Finvest
MRVL Semiconductors · AI infrastructure · Data center · Fabless chips · Thesis updated June 11, 2026

AI demand raised the bar

01 Running thesis

Big AI plan, little room for error

Marvell's story changed again after Q1 FY27. Management lifted fiscal 2027 revenue guidance by more than $500 million to nearly $11.5 billion. It also lifted fiscal 2028 guidance by $1.5 billion to about $16.5 billion. That is a major reset for a company that was at $8.19 billion of revenue in fiscal 2026.

The bull case is simple: AI data centers need faster links between chips, servers, and clusters. Marvell sells those links through optical chips, electrical links, switches, storage controllers, and custom ASICs. An ASIC is a chip built for one customer's special job. Data Center revenue is now expected to grow about 50% in fiscal 2027 and about 55% in fiscal 2028.

The bear case is also simple. The stock is being valued as if the new $16.5 billion fiscal 2028 plan happens. That leaves little safety if a large custom chip program slips, if supply is tight, or if cloud companies slow their AI spending.

The next test is quarterly execution. Management guided to at least 10% sequential revenue growth for Q3 and Q4, which would put the company at a $3 billion quarterly revenue run rate in Q3 FY27. Investors should also watch whether custom silicon really sets up to more than double in fiscal 2028.

May 2026The Q1 FY27 10-Q confirmed the new AI-led outlook and added sharper concentration data. One distributor was 45% of revenue, one direct customer was 16%, and Asia was 83% of revenue.
May 2026Management raised fiscal 2027 revenue guidance to nearly $11.5 billion and fiscal 2028 guidance to about $16.5 billion. Data Center, interconnect, and custom silicon all drove the stronger plan.
Mar 2026The fiscal 2026 10-K confirmed the new two-end-market structure. Data Center was 74% of fiscal 2026 revenue, and the filing added a long-term risk that AI tools could lower chip design barriers.
Mar 2026The Q4 FY26 call raised fiscal 2027 revenue guidance toward $11 billion and introduced a fiscal 2028 target near $15 billion. The main driver was stronger AI data center demand.
Dec 2025The Q3 FY26 10-Q supported the earnings-call view. Revenue rose 37% year over year, with Data Center up 38%.
Dec 2025Marvell announced the Celestial AI acquisition for about $3.25 billion upfront, plus a possible earnout. The deal strengthened the long-term optical interconnect story but added integration risk.
Aug 2025The Q2 FY26 10-Q confirmed strong data center growth and the completed auto business sale. It did not change the core thesis.
Aug 2025The Q2 FY26 call showed near-term lumpiness in custom silicon, with a guided sequential dip before a possible Q4 rebound. Strong electro-optics and recovering non-data-center markets partly offset that risk.
02 Business model

Selling the picks for AI clusters

Marvell is a fabless chip company. That means it designs chips but uses outside manufacturers to make them. Its main customers are large cloud and networking companies that need fast, low-power data movement.

The company makes money in two main ways. First, it sells merchant products, which are standard chips such as optical DSPs, switches, retimers, and storage controllers. These tend to carry higher gross margins. Second, it builds custom silicon for Tier 1 cloud customers. These chips can be lower gross margin, but customers often pay non-recurring engineering fees, called NRE, to help cover design costs.

The moat comes from hard-to-copy chip IP and years of trusted work with major cloud buyers. Key building blocks include high-speed SerDes, ARM compute, security, silicon photonics, high-bandwidth memory interfaces, and advanced packaging. The Celestial AI deal adds Photonic Fabric technology, which is meant to move data inside AI clusters with lower power and lower delay.

The weak point is concentration. A few customers and programs matter a lot. If one large cloud roadmap changes, Marvell can lose revenue faster than a broad chip supplier.

03 Product portfolio

Where the chips fit

Growth engine

Electro-optics and interconnect

These chips help data move across optical modules, cables, and data center links. Management now expects interconnect revenue to grow more than 70% year over year in fiscal 2027.

Growth engine

Custom silicon and ASICs

Marvell designs special chips for large cloud customers and AI workloads. Management expects custom revenue to more than double in fiscal 2028, making this one of the biggest swing factors.

Steady

Switching and networking

Ethernet switches and related networking chips move traffic inside data centers, campuses, and carrier networks. This line benefits when AI clusters need more network bandwidth.

Steady

Storage controllers

Marvell sells controllers used in cloud, enterprise, and consumer storage devices. Storage is less flashy than AI accelerators, but it remains part of the data infrastructure bundle.

Cash cow

OCTEON DPUs and carrier chips

DPUs, PHYs, and carrier networking chips serve enterprise and telecom buyers. These markets can recover sharply, but they are still cyclical.

Option

Photonic Fabric from Celestial AI

This acquired technology is aimed at optical connections inside AI systems. Management expects scale-up optics to contribute more than $300 million in fiscal 2028.

04 Business segments

Now mostly data center

Data Center76%growing fast
Communications and Other24%modest

Segment mix uses Q1 FY27, the quarter ended May 2, 2026. The 10-Q also shows heavy concentration: one distributor was 45% of net revenue and one direct customer was 16%.

05 Risk factors

What could break the story

Custom program slip

High impact · Medium odds

A large part of the upside depends on custom silicon programs for Tier 1 cloud customers. If a new XPU program is delayed, canceled, or ramps below plan, the fiscal 2028 target becomes harder to hit.

We watchListen for updates on the new Tier 1 XPU program and whether custom revenue is still expected to more than double in fiscal 2028.

Cloud capex pause

High impact · Medium odds

Marvell is tied closely to AI data center spending. If large cloud companies slow capital spending, orders for interconnect, switching, and custom silicon could fall short.

We watchTrack management comments on cloud bookings, Data Center growth, and the fiscal 2028 revenue target of about $16.5 billion.

Customer concentration

High impact · High odds

Revenue is packed into a small group of buyers. In Q1 FY27, one distributor was 45% of net revenue and one direct customer was 16%. In fiscal 2026, the ten largest customers were 82% of total net revenue.

We watchWatch customer concentration in each 10-Q and any sign that a major cloud customer is changing chip suppliers.

Taiwan and Asia supply exposure

High impact · Medium odds

Marvell depends on third-party manufacturing and assembly partners, with major exposure to Taiwan. Sales shipped to customers with operations in Asia were 83% of net revenue in Q1 FY27, up from 74% a year earlier.

We watchMonitor supply chain commentary, Taiwan-related disruptions, and any warning about product availability.

Lower-margin custom mix

Medium impact · Medium odds

Custom silicon can carry lower gross margins than standard merchant chips. The model works best when customers pay NRE fees and volumes scale enough to cover high design costs.

We watchWatch non-GAAP gross margin, operating margin, and any disclosure about NRE collections.

Celestial AI integration

Medium impact · Medium odds

The Celestial AI deal is a large bet on photonic interconnect. If Marvell cannot integrate the team and technology, the deal could dilute returns or lead to impairment.

We watchLook for scale-up optics revenue progress toward more than $300 million in fiscal 2028 and management comments on product milestones.
06 Quick answers

In one breath

What does Marvell Technology do?

Marvell designs chips for data infrastructure. Its main growth areas are AI data centers, optical interconnect, networking, storage, and custom chips for large cloud companies.

Why is Marvell linked to AI?

AI clusters need huge amounts of data to move quickly between chips, servers, and data centers. Marvell sells the chips that help move that data, and it also builds custom silicon for cloud customers.

What is the biggest risk for MRVL stock?

The biggest risk is execution against a very high growth plan. Management is targeting about $16.5 billion of revenue in fiscal 2028, so a custom chip delay or cloud spending slowdown could hit the stock hard.