4 FY26 call 2026-03-05 |
| · Custom AI silicon (xPU + NIC) | 1,500 | 1,900 | 3,800 | 5,320 | 6,650 | Mgt: FY27 +20%+; FY28 “double” vs. FY27; ✓ Q4 FY26 call |
| · Optical interconnect (PAM4 + 1.6T DSP + coherent + AECs) | 2,400 | 3,360 | 4,700 | 5,640 | 6,200 | Inphi franchise; mgt “1.6T DSP in production”; LightCounting 800G→1.6T 2026-2028 doubling cadence (comps valuation) [Confidence: estimate — basis: Inphi ~$1.5B FY24 baseline + PAM4 unit doubling 2026/2027 per Murphy Q3 FY26] |
| · Teralynx switch + Tomahawk-class | 800 | 1,120 | 1,500 | 1,800 | 2,000 | Innovium ramp; CY2024-2026 51.2T cycle [Confidence: estimate — basis: 10-15% hyperscale switching share cited in comparable transactions] |
| · OCTEON DPU + smartNIC | 700 | 850 | 1,000 | 1,150 | 1,300 | Pensando comp baseline (~$50M in 2022 → DPU TAM growing); OCTEON installed base [Confidence: estimate — basis: management language “custom NIC program underway” Q4 FY25] |
| · CXL / Tanzanite / memory fabric / other | 700 | 1,310 | 1,810 | 2,102 | 2,745 | Includes Celestial AI ramp (H2 FY28: ~$200M; FY29: $700M; FY30: $1.2B per Murphy Q3 FY26 ✓) and Polariton (no revenue through FY28; modest FY29-FY30) and residual “other” [Confidence: estimate — basis: Celestial revenue framework Q3 FY26 call + bottom-up reconcile to mgt FY28 $15B total] |
| Communications & Other | 2,094 | 2,250 | 2,300 | 2,255 | 2,210 | Stabilizing; mgt mix-shift narrative (DC 75% → 80%+ in FY28). segment revenue |
| · Carrier infrastructure | 600 | 700 | 720 | 700 | 680 | Cycle recovery off Q3 FY25 trough [Confidence: estimate — basis: Q2 FY26 +71% YoY rebound, normalizing to flat] |
| · Enterprise networking | 700 | 760 | 800 | 800 | 790 | Steady book of business [Confidence: estimate — basis: Q3 FY26 disclosure ranges] |
| · Storage (sunset) | 550 | 500 | 450 | 415 | 390 | Long-term sunset; storage sunset cites flat-to-down trajectory. ✓ |
| · Consumer / other | 244 | 290 | 330 | 340 | 350 | Residual; gaming (Sony/MSFT consoles) + IoT [Confidence: estimate — basis: Q4 FY26 mix walk] |
| Auto Ethernet (divested) | (~150 partial yr) | 0 | 0 | 0 | 0 | Sold to Infineon Aug 14 2025 for $2.5B cash; ✓ Q2 FY26 announce + 10-K Note 4 |
| TOTAL REVENUE | 8,194 | 10,790 | 15,110 | 18,267 | 21,105 | Mgt: FY27 ~$11B; FY28 ~$15B ✓ Q4 FY26 call |
Implied vs. management framing:
- FY27 model $10.79B vs. mgt “approaching $11B” → in line (slight conservatism vs. mid-point) ✓
- FY28 model $15.11B vs. mgt “approximately $15B” → in line ✓
- FY29-FY30 are analyst extrapolations: deceleration to +21% then +16% reflects (a) custom silicon “ramps once-then-grows” vs. doubles every year, (b) optical 1.6T → 3.2T cycle moderating, (c) Celestial AI revenue stepping up but from low base. [Confidence: estimate — basis: no management framing beyond FY28]
2.2 Custom AI silicon ramp curve (load-bearing assumption)
| FY | Custom xPU + NIC ($M) | YoY | Notes |
|---|---|---|---|
| FY26A | 1,500 | n/a | ”Scaled from $0 to $1.5B” — Murphy Q4 FY26 ✓ |
| FY27E | 1,900 | +27% | “At least 20%” mgt floor ✓; market split AWS Trainium 2/3, Microsoft Maia, Google TPU 8i / MPU, Meta Arke at staggered ramps |
| FY28E | 3,800 | +100% | “Forecast to double from FY27 levels” — Murphy Q3 FY26 ✓ |
| FY29E | 5,320 | +40% | Trainium 3 mature; Maia 200 ramping; assumes no major share-loss event |
| FY30E | 6,650 | +25% | Steady-state hyperscaler refresh; new sockets compensate maturing ones |
Per-customer split (FY27E qualitative; explicit allocation withheld by mgt):
- AWS Trainium 2 (production) + Trainium 3 (ramp): ~40-50% of custom xPU revenue ◐ (Stifel Mar 2026 customer-cliff thesis flagged share loss to Alchip on T3/T4 — contested, ungated)
- Microsoft Maia 100 (production) + Maia 200 (ramp): ~25-30% ◐
- Google MPU + TPU 8i companion: ~10-15% ◐
- Meta Arke + others: balance ◐
- Source: triangulated from earnings calls “four major hyperscalers, two in production, third NIC” + analyst coverage field checks +
03_ecosystem/company_profiles/. - [Confidence: estimate — basis: management refuses to name customers; mix inferred from public disclosures]
2.3 Polariton / Celestial AI integration timing
- Celestial AI (closed Feb 2 2026): per Murphy Q3 FY26, no revenue FY26/FY27, $200M (H2 FY28 ramp), ~$500M run-rate Q4 FY28, $1.0B+ run-rate Q4 FY29. Modeled as ~$200M FY28 / ~$700M FY29 / ~$1.2B FY30. ✓
- XConn (closed Feb 10 2026): CXL switch IP; folds into “CXL/Other” line. No standalone disclosure; estimate $50-100M FY27, ramping with CXL spec adoption [Confidence: estimate — basis: 10-K Note 16 silent on revenue]
- Polariton (announced Apr 22 2026): IP/team acquisition; plasmonic modulator at >3.2T port speeds; no revenue through FY28; embedded in optical interconnect roadmap from FY29 onward [Confidence: estimate — basis: pre-revenue research-stage IP per comparable transactions]
3. Margin model
3.1 Gross margin trajectory (non-GAAP; segment + blended)
| Sub-segment | Approx. GM | Source / basis |
|---|---|---|
| Custom AI silicon | 50-55% | Mgt: “structurally lower” than blended; ✓ Q2 FY26 Meintjes |
| Optical interconnect (DSP / coherent) | 65-70% | Inphi historical 70%+; FY26 mix-down ✓ |
| Switching (Teralynx) | 55-58% | Hyperscale ASIC pricing; modest |
| OCTEON DPU | 60-62% | Mature line |
| Celestial / CXL / new IP | 55-60% pre-scale → 60%+ at scale | Mgt “accretive to overall data center” ✓ Q3 FY26 Murphy |
| Carrier / Enterprise | 60-62% | Stable mature |
| Storage | 55-58% | Declining HDD/SSD controllers |
| Consumer | 30-40% | Console / IoT |
Blended non-GAAP GM forecast:
| FY | Non-GAAP GM | Driver |
|---|---|---|
| FY26A | 59.5% | Q4 FY26 59.0% exiting; full-year ~59.5% ✓ |
| FY27E | 59.0% | Mgt: “Q1 FY27 58-60%; full year 59-60%” ✓ Q4 FY26 |
| FY28E | 58.5% | Custom mix grows (xPU 1,900 → 3,800), pulling blend down ~50bps |
| FY29E | 58.7% | Celestial ramp partially offsets (60%+ accretive at scale) |
| FY30E | 59.0% | Optical 3.2T cycle and Celestial scale lift mix |
3.2 OpEx / SBC / 3-acquisition cost-absorb
FY26A baseline (quarterly trend):
- Non-GAAP OpEx run-rate Q4 FY26: ~$285M/quarter → ~$1,140M annualized
- FY26 R&D: ~$1.85B (GAAP); SG&A: ~$0.45B (GAAP)
- Non-GAAP OpEx FY26: ~$1.95-2.0B
Forward OpEx (non-GAAP):
| FY | Non-GAAP OpEx ($M) | YoY | Acquisition layer |
|---|---|---|---|
| FY27E | 2,200 | +10% | + $50M Celestial annual run-rate ✓ Q3 FY26 Meintjes; + ~$30M XConn; + ~$20M Polariton [Confidence: estimate] |
| FY28E | 2,400 | +9% | Full Celestial integration; Polariton design ramp |
| FY29E | 2,580 | +7.5% | Steady; OpEx/Rev declines as revenue grows |
| FY30E | 2,720 | +5.4% | Low-teens R&D % of revenue |
SBC dilution: Marvell’s SBC has historically run ~$700-900M annually (FY25-FY26 GAAP). Diluted share count step-ups: ~27.2M from Celestial close + up to 27.2M earnout (1/3 trigger at $500M cumulative Celestial revenue by FY29; full at $2.0B cumulative — full earnout unlikely on management’s stated ramp), ~2.1M from XConn. Modeled diluted share growth: ~860M FY27 → ~877M FY28 → ~885M FY29 → ~890M FY30 (assuming aggressive buybacks offset SBC run-rate at ~50% from $1.0-1.2B annual buyback). ◐
3.3 Operating margin → EBITDA → FCF
| FY | Revenue | Non-GAAP GM | Non-GAAP OpEx | Non-GAAP OI | Non-GAAP OM | EBITDA est. | EBITDA mgn | FCF est. | FCF mgn |
|---|---|---|---|---|---|---|---|---|---|
| FY26A | 8,194 | 59.5% | 1,950 | 2,925 | 35.7% | 3,300 | 40.3% | 1,750 | 21.4% |
| FY27E | 10,790 | 59.0% | 2,200 | 4,166 | 38.6% | 4,540 | 42.1% | 2,250 | 20.9% |
| FY28E | 15,110 | 58.5% | 2,400 | 6,439 | 42.6% | 6,920 | 45.8% | 3,800 | 25.1% |
| FY29E | 18,267 | 58.7% | 2,580 | 8,144 | 44.6% | 8,720 | 47.7% | 4,950 | 27.1% |
| FY30E | 21,105 | 59.0% | 2,720 | 9,732 | 46.1% | 10,360 | 49.1% | 6,100 | 28.9% |
Non-GAAP EPS bridge:
| FY | OI | Net interest | Pretax | Tax @ 13% (mgt non-GAAP rate FY26) ✓ | Net Inc | Diluted shares | EPS |
|---|---|---|---|---|---|---|---|
| FY26A | 2,925 | (160) | 2,765 | 360 | 2,406 | 850 | $2.83 (vs. $2.84 reported ✓) |
| FY27E | 4,166 | (240) | 3,926 | 510 | 3,416 | 860 | $3.97 |
| FY28E | 6,439 | (240) | 6,199 | 805 | 5,394 | 877 | $6.15 |
| FY29E | 8,144 | (220) | 7,924 | 1,030 | 6,894 | 885 | $7.79 |
| FY30E | 9,732 | (200) | 9,532 | 1,239 | 8,293 | 890 | $9.32 |
Mgt FY28 framing: “non-GAAP EPS well over $5” ✓ Q4 FY26. Model FY28 $6.15 = on guidance, modestly above mid-point. Implies operating leverage on track.
FCF reconciliation (FY27E):
- Non-GAAP OI: $4,166M → less D&A swap ~$0 (already non-GAAP) → less non-GAAP tax (13%): ~$540M → less interest ~$240M → less capex ~$220M (fab-light per Murphy Q4 FY26 ✓) → less working-capital: ~$300M (revenue growth) → less SBC paid in cash equivalent: $0 (excluded from non-GAAP) ≈ $2,250M FCF.
- Note: GAAP FCF is structurally lower by SBC ($800-900M run-rate); non-GAAP definition used for valuation per sell-side convention.
4. DCF
4.1 WACC build
| Component | Input | Source |
|---|---|---|
| Risk-free rate (10Y UST) | 4.345% | Pricing benchmark for MRVL 5.300% 2036 notes; 4.125% UST 2036-02-15 priced 98-08 / 4.345% YTM ✓ (credit market positioning §1a) |
| Equity risk premium | 5.5% | Damodaran Jan 2026 implied ERP for US large-cap [Confidence: estimate — basis: Damodaran Jan 2026 update] |
| Beta (β) | 1.5 | STOCK_PRICE_DATA.json ◐ stub-default; sell-side range 1.4-1.7 (Bloomberg Apr 2026) ◐ |
| Cost of equity (CAPM) | 4.345% + 1.5 × 5.5% = 12.6% | |
| Pre-tax cost of debt (blended) | 5.10% | Weighted: $1.0B 5.300% (2036) + $0.5B 5.450% (2035) + $0.5B 4.750% (2030) + $0.75B 2.450% (2028) + $0.75B 2.950% (2031) + $0.5B 5.750% (2029) + $0.5B 5.950% (2033). ✓ from credit market positioning §1 |
| Tax rate (effective) | 13% non-GAAP / 18% GAAP statutory | FY26 10-K effective rate (post-Pillar Two impact, Bermuda-redomicile-Delaware) ◐ |
| After-tax cost of debt | 5.10% × (1-0.18) = 4.18% | |
| Capital structure (E / D) | $133.1B / $4.5B → 96.7% E / 3.3% D | Spot 2026-04-28 + 10-K Note 7 ✓ |
| WACC | 0.967 × 12.6% + 0.033 × 4.18% = 12.3% |
Sensitivity: if β = 1.4 → WACC 11.8%; if β = 1.7 → WACC 13.4%.
4.2 Terminal value — exit multiple AND perpetuity growth
Method A — Exit-multiple (FY30E EBITDA × 18x):
- FY30E EBITDA: $10,360M
- Exit multiple: 18x (mid-point of AVGO 19x, NVDA 22x, COHR 12x — comps valuation peer set; reasonable for “AI-infrastructure incumbent in maturation phase”)
- Terminal EV: 10,360 × 18 = $186.5B
- Discount back 5 years at 12.3%: $186.5B / (1.123)⁵ = $104.8B PV
Method B — Perpetuity growth (Gordon):
- FY30E FCF: $6,100M
- Long-term growth rate: 4% (semis maturing; above GDP given AI infrastructure secular demand) [Confidence: estimate]
- Terminal FCF: 6,100 × 1.04 = $6,344M
- Terminal value: 6,344 / (0.123 - 0.04) = $76,434M
- Discount back at 12.3%: $76,434 / (1.123)⁵ = $42.9B PV
TV reconciliation: Exit-multiple gives ~$105B PV; perpetuity ~$43B PV. Gap reflects (a) the 18x exit multiple bakes in continued above-GDP growth beyond the explicit forecast, (b) perpetuity 4% is conservative for a still-growing custom-silicon platform. Average: $74B TV PV.
4.3 Enterprise value & per-share
| Item | $M |
|---|---|
| Sum of FY27-FY30 discounted FCF | 11,580 |
| · FY27E FCF $2,250 / 1.123 = | 2,003 |
| · FY28E FCF $3,800 / 1.123² = | 3,011 |
| · FY29E FCF $4,950 / 1.123³ = | 3,491 |
| · FY30E FCF $6,100 / 1.123⁴ = | 3,829 |
| (sum: $12,334M before rounding) | 12,334 |
| Terminal value PV (avg of two methods) | 73,850 |
| Enterprise Value (DCF) | 86,184 |
| (+) Cash | 2,700 |
| (–) Total debt | (4,500) |
| Equity value | 84,384 |
| Diluted shares (FY27 mid-year) | 865M |
| DCF per-share fair value (base) | $97.6 |
Bull / base / bear sensitivity grid (per-share, FY27 mid-year diluted):
| Scenario | Revenue FY30 | GM% blend | Exit multiple | DCF $/share |
|---|---|---|---|---|
| Bear | $17.0B (FY28 $13B / no FY29-30 acceleration) | 56% | 14x | $58 |
| Base | $21.1B | 58.7% | 18x | $98 |
| Bull | $25.5B (xPU $4.5B/$8B/$11B/$14B; Celestial $1.5B FY30) | 60% | 22x | $148 |
Note: even the base DCF ($98) is materially below spot ($154.83). DCF discipline assigns no value to “optionality” beyond the explicit FY30E horizon and applies a high WACC to a high-beta name. The market is pricing FY28+ acceleration that DCF only rewards at the bull-case.
4.4 Per-share output by sensitivity to GM and revenue (FY27 P&L pivot)
DCF $/share at base WACC 12.3%, exit multiple 18x, varying FY27 revenue and FY28 GM:
FY28 GM%
56% 58% 60% 62%
FY27 Rev
$9.0B $66 $74 $81 $88
$10.0B $76 $84 $92 $100
$11.0B $86 $95 $103 $112
$12.0B $96 $105 $114 $124
Implied: even at $12B FY27 revenue with 62% GM (well above mgt frame), DCF base at 18x exit only reaches $124/share — still below spot.
5. Sum-of-parts
5.1 Per-segment EV/Sales (applied to FY27E)
| Segment | FY27E Revenue ($M) | EV/Sales | EV ($M) | Comp basket |
|---|---|---|---|---|
| Custom AI silicon (xPU + NIC) | 1,900 | 13x | 24,700 | AVGO custom-silicon division (implied 13-15x within AVGO blend) ◐; AMD-Pensando 35x fwd (pre-revenue, less applicable); Inphi precedent 14.6x for analogous DC silicon IP. 13x = high end of mature comps but below AMD-Pensando peak. |
| Optical interconnect (PAM4 / coherent / 1.6T) | 3,360 | 11x | 36,960 | Inphi M&A 14.6x ✓; COHR public 6-7x EV/Sales ✓ (comps valuation); ANET 9-10x EV/Sales (Apr 2026) ◐. 11x = mid-point reflecting MRVL’s DSP IP moat. |
| Switching / Teralynx | 1,120 | 6x | 6,720 | ANET 9-10x; Cisco 4-5x; Marvell switch is sub-leader vs. AVGO Tomahawk → discount to ANET. 6x. |
| OCTEON DPU / NIC | 850 | 9x | 7,650 | AMD-Pensando $1.9B / ~$50M = 35x at deal (forward); applied to producing revenue → 8-10x. 9x. |
| CXL / Tanzanite / Celestial / Other | 1,310 | 14x | 18,340 | ALAB (Astera Labs) 18-22x EV/Sales (Apr 2026, pure CXL pure-play) ◐; Celestial M&A 65x fwd (sub-set valuation); Ayar Labs Series E $3.75B at zero rev (private) ◐. 14x = significant haircut to ALAB given mix of mature+early. |
| Datacenter total | 8,540 | ~11x avg | 94,370 | |
| Carrier infrastructure | 700 | 4x | 2,800 | Cirrus Logic 2-3x; QCOM 4-5x; legacy comm silicon. 4x. |
| Enterprise networking | 760 | 4x | 3,040 | Same comp set. |
| Storage (sunset) | 500 | 2x | 1,000 | WDC/STX storage divisions trade 1-2x; declining → discount. 2x. |
| Consumer / other | 290 | 3x | 870 | Mid-tier analog. 3x. |
| Communications & Other total | 2,250 | ~3.4x avg | 7,710 | |
| Total Enterprise Value (SOTP) | 10,790 | ~9.5x blended | 102,080 |
5.2 Equity value & per-share
| Item | $M |
|---|---|
| SOTP Enterprise Value | 102,080 |
| (+) Cash | 2,700 |
| (–) Debt | (4,500) |
| (–) Celestial earnout fair value (50% PV expected) | (1,500) |
| SOTP equity value | 98,780 |
| Diluted shares | 865M |
| SOTP per-share (base) | $114 |
5.3 Cross-check vs. DCF
- DCF base: $98/share
- SOTP base: $114/share
- Spot: $155
- Reconciliation: SOTP > DCF by ~16%. The gap reflects (a) DCF embeds a 12.3% WACC haircut on every cash flow whereas SOTP applies multiples set by markets that already implicitly use lower discount rates for high-growth segments, (b) DCF terminal-multiple at 18x is below the 18-22x weighted-average SOTP multiple, (c) SOTP credits the “platform value” of optical-DSP-leadership scarcity that DCF prices only at run-rate FCF. Both are below spot. The market’s $155 ⟹ ~$140B equity value implies SOTP multiples ~14x on FY27 (vs. 9.5x model) or DCF growth assumptions ~25% beyond the explicit horizon. For the bull case, those are achievable; for base, market appears stretched 25-35%.
5.4 SOTP applied to FY28E (forward)
If we roll forward to FY28E (a more aggressive base), SOTP at slightly compressed multiples (markets de-rate as growth moderates):
| Segment | FY28E Revenue | EV/Sales | EV |
|---|---|---|---|
| Custom AI | 3,800 | 11x | 41,800 |
| Optical | 4,700 | 9x | 42,300 |
| Switching | 1,500 | 5x | 7,500 |
| OCTEON | 1,000 | 7x | 7,000 |
| CXL/Celestial/other | 1,810 | 12x | 21,720 |
| Comms & Other | 2,300 | 3x | 6,900 |
| Total EV (FY28 SOTP) | 15,110 | ~8.4x | 127,220 |
| (+) Cash growth | 4,500 | ||
| (–) Debt | (4,500) | ||
| (–) Earnout PV | (1,500) | ||
| Equity | 125,720 | ||
| Diluted shares | 877M | ||
| FY28 SOTP / sh (1-yr forward) | $143 |
PV-back-to-2026 at 12.3%: $143 / 1.123 = $127/share = “1-yr forward fair value.”
6. Stress tests
| Stress | Scenario detail | FY28 revenue impact | FY28 EPS impact | DCF $/sh impact (base) | SOTP $/sh impact |
|---|---|---|---|---|---|
| Custom AI silicon −25% | xPU lifetime ramp delayed / Trainium 3 share to Alchip; FY28 xPU $2.85B vs. $3.80B base | −$950M (−6%) | −$0.95 EPS (−15%) | −$11 (−11%) → $87 | −$13 (−11%) → $101 |
| NVIDIA in-sources networking | NVIDIA reclaims InfiniBand/Spectrum-X share previously open to MRVL DSP/optical (lower-end exposure); FY28 optical $4.0B vs. $4.7B | −$700M (−5%) | −$0.65 EPS | −$8 → $90 | −$10 → $104 |
| China-export tightening | Innolight + Eoptolink (China-based optical module customers using MRVL DSP) restricted; ~10-15% of optical at risk → $500-650M FY28 | −$575M (−4%) | −$0.55 EPS | −$7 → $91 | −$8 → $106 |
| Polariton integration delays | >3.2T port roadmap pushed FY30→FY31; modest near-term (FY28-29 unaffected); FY30 optical −$300M | (−$300M FY30) | −$0.20 EPS FY30 | −$3 → $95 | flat near-term → $114 |
| GM compression −150bps | Aggressive Broadcom DSP / coherent pricing pulls blended GM to 57% in FY28-29 | (no rev impact) | −$0.85 EPS | −$10 → $88 | −$11 → $103 |
| Combined “soft bear” (xPU −15% + GM −100bps + China −5%) | −$1.4B FY28 | −$1.40 EPS | −$24 → $74 | −$26 → $88 | |
| Combined “hard bear” (xPU −25% + GM −150bps + China −10% + NVIDIA in-source) | −$2.5B FY28 | −$2.50 EPS | −$40 → $58 | −$40 → $74 |
Asymmetry: the “hard bear” scenario (multiple tail risks compounding) lands at $58-74/share — a ~50% drawdown from spot. This is the load-bearing input to position-sizing for a long-MRVL thesis. By contrast, the “single-stress” scenarios are absorbable in a 5-15% drawdown range, indicating the stock is sensitive but not fragile to any individual risk factor.
7. Conclusion: implied fair-value range
| Case | DCF $/sh | SOTP $/sh | Blended (50/50) | Rationale |
|---|---|---|---|---|
| Bear | $58 | $74 | $66 | Custom xPU underperforms ramp; GM compresses; multiple stress drivers compound. Implied at 11-12x FY28 EPS. |
| Base | $98 | $114 | $106 | Mgt FY27 ~$11B / FY28 ~$15B realized; GM 58.5%; 18x exit / 9.5x SOTP blended. Below spot ~32% — market pricing higher growth-after-FY30 acceleration. |
| Bull | $148 | $165 | $157 | xPU doubles each FY27→28→29; optical 1.6T→3.2T cycle pulls ahead; GM stable at 59-60%; 22x exit / 11x SOTP blended. Roughly at spot. |
| Spot | — | — | $155 | Market implies ~bull-case fundamentals + base-case multiples, OR base-case fundamentals + bull-case multiples. |
One-line rationale per case:
- Bear $66: Custom AI silicon ramp slows + GM erodes; multiple compresses as growth moderates. Triggered by (i) Trainium 3 share loss to Alchip (Stifel customer-cliff thesis bears out) AND (ii) Microsoft Maia rotation to AVGO AND (iii) optical China-export tightening. Probability ~20%.
- Base $106: Management FY28 ~$15B framing realized at face value; GM stable at 58.5%; SOTP/DCF average compounds at ~12% from current; multiples revert to historical median. Probability ~50%.
- Bull $157: All current design wins ramp on schedule + Celestial AI delivers $1.5B FY30 + Polariton enables 3.2T transition + multiples sustain at AI-incumbent peer levels. Probability ~30%.
Probability-weighted fair value: 0.2 × $66 + 0.5 × $106 + 0.3 × $157 = $13.2 + $53.0 + $47.1 = $113/share ≈ −27% downside vs. spot.
This positions MRVL as fairly-to-richly priced under base-case fundamentals; the bull scenario must obtain to justify spot. The upside is path-dependent on (a) custom xPU growth not just hitting “double in FY28” but accelerating in FY29-30, (b) GM holding 59%+ despite custom mix, (c) Celestial AI execution. The downside floor (~$66) is a 50% drawdown from spot — significant tail risk if hyperscaler concentration bites.
Triangulation: Note this output ($113 prob-weighted) sits closer to the sell-side consensus ($120 PT per comps valuation) than to spot ($155), suggesting the credit market and consensus sell-side share the “fairly-to-richly priced” view. Single-strategy crowding risk noted in credit market positioning §7 applies: “high P/E + tight credit spreads = synchronized repricing risk on a 5-10% miss.”
8. Cross-references
- segment revenue — historical segment trend underpinning the 2.1 build
- earnings calls — verbatim mgt FY27/FY28 framing (Q3/Q4 FY26 calls) used for the 2.2 ramp curve
- quarterly trend — historical OpEx and FCF base for §3.2 / §3.3
- comps valuation — peer trading multiples for §5.1
- comparable transactions — M&A multiples (Inphi 14.6x, Acacia 9.7x, Pensando 35x, Mellanox 5.2x) for §5.1 calibration
- balance sheet — debt/cash for EV→equity bridge
- credit market positioning — §1a 5.300% senior-notes coupon, blended cost-of-debt input to §4.1
- valuation framework — investment-thesis lens
- m and a history — cumulative ~$22B M&A spend underpins SOTP segment count
- storage sunset — storage revenue trajectory
- celestial ai (if present) — Celestial revenue ramp framework
- polariton deal terms (if present) — Polariton terms
Author note on confidence: All segment revenue inputs through FY27 anchor to either management framing (✓) or earnings-call disclosure (✓ to ◐). FY28 revenue is mgt-framed (✓ “approximately $15B”). FY29-FY30 are analyst extrapolation [Confidence: estimate]. Margin assumptions are anchored to recent quarters (✓) with forward trends modeled. WACC inputs: Rf and cost-of-debt are primary-sourced from MRVL bond pricing (✓); β is stub-default 1.5 from STOCK_PRICE_DATA.json ◐ (sell-side range 1.4-1.7); ERP is Damodaran convention (estimate). The model output should not be cited as a price target; it is a triangulation framework. Underwriting an MRVL position requires accepting either the “growth persists past FY30” assumption (validates DCF-bull) or a 12-15x FY28 multiple expansion (validates SOTP-bull).