Marvell is becoming an AI data-center interconnect company: what the filings confirm

Draft for EvidInvest review. This is financial research content, not investment advice. Built from a research-agent style Aether workflow over Marvell SEC filings: FY2026 10-K, Q1 FY2027 10-Q, and Q1 FY2027 8-K earnings exhibit. Evidence labels distinguish confirmed SEC facts from inferred supply-chain relationships and claims that still need sourcing.

Run your own SEC filing analysis with Aether: aether.evidinvest.com. This draft uses the same claim-discipline pattern: confirmed SEC facts, inferred relationships, and claims that still need a source.

Thesis

Marvell Technology ($MRVL) is not the easiest AI story to explain in one sentence. It is not NVIDIA. It is not a cloud operator. It is not an AI application company.

But that is exactly why the filings are useful.

Marvell's SEC evidence points to a company increasingly positioned in the connectivity layer of AI data centers: custom silicon, Ethernet switching, DSPs, optical interconnect, silicon photonics, CPO/LPO, retimers, CXL/PCIe switching, and data-center interconnect.

The core filing-backed point is simple: data center has become Marvell's dominant end market.

In FY2026, Marvell reported $8.195B of total revenue. Data-center revenue was $6.100B, or 74% of the company total. In Q1 FY2027, data-center revenue was $1.833B, or 76% of total revenue.

That makes MRVL a useful EvidInvest/Aether case study: the investment story is not just “AI demand is strong.” The better question is what the filings confirm about where Marvell sits in the AI infrastructure stack — and where the evidence stops.

What Marvell actually sells into the AI infrastructure stack

Marvell says it operates in one reportable segment: the design, development, and sale of integrated circuits. Its end-market disclosure separates the business into Data center and Communications and other.

The data-center application list is broad. In the FY2026 10-K, Marvell describes applications including cloud and on-premise AI systems, Ethernet switching, network-attached storage, AI servers, general-purpose servers, storage area networks, storage systems, and data-center interconnect.

That language matters because it anchors the AI claim in filing evidence.

The stronger part of the MRVL story is the product layer. Marvell's filings describe interconnect products including PAM DSPs, coherent and coherent-lite DSPs, laser drivers, trans-impedance amplifiers, silicon photonics, co-packaged optics, linear pluggable optics chipsets, data-center interconnect, active electrical cable DSPs, and PCIe retimer solutions.

Its custom ASIC disclosures add another layer: ultra-high-speed SerDes, ARM compute, security, storage, silicon photonics, advanced packaging, die-to-die interconnects, chiplets, co-packaged optics, and custom HBM.

Evidence label: the product categories above are confirmed SEC facts from Marvell's FY2026 10-K. The conclusion that Marvell is a connectivity/interconnect supplier to AI data-center buildout is an inferred relationship supported by those disclosures.

The financial signal: data center is now most of the company

The revenue mix is the cleanest signal.

Marvell's FY2026 total revenue was $8.195B, up from $5.767B in FY2025. That is a calculated 42.1% year-over-year increase from the figures in the annual filing.

Data-center revenue was even more important:

FY2024 data-center revenue: $2.217B, 40% of total revenue.
FY2025 data-center revenue: $4.164B, 72% of total revenue.
FY2026 data-center revenue: $6.100B, 74% of total revenue.

That is not a small mix shift. In two fiscal years, data center moved from a large end market to the center of the company.

The latest quarterly filing extends the same pattern. Q1 FY2027 net revenue was $2.418B, up 27.6% from the prior-year quarter. Q1 FY2027 data-center revenue was $1.833B, up 27.2% year over year and equal to 76% of total revenue.

In the Q1 FY2027 earnings exhibit, CEO Matt Murphy said the company expected revenue growth to continue accelerating through FY2027, driven by continued strength in data center, and cited “exceptional AI-related bookings.”

Evidence label: the revenue figures and CEO quotation are confirmed SEC facts from Marvell's 10-K, 10-Q, and 8-K earnings exhibit. Any valuation conclusion from those numbers is needs model.

The interconnect angle: why MRVL belongs in AI infrastructure maps

The interesting MRVL thesis is not simply “more AI revenue.” It is that AI infrastructure creates stress in the movement of data.

Clusters need scale-up links inside systems, scale-out networking across racks, optical links across the data center, coherent links over longer reaches, retimers, DSPs, switching, custom silicon, packaging, and power-efficient interconnect. Marvell's filings map directly to several of those layers.

The Q1 FY2027 earnings exhibit cited demand across 800G and 1.6T scale-out optics, 51.2T Ethernet scale-out switches, scale-up optical solutions for NPO and CPO applications, scale-across data-center interconnect modules, and custom XPU / XPU-attach solutions.

The acquisition evidence points in the same direction. Marvell completed the acquisition of Celestial AI on February 2, 2026, describing Celestial's Photonic Fabric technology as purpose-built for next-generation scale-up interconnect. It also completed the acquisition of XConn Technologies on February 10, 2026, describing XConn as a provider of PCIe and CXL switching silicon that expands Marvell's switching portfolio.

Evidence label: the product demand, Celestial AI, and XConn details are confirmed SEC facts. The broader conclusion that MRVL is becoming a key AI data-center interconnect platform is an inferred relationship.

Customer evidence: concentration is confirmed, identities are not

This is where claim discipline matters.

Marvell's filings confirm significant customer concentration. In FY2026, Direct Customer A represented 14% of revenue and Distributor A represented 37%. In Q1 FY2027, Direct Customer A represented 16% and Distributor A represented 45%.

The ten largest customers, including distributors and direct customers, represented 82% of FY2026 revenue. That is a real concentration signal.

But the reviewed filings do not name the hyperscalers or end customers behind those labels. The copy-safe claim is that Marvell has material unnamed customer and distributor concentration. The unsafe claim would be to map Customer A to a named cloud company without a separate primary source.

Evidence label: customer concentration is a confirmed SEC fact. Any named-customer mapping is needs source.

Supplier-chain evidence: fabless exposure, but unnamed partners

Marvell outsources fabrication to independent foundries. The FY2026 10-K says the company places firm supplier orders up to 26 weeks before anticipated customer delivery and may make supply commitments up to 52 weeks to secure capacity.

It also says Marvell has entered into, and expects to continue entering into, capacity reservation arrangements with certain foundries and substrate partners.

The filing risk factors are important. Marvell flags exposure to limited advanced-node foundry availability, substrate constraints, IC packaging capacity, and geographic concentration, including Taiwan and the Pacific Rim.

That supports the supply-chain thesis, but with a boundary. The reviewed filings support “advanced foundry, packaging, and substrate dependency.” They do not support naming a specific supplier such as TSMC, ASE, Amkor, or a substrate vendor unless a separate source is added.

Evidence label: outsourced manufacturing and capacity reservation language are confirmed SEC facts. Named supplier claims are needs source.

Risk side: AI upside is not linear

The same filings that support the AI infrastructure thesis also show the risks.

First, data center is now most of the company. That creates upside when AI infrastructure spending accelerates, but it also makes Marvell more exposed if data-center capex slows, shifts, or becomes more concentrated among fewer customers.

Second, customer concentration is high. One distributor represented 37% of FY2026 revenue and 45% of Q1 FY2027 revenue. The top ten customers represented 82% of FY2026 revenue.

Third, the supply chain is externally dependent. Advanced-node foundry availability, substrates, packaging capacity, and geographic concentration are not optional details for a fabless semiconductor company selling into AI infrastructure.

Fourth, geography matters. In Q1 FY2027, shipment destination/customer operations exposure included China at 44% of revenue and Taiwan at 21%. Marvell says most sales are to customers with operations outside the United States, primarily in Asia. That should be read carefully: geography of shipment is not the same thing as end-user demand, but it is still a relevant risk and exposure signal.

Bottom line

MRVL is a good example of why AI infrastructure research needs more than headline narratives.

The filings confirm that Marvell's business has shifted heavily toward data center. They also confirm a product stack closely tied to the hard physical bottlenecks of AI infrastructure: custom silicon, high-speed SerDes, switching, DSPs, optics, silicon photonics, CPO/LPO, retimers, and scale-up/scale-out interconnect.

The clean EvidInvest thesis is:

Marvell is a filing-confirmed AI data-center infrastructure supplier, especially in the interconnect/connectivity layer.

The clean caveat is just as important:

The filings do not let us name the hidden customers or specific manufacturing partners without separate sourcing.

That is the value of the Aether workflow: separate what is confirmed, what is inferred, and what still needs a source before the market narrative gets ahead of the evidence.

Do your own Aether analysis

You can run the same SEC-first workflow on MRVL or another ticker with Aether: aether.evidinvest.com.

Use it to pull filings, map business segments and customer/supplier disclosures, and label every claim as confirmed SEC fact, inferred relationship, or needs source before turning the research into a thesis.

Source notes

Marvell Technology, Inc. FY2026 10-K, filed 2026-03-11, accession 0001835632-26-000011.
Marvell Technology, Inc. Q1 FY2027 10-Q, filed 2026-05-28, accession 0001835632-26-000019.
Marvell Technology, Inc. Q1 FY2027 8-K earnings exhibit, filed 2026-05-27, accession 0001835632-26-000014.
Evidence labels: confirmed SEC fact / inferred relationship / needs source.