MRVL
Marvell Technology (MRVL): Custom Silicon, NVIDIA Validation, and a Macro-Driven Mispricing
- Sector
- Semiconductors
- Date
- Mar 25, 2026
Hold / Trim (Initiated Buy)$130 initial, surpassed
A macro discount masked a real AI infrastructure inflection. The thesis played out faster than modeled.
Thesis
Marvell was caught in a macro discount, not a comps discount. At 9.4x EV/NTM revenue against a peer median of 10.5x, the stock was pricing in two-plus years for Marvell to grow into a $120-150B valuation, despite guiding to 30% revenue growth in FY2027E. Revenue had already jumped 42% year over year, EBITDA margins were expanding from 32% toward 37% by FY2030E, and Cash flow available to all capital providers before interest, used to value the whole enterprise regardless of debt was set to grow from roughly $1.3B to over $8B across the projection window. None of that was speculative. It was sitting in reported results and conservative forward assumptions, and the market was pricing the macro environment, not the business.
Key Model Outputs
| Metric | Value |
|---|---|
| FY2027E Revenue | $10.7B (+30% YoY) |
| FY2027E EBITDA | $3.6B (34% margin) |
| FY2027E UFCF | $3.5B (32% margin) |
| WACC | 15.1% (Beta: 1.99) |
| EV / NTM Revenue | 9.4x (Peer: 10.5x) |
| EV / EBITDA (LTM) | 36.5x (Peer: 37.3x) |
| Net Debt (FY2026A) | $1.8B |
Custom Silicon, Not Commodity Exposure
Marvell’s AI exposure isn’t a rising-tide story. It co-designs custom XPUs and Application-Specific Integrated Circuit, a chip designed for one specific task rather than general-purpose computings embedded directly into hyperscaler infrastructure. The detail the market was underweighting: Marvell’s memory heritage didn’t disappear when it sold its memory business. That architecture knowledge carried into custom 2nm SRAM modules built directly into its ASICs, addressing the latency and bandwidth penalty of off-chip memory access at the design level rather than patching around it. Pure-play chip designers without that heritage don’t have the same muscle memory.
The clearest external validation arrived when NVIDIA announced a $2B strategic investment in Marvell, paired with a partnership connecting Marvell’s custom XPUs and Technology using light instead of electrical signals to move data, reducing power and increasing speed for short, high-bandwidth links into the NVIDIA's interconnect ecosystem allowing third-party chips to connect directly into NVIDIA's GPU architecture ecosystem. It also opened two monetization vectors not yet reflected in consensus: AI-RAN/5G infrastructure and optical interconnect.
Valuation Summary
| Method | Assumptions | Low | High | Mid |
|---|---|---|---|---|
| DCF, Base Case | WACC 9.5%, TGR 3.0% | $81 | $109 | $100 |
| DCF, Bull Case | WACC 8.5%, TGR 3.5% | $118 | $143 | $130 |
| DCF, Bear Case | WACC 11.5%, TGR 2.0% | $65 | $73 | $69 |
| EV/Revenue Comps | 8.0x to 12.0x FY26E Rev | $72 | $150 | $116 |
| EV/EBITDA Comps | 22x to 30x FY26E EBITDA | $81 | $136 | $110 |
| Sum-of-the-Parts | Segment EV/EBITDA | $105 | $157 | $131 |
| 52-Week Range | Market price | $47 | $103 | $75 |
| TGR / WACC | 8.5% | 9.0% | 9.5% | 10.5% | 11.0% | 11.5% | 12.0% |
|---|---|---|---|---|---|---|---|
| 2.0% | WACC 8.5%, TGR 2.0%: $102 | WACC 9.0%, TGR 2.0%: $95 | WACC 9.5%, TGR 2.0%: $88 | WACC 10.5%, TGR 2.0%: $77 | WACC 11.0%, TGR 2.0%: $73 | WACC 11.5%, TGR 2.0%: $69 | WACC 12.0%, TGR 2.0%: $65 |
| 2.5% | WACC 8.5%, TGR 2.5%: $110 | WACC 9.0%, TGR 2.5%: $101 | WACC 9.5%, TGR 2.5%: $94 | WACC 10.5%, TGR 2.5%: $81 | WACC 11.0%, TGR 2.5%: $76 | WACC 11.5%, TGR 2.5%: $72 | WACC 12.0%, TGR 2.5%: $68 |
| 3.0% | WACC 8.5%, TGR 3.0%: $119 | WACC 9.0%, TGR 3.0%: $109 | WACC 9.5%, TGR 3.0%: $100 | WACC 10.5%, TGR 3.0%: $86 | WACC 11.0%, TGR 3.0%: $80 | WACC 11.5%, TGR 3.0%: $75 | WACC 12.0%, TGR 3.0%: $71 |
| 3.5% | WACC 8.5%, TGR 3.5%: $130 | WACC 9.0%, TGR 3.5%: $118 | WACC 9.5%, TGR 3.5%: $108 | WACC 10.5%, TGR 3.5%: $92 | WACC 11.0%, TGR 3.5%: $85 | WACC 11.5%, TGR 3.5%: $80 | WACC 12.0%, TGR 3.5%: $75 |
| 4.0% | WACC 8.5%, TGR 4.0%: $143 | WACC 9.0%, TGR 4.0%: $129 | WACC 9.5%, TGR 4.0%: $116 | WACC 10.5%, TGR 4.0%: $98 | WACC 11.0%, TGR 4.0%: $91 | WACC 11.5%, TGR 4.0%: $84 | WACC 12.0%, TGR 4.0%: $79 |
Shading reflects implied share price (lighter = lower, fuller = higher). Bordered cell marks the published base case.
Blended price target: $130, ~33% upside from the $97.88 entry price.
Catalysts
NVIDIA strategic partnership and NVLink Fusion integration. AI-RAN/5G infrastructure buildout. Continued hyperscaler custom XPU design wins. Silicon photonics ramp. Continued margin expansion as Data Center mix grows.
Risks
Memory pricing pressure from Micron, Samsung, and SK Hynix compressing margins faster than revenue can offset. China export control escalation. Customer concentration. An AI capex slowdown. Broader macro and geopolitical risk, which was the actual driver of the entry-point discount.
Outcome
The thesis played out faster and more emphatically than modeled. MRVL has traded north of $300, more than 200% above the entry price and well past the original $130 target. At this level, the risk/reward that justified the original Buy no longer holds at the same magnitude, and the current stance is hold/trim rather than holding for further upside against a now-stale target. The case for revisiting Marvell going forward would need a fresh entry point and a fresh valuation, not an extension of the original price target.