Better Buy After the Semiconductor Sell-Off: Marvell or Broadcom?

The key takeaway is that the AI infrastructure trade is starting to bifurcate between narrative premium and cash-generating scale. Marvell’s endorsement-driven rerating looks fragile because its upside depends on a narrower set of hyperscale program wins, while Broadcom’s mix is broadening from “AI optionality” into a demonstrable earnings engine with far better self-funding capacity. That matters because in the next 1-2 quarters, the market will likely reward the name that can absorb disappointment without multiple compression, not the one with the most exciting headline beta.

Broadcom’s second-order advantage is that its cash flow gives it more room to keep winning sockets through packaging, networking, and software adjacency without needing perfect execution. If AI capex keeps broadening beyond the first-wave GPU clusters, Broadcom should capture more of the plumbing spend than Marvell, whose earnings sensitivity is higher to a handful of design cycles and customer concentration. The flip side is that any slowdown in hyperscaler order timing will hit MRVL first and harder, because the stock is priced like a durable monopoly rather than a cyclical supplier.

The contrarian miss is that the current debate may underweight duration risk for both names. Even if AI demand remains healthy, the next leg of returns will likely come from margin durability and diversification, not from revenue growth alone; on that score, AVGO has the cleaner setup. MRVL could still outperform on short bursts of sentiment, but it now needs flawless follow-through, while AVGO merely needs to keep converting AI demand into cash at scale.