Marvell Technology, Inc. (MRVL)

Let's get the hard part over with first and address the risks because they're not trivial.

Headwinds

1. Marvell is expensive on nearly any metric.
2. The valuation leaves no room for error; it requires near-perfect execution.
3. It also needs a cooperative macro: a dovish Fed and hyperscaler capex that keeps climbing without a sharp risk-off shock.
4. Competition is intensifying, most directly from Broadcom.

Tailwinds

1. AI data-center revenue is scaling fast: the FY2028 guide of $16.5B has Marvell roughly doubling from FY2026.
2. Custom silicon and optics are the engine: hyperscalers need ASICs, optical interconnect, switching, and the fabric that moves data between accelerators.
3. As the industry works to reduce its dependence on Nvidia, Marvell is well positioned for a more heterogeneous compute world.
4. The stock god, Jensen Huang, said Marvell will go up.
5. See point above.

Data Center - $1.8B revenue, 76% of total revenue

The data center segment is growing 27% year over year and 11% sequentially. What makes the figure interesting is not its size but its slope. Management guided the segment to grow about 50% for the full year and framed the trajectory in plain terms: “we were 46% data center growth last year. The cherry on top is that Murphy is guiding to $10B in custom XPU sales in FY 29.

Growth is broad-based across AI-driven product lines, including strong 800G PAM4 demand, a rapid 1.6T ramp, scale-out switching, scale-up optics, DCI modules, and custom silicon. Management's outlook points to multiple billion-dollar revenue opportunities emerging across optics, switching, and custom silicon, with custom revenue expected to grow more than 20% in FY27 and more than double in FY28 as multiple programs ramp.

Custom silicon: the XPUs Marvell co-designs for hyperscalers — did about $1.5 billion in FY2026, is guided to grow more than 20% in FY2027, and to more than double in FY2028 — a path management still underwrites to exceed $10 billion by FY2029 Since last quarter, Marvell won several new designs as customers continue expanding their adoption of custom silicon. Marvell expects new sockets to begin contributing incremental revenue following their typical development cycle of approximately 2 years. The level of custom engagement with key customers remains unprecedented. And Marvell continue to be deeply involved in a broad set of significant additional opportunities.

Interconnect: is inflecting even harder. Interconnect, the largest piece of the data center business, is now guided to grow more than 70% year over year in FY2027 — up from a prior 50% expectation — on strong 800G demand and a fast 1.6T ramp, with 400-gig-per-lane already demonstrated. Datacenter-interconnect modules now ship to all five major US hyperscalers and are tracking from roughly $500 million in FY2026 toward $1 billion in FY2028; scale-out switching more than doubles past $600 million in FY2027 on the 51.2T platform, with a $1 billion-plus FY2028 target and a 100T part in development. This is not a roadmap deck — “it is not a PowerPoint,” as Murphy put it, pointing to 15 billion field hours across four generations of silicon photonics and 224-gig SerDes in production. Optics, he argued, “is the future of data center connectivity,” and Marvell is one of “maybe a couple” of companies that can integrate at this level. The pitch to hyperscalers is simply that “we have all the pieces.”

Communications - $0.5B revenue, 24% of total revenue

The communications and other end market has now largely recovered from inventory corrections at our customers. Going forward, Marvell expects revenue in this end market to broadly reflect the underlying trends in the enterprise networking carrier and consumer businesses.

Marvell helps reduce Nvidia dependencies

Marvell offers differentiated exposure to AI infrastructure versus a pure Nvidia position. While Nvidia remains the dominant GPU platform, hyperscalers are increasingly investing in custom silicon to optimize performance, cost, and power consumption for specific workloads, and Marvell captures that shift through custom XPUs, ASIC programs, optical interconnect, and high-speed networking. This does not make Marvell a hedge against AI demand — both companies depend on continued data-center capex — but it does make it a hedge against a GPU-only view of the AI compute stack.

A concrete example is the five-year agreement Marvell signed with AWS in December 2024 to supply a broad range of data-center semiconductors, spanning custom AI silicon, optical digital signal processors (DSPs), active electrical cable (AEC) DSPs, PCIe retimers, data-center interconnect (DCI) optical modules, and Ethernet switching silicon.¹

The structural point is share. Marvell and Broadcom together account for roughly 95% of the custom AI ASIC co-design market, and Marvell holds an estimated 20–25% of it, anchored by AWS's Trainium and Microsoft's Maia programs.²

Crucially, this is less a bet against Nvidia than a bet on the plumbing. Marvell also partners with Nvidia directly — co-developing silicon photonics, integrating with NVLink Fusion so its custom chips interface with Nvidia systems, and pairing its OCTEON silicon with Nvidia GPUs for AI RAN. It sells into the cluster whether the compute is a GPU or a custom XPU, which is why the connectivity franchise is the real hedge. That positioning is what led Nvidia's Jensen Huang, at COMPUTEX in June 2026, to name Marvell a candidate for the trillion-dollar club and to call connectivity “the next bottleneck in the AI infrastructure space.”³ Analysts flagged the remark as aspirational, but the underlying claim — that data movement, not just compute, is where the next constraint binds — is precisely the ground Marvell owns.

Key Metrics

1. FY28 revenue guided to $16.5B. Management raised the FY2028 outlook to $16.5B, about 45% above the FY2027 base of nearly $11.5B and roughly $1.5B higher than the prior outlook.
2. Data center is 76% of revenue. Data center reached $1,832.7M in Q1 FY27 out of $2,417.8M total, and management expects it to grow about 50% this fiscal year on interconnect, switching, and custom silicon.
3. Path to 50% returns by 2028. A 20x sales multiple on the guided $16.5B FY2028 revenue implies a $330B market cap, about 1.5x today's $220B.
4. Non-GAAP operating margin target 38% to 40%. Q1 FY27 non-GAAP operating margin was 35%, and management targets the upper end of its 38% to 40% model as data center scales.
5. Forward P/E glides to 69x by FY28. Assuming net income margins of 29%, FY 28 net income is ~5 bln, resulting in a P/E ratio of 69x.