Marvell Technology is executing Broadcom's custom silicon playbook, targeting $10 billion in AI chip revenue by fiscal 2029.
Marvell Technology is executing Broadcom's custom silicon playbook, targeting $10 billion in AI chip revenue by fiscal 2029.

Marvell Technology is executing Broadcom's custom silicon playbook, targeting $10 billion in AI chip revenue by fiscal 2029.
Marvell Technology's data center revenue has swelled to 76% of total sales as the chipmaker locks in more than 50 custom AI design opportunities across over 10 hyperscaler customers, mirroring the strategy that turned Broadcom into a $1.71 trillion company.
"We are seeing exceptional AI-related bookings, and as a result, we are significantly raising Marvell's revenue outlook for both fiscal 2027 and fiscal 2028," Chief Executive Officer Matt Murphy told investors on the Q1 earnings call.
Revenue hit $2.418 billion in the fiscal first quarter, up 27.6% year over year, with data center climbing 11% sequentially. Free cash flow more than doubled to $483.1 million. Management guided second-quarter revenue to $2.70 billion, implying roughly 35% year-over-year growth. The custom XPU pipeline, which Reuters reported could top $10 billion by fiscal 2029, spans lead 3-nanometer programs with a top US hyperscaler and follow-on architecture work already underway.
At roughly $251, Marvell trades at 67 times forward earnings — more than three times Broadcom's 20x multiple — pricing in the custom silicon ramp before it materializes in full. Any delay in the lead 3nm XPU program, expected to enter production in calendar 2026, could trigger a violent rerating.
The Custom Silicon Flywheel
Marvell has recycled proceeds from asset sales — the automotive Ethernet unit went to Infineon for $2.5 billion — into acquisitions that deepen its interconnect moat. The company closed Celestial AI, a photonic fabric startup, in February 2026 and picked up XConn Technologies for chiplet connectivity, then raised $2 billion through a Series A convertible preferred to fund the buildout. The strategy mirrors Broadcom's: sell the picks and shovels, let Nvidia fight the merchant GPU war.
The NVIDIA NVLink Fusion partnership adds another layer. Marvell's high-speed optics, 51.2-terabit Ethernet switches, and custom silicon now touch every layer of the AI rack except the GPU itself. That breadth gives the company pricing power and customer stickiness that pure-play chip designers lack.
The Risks Priced Into 67x Earnings
The bear case is concentrated. Custom silicon revenue depends on a handful of hyperscalers, any of which could dual-source to Broadcom on the next node. Murphy acknowledged customers "may be pursuing multiple paths" on XPU supply. GAAP net income collapsed 80.4% year over year last quarter on a $331.8 million contingent consideration charge, while stock-based compensation climbed to $207.6 million. Insiders have logged 129 transactions, net selling.
At 67 times forward earnings versus the semiconductor industry average of 65.7, the stock leaves little room for execution missteps. Gross margin guidance of 58.25% to 59.25% for the current quarter will be the first test — if it holds and revenue clears $2.7 billion, the thesis gains credibility. If it slips, the multiple compression could erase months of gains.
Marvell's path to $10 billion in custom chip revenue by fiscal 2029 would compress today's forward multiple rapidly even without further expansion. The risk-reward is asymmetric at $251, but only for investors willing to bet that hyperscaler demand for custom silicon outruns the competitive response from Broadcom. Watch Q2 FY27 gross margins and the 3nm XPU timeline — those two data points will determine whether Marvell becomes Broadcom's equal or its cautionary tale.
This article is for informational purposes only and does not constitute investment advice.