⚠ The AI Capex Cycle and the MultipleHigh threat

Marvell trades at roughly $227 billion — about 87 times trailing earnings and some 24 times sales¹ — after the shares more than tripled in a year. That price embeds two assumptions: that hyperscaler capital expenditure keeps rising, and that Marvell keeps winning a growing share of it.

The first assumption is outside the company's control entirely. Data center is about 74% of revenue², and the customers funding it are a handful of firms whose combined capital budgets have been growing at rates no infrastructure cycle has sustained indefinitely. Semiconductors are a cyclical industry that has repeatedly convinced itself otherwise, and Marvell's own history contains a telecom trough deep enough to hold carrier revenue down for years.

The second is where the multiple becomes uncomfortable. Marvell must not merely participate in the build-out but keep gaining — a company holding 20-25% of custom design share against a rival with roughly 70%³ is priced as though the gap narrows. Any quarter in which a program slips, a socket is lost or capital expenditure guidance softens will be repriced violently, because at more than eighty times trailing earnings there is no margin for disappointment.

The number to watch is the sequence of data-center revenue growth rates rather than any single quarter, alongside hyperscaler capital expenditure guidance. Deceleration in the customers' spending plans arrives before deceleration in Marvell's revenue, and both arrive before the multiple adjusts. It is worth remembering that the same fiscal 2026 that produced record revenue also produced a GAAP result flattered by a $2.5 billion divestiture gain⁴ — reported profitability here is younger and less settled than the share price implies.