MRVL Marvell Technology

Research status: Primary-source verified through Marvell Q2 FY2027 results and the 27 August 2026 earnings call. Decision status: operating thesis strengthened, but valuation/expectation risk remains high; keep MRVL as investigate rather than chase.

Marvell delivered Q2 FY2027 revenue of $2.739B, +37% YoY and +13% QoQ, with Data Center revenue of $2.172B, +46% YoY and +18% QoQ. GAAP/non-GAAP gross margin was 53.1%/58.9%, GAAP/non-GAAP operating margin 16.8%/36.6%, GAAP/non-GAAP EPS $0.33/$0.94, and operating cash flow $605.5M. Q3 revenue guidance is $3.15B ±5%, roughly +15% QoQ and >50% YoY at midpoint, with 52.9–53.9% GAAP gross margin and 57.5–58.5% non-GAAP gross margin.

The AI read-through is stronger than the headline beat. Management raised FY2027 revenue to roughly $12B, ~45% growth, and now expects Data Center revenue to grow about 60% in FY2027 versus ~50% previously. FY2028 company revenue is now expected around $18B, up $1.5B from the prior outlook, with Data Center revenue expected to grow more than 60% YoY and custom silicon expected to more than double.

Connectivity is broadening rather than rotating from one technology to another: 800G optical DSP demand remains strong while 1.6T ramps rapidly; 51.2T scale-out switching is on track to more than double this year; DCI, broadband analog, NPO/CPO scale-up optics and purpose-built UAL/ESUN/NVLink switches are all seeing stronger engagement. Management emphasized that 800G and 1.6T, and copper versus NPO/CPO, will coexist across different architectures rather than transition in a single cutover.

For PowerFund, the result is a strong positive demand read-through for CLS, ANET and CRDO, and a mixed-positive read-through for NVDA: networking/interconnect intensity is rising, but custom XPUs and XPU-attached silicon also validate hyperscaler alternatives to merchant GPUs. Marvell itself remains unattractive to chase after a near-tripling in 2026 and an after-hours decline despite raised forecasts, which signals that expectations are already demanding.

Investment case

Marvell owns a broad portfolio of the connectivity building blocks required as AI clusters scale: electro-optics, switch silicon, DCI, custom accelerators/attach silicon, storage and now optical scale-up through Celestial AI plus PCIe/CXL scale-up through XConn.

> The PowerFund thesis is that Marvell can become one of the few vendors monetizing multiple AI-interconnect layers simultaneously. The thesis only works per share if data-center growth remains >25–30%, custom/optical programs scale on schedule, and GAAP economics improve fast enough to offset acquisition dilution, SBC and amortization.

Verified operating baseline

• FY2026 revenue: $8.195B, +42% YoY.
• FY2026 data-center revenue: $6.100B, 74% of total and +46% YoY.
• FY2026 top-10 customers: 82% of revenue.
• One direct customer represented 14% of FY2026 revenue; one distributor represented 37%.
• Q1 FY2027 revenue: $2.418B, +28% YoY.
• Q1 operating cash flow: $638.8M, a record.
• Q1 GAAP / non-GAAP gross margin: 52.1% / 58.9%.
• Q1 GAAP / non-GAAP diluted EPS: $0.04 / $0.80.
• Q2 FY2027 guide: $2.70B ±5% revenue, 52.1–53.1% GAAP GM, 58.25–59.25% non-GAAP GM, $0.37 ±0.05 GAAP EPS, $0.93 ±0.05 non-GAAP EPS.

Acquisition bridge

Celestial AI: completed 2 Feb 2026 for total purchase consideration of approximately $3.54B, including about $1.28B cash, 24.5M Marvell shares and ~$316M fair value of contingent consideration. Up to roughly 24.4M additional shares may be issued if revenue milestones through FY2029 are met. Celestial's revenue/earnings were not material in Q1, so current valuation largely capitalizes future Photonic Fabric success.

XConn: completed 10 Feb 2026. Marvell expects initial revenue in Q3 FY2027, a ~$50M annualized run rate in Q4 FY2027, and roughly $100M FY2028 revenue. This is strategically useful for PCIe/CXL/UALink scale-up but financially small today.

KPI ladder

Valuation scenarios

Reference: $237.04 on 21 Aug 2026. Using ~899M basic shares implies ~$213B equity value before possible Celestial earnout dilution.

The following are PowerFund estimates, not company guidance.

24 months

Probability-weighted midpoint is roughly $235–245/share, close to the current mark.

60 months

The long-run opportunity is credible, but today's valuation requires near-flawless conversion of backlog/design wins into per-share earnings.

• FY2027 Q2 results on 27 August 2026 versus the $2.70B midpoint guide.
• Evidence management's expected FY2027 acceleration is visible in bookings, backlog and revenue.
• 800G/1.6T electro-optical ramps and 51.2T switching growth.
• New custom XPU/XPU-attach production ramps at multiple hyperscalers.
• Celestial Photonic Fabric customer milestones and visible FY2028/FY2029 revenue.
• XConn PCIe/CXL revenue beginning in Q3 FY2027.
• GAAP gross margin and GAAP EPS improving faster than non-GAAP presentation.
• Continued strong operating cash flow and disciplined capital return.

• Valuation: current equity value near $213B embeds years of exceptional AI growth.
• Customer concentration: top 10 customers were 82% of FY2026 revenue; large program timing can materially move results.
• Data-center concentration: 74% of FY2026 revenue came from data center.
• GAAP/non-GAAP gap: SBC and acquired-intangible amortization make headline adjusted EPS substantially higher than GAAP earnings.
• Celestial dilution: up to ~24.4M additional shares may be issued on earnout milestones, on top of 24.5M shares issued at close.
• Acquisition execution: Celestial and XConn must create product/revenue synergies, not simply more goodwill and amortization.
• Custom-silicon concentration: hyperscaler program wins can be large but binary and customer-controlled.
• Optical competition: Broadcom, Credo and optical component vendors compete across DSP/interconnect layers.
• Hyperscaler insourcing: customers increasingly design portions of their own silicon/connectivity stack.
• Factor concentration: MRVL is another high-beta AI-capex exposure for PowerFund.

Warning — freeze additions and re-underwrite

• Data-center revenue growth falls below ~25% while hyperscaler AI capex remains strong.
• Q2/Q3 revenue misses management's acceleration narrative by >5–10%.
• Non-GAAP gross margin falls below ~57% or GAAP gross margin below ~50% without a clear temporary mix reason.
• GAAP/non-GAAP EPS gap remains very large while growth slows.
• Celestial product/revenue milestones shift materially right.
• Diluted share count grows materially faster than operating earnings.

Reduce

• Data-center growth <20% for two quarters before Celestial/custom ramps compensate.
• Major custom silicon program is lost or delayed enough to reduce forward earnings power by ~15%+.
• Celestial requires significant additional dilution without matching per-share value creation.
• Cash conversion materially trails adjusted earnings for multiple quarters.

Invalidate

> Invalidate if Marvell's AI design-win growth fails to convert into rising GAAP per-share earnings—specifically, if data-center growth falls below ~15–20%, margins structurally weaken, and acquisition/SBC dilution causes per-share earnings to stagnate despite high reported non-GAAP growth.

Additional hard invalidators: sustained loss of optical/switching technology leadership, major hyperscaler customer loss, or evidence Celestial's scale-up optical architecture fails to gain production adoption.

Marvell's strength is breadth across custom compute, electro-optics, Ethernet switching, DCI, PCIe/CXL and emerging optical scale-up. Few vendors can offer all of these layers with hyperscaler relationships.

• Broadcom: strongest direct strategic comparator; enormous custom-silicon and networking scale.
• Credo: more focused and faster-growing in AEC/SerDes/connectivity, but narrower.
• NVIDIA: increasingly integrates networking tightly with its compute platform.
• Astera/XConn-class vendors: compete in scale-up, PCIe/CXL and retimer categories.

The moat is system-level IP, advanced SerDes, long qualification cycles and hyperscaler co-design relationships. The risk is that breadth masks uneven economics: a design-win-heavy portfolio can grow revenue rapidly while acquisition amortization, SBC and customer bargaining power dilute shareholder returns.

1. Rebuild valuation scenarios using the new ~$12B FY2027 and ~$18B FY2028 company revenue framework and current fully diluted share count.
2. At the 6 October Investor Day, quantify FY2028/FY2029 revenue by custom XPU, XPU-attach, optical DSP, DCI, scale-out switching, scale-up optics and scale-up switching.
3. Track the large hyperscaler warrant/program economics separately; distinguish programs already embedded through FY2028 from incremental FY2029+ opportunity.
4. Track 800G versus 1.6T optical revenue and whether 1.6T adds to, rather than cannibalizes, the established DSP base.
5. Quantify NPO/CPO scale-up optics revenue and customer concentration; verify the magnitude of the raised FY2028 scale-up optics outlook.
6. Track 51.2T switching production and Tier-1 scale-up switch engagements across UAL, ESUN and NVLink Fusion.
7. Reconcile GAAP versus non-GAAP EPS each quarter; Q2 SBC was large and the GAAP/non-GAAP earnings gap remains material.
8. Compare Marvell's expanding optical/switching stack with CRDO and ANET before allocating the next AI-networking dollar.