Broadcom vs. Marvell: The AI Supercycle Is Big Enough for Both. Here's the Better Buy.

Broadcom expects AI semiconductor revenue of approximately $58B in fiscal 2026, $115B in fiscal 2027, and $230B in fiscal 2028, after Q3 AI revenue rose 221% year over year to $16.7B. Marvell raised its fiscal 2027 revenue outlook to $12B from $11.5B and forecasts roughly $18B in fiscal 2028, supported by Google custom-silicon work and data-center growth above 60%. The article favors Broadcom on risk-reward: it trades at 21.6x estimated fiscal 2027 earnings versus Marvell at 57.3x estimated fiscal 2028 earnings, despite Broadcom's faster projected revenue growth.

Analysis

The key differentiation is not headline AI exposure but quality of economics. AVGO converts AI demand into cash flow across custom silicon, switching and software, while MRVL is underwriting a narrower set of design wins whose returns may be diluted by customer-specific pricing and the economics of Google-linked warrants. Those warrants align incentives, but they also make reported revenue an incomplete measure of value creation: investors should focus on incremental gross margin, warrant dilution and customer concentration rather than top-line growth alone.

Google’s multi-sourcing strategy is structurally positive for the custom-ASIC ecosystem but reduces the probability that any single supplier captures a winner-take-most profit pool. MRVL can gain share in TPUs without necessarily displacing AVGO; the more relevant competitive threat is that hyperscalers use dual sourcing to negotiate lower ASIC and connectivity pricing. This favors AVGO’s broader attach rate and recurring software cash flows in a 6-18 month capex normalization scenario, whereas MRVL’s multiple remains more sensitive to any delay in TPU ramps, optical demand digestion, or a lower-than-expected gross-margin mix.

Near term, the valuation spread supports AVGO over MRVL, but AVGO’s embedded multi-year AI trajectory leaves little room for a hyperscaler capex pause. The contrarian view is that MRVL’s higher beta could outperform over 1-3 months if Google volume commitments become externally quantifiable or if connectivity revenue proves less cyclical than custom compute; absent that evidence, the market is paying for execution before margins are demonstrated. Falsify the AVGO/MRVL relative thesis if MRVL reports sustained data-center gross-margin expansion and raises FY2028 earnings materially, or if AVGO reduces its AI outlook, reports lower ASIC margins, or discloses a major customer-program delay.

Key Decisions for Investors

• Initiate a 6-12 month market-neutral pair: long AVGO / short MRVL, sized beta-neutral. Thesis is valuation convergence through superior cash conversion and lower customer-concentration risk; target 15-25% relative return. Stop/reassess if MRVL raises FY2028 EPS by more than 15% with gross-margin expansion rather than revenue-only upside.
• For long-only technology exposure, rotate incremental AI-infrastructure allocation from MRVL into AVGO ahead of the next earnings cycle; retain MRVL only as a smaller satellite position pending disclosure of Google revenue mix, warrant dilution and program-level profitability.
• Set an event-driven alert on Google TPU deployment commentary and MRVL data-center gross margin. A verified acceleration in TPU volumes plus margins—not merely design-win announcements—would justify covering the MRVL short leg and potentially reversing into a 3-6 month tactical long.
• Hedge AVGO exposure around hyperscaler capex results with short-dated SOXX puts rather than single-name puts; the principal near-term risk is correlated AI-capex de-rating across NVDA, AVGO and networking suppliers, not an isolated AVGO execution failure.