Better Custom ASIC Stock: Marvell vs. Broadcom

Broadcom is seeing steep AI custom-ASIC growth, with AI semiconductor revenue up 143% YoY to $10.8B in fiscal Q2 and management guiding to >200% growth this quarter to $16B. Broadcom also targets $100B AI revenue in fiscal 2027, versus Marvell’s outlook of +20% custom AI chip revenue this fiscal year and >100% acceleration next year. Despite Marvell stock up 209% YTD (making it expensive), Broadcom’s 12-month median price target of $525 implies 41% upside vs Marvell’s $240 target implying a 14% decline.

Analysis

The tradeable signal here is not “AI semis are good,” but that custom silicon is starting to separate the winners by operating leverage and customer lock-in. Broadcom has already crossed the threshold where incremental AI revenue is meaningfully accretive to margins and sentiment, while Marvell still looks like a later-cycle execution story that needs proof of durable design-win conversion. That makes AVGO the cleaner compounder and MRVL the more fragile multiple, even if both ride the same structural demand wave.

The second-order effect is on GPU share, not just on the two named names: if hyperscalers keep shifting inference and fixed-function workloads to ASICs, NVDA’s pricing power should remain intact in training, but the mix tailwind to its overall AI growth can fade at the margin over the next 6-18 months. Supply-chain beneficiaries are also likely to be the advanced packaging / networking ecosystem rather than the chip designers alone, because custom chips pull through memory, substrates, and high-speed interconnect. The market is likely underpricing how concentrated customer risk is for MRVL; one delayed program can push out the whole earnings ramp.

Contrarian view: the consensus may be overconfident that high growth automatically means higher upside. For MRVL, the issue is not TAM, it is whether the current valuation already discounts the 2027-2029 revenue curve before it is visible in reported numbers. For AVGO, the main risk is not competition but customer bargaining power; if a handful of hyperscalers internalize more silicon, Broadcom’s long-term mix could become more lumpy than the street assumes.

Near-term, this is mostly a relative-value setup, not a broad sector long. The next 1-3 month catalysts are quarterly AI revenue prints, hyperscaler capex commentary, and any new custom-chip design wins; the thesis breaks if AVGO’s AI growth decelerates materially or if MRVL surprises with a step-change in booked demand. Over 6-18 months, the key variable is whether ASIC adoption stays confined to inference or expands into more of the AI stack.

Key Decisions for Investors

• Enter a beta-adjusted long AVGO / short MRVL pair trade over the next 1-2 weeks; thesis is valuation convergence as Broadcom’s growth visibility converts to multiple support while Marvell’s premium remains execution-dependent. Risk/reward is attractive if MRVL fails to raise forward AI revenue guidance at the next print.
• Accumulate AVGO on any 3-5% post-earnings pullback rather than chasing strength; use a 6-12 month horizon targeting continued multiple re-rating if AI revenue growth stays above 100% and management reaffirms the 2027 framework.
• If you want convexity, use AVGO call spreads six to nine months out instead of outright stock; the market is still underestimating the upside if the company proves it can sustain >$15B quarterly AI run-rate without margin dilution.
• Do not short NVDA outright on this theme, but watch for relative underperformance versus AVGO if custom ASIC adoption broadens in inference; that would be a signal to rotate within AI semis rather than exit the theme.
• Set a watch item on MRVL: if the next quarterly update shows any delay in custom silicon revenue inflection or softer gross margin mix, use it as a catalyst to add to the AVGO/MRVL pair rather than betting on MRVL mean reversion.