Short answer: Marvell’s disclosed AI exposure is more visibly weighted toward optical connectivity and custom silicon, while Broadcom reports a much larger AI-semiconductor category that combines custom accelerators with networking. Neither company can be declared an overall winner from the available figures: Marvell reports fourth-quarter fiscal 2026 data-center revenue, whereas Broadcom reports third-quarter fiscal 2026 AI-semiconductor revenue, with different scopes and periods.
Executive answer
The better-positioned company depends on which part of an AI data-center buildout you are evaluating.
As an Amazon Associate I earn from qualifying purchases.
- Choose Marvell as the closer fit for an optical-connectivity and custom-silicon thesis. Marvell reported $1.651 billion of data-center revenue in fiscal 2026 fourth-quarter results, up 21% year over year. Its fiscal 2026 proxy described optical interconnect as roughly half of data-center revenue and custom silicon as approximately one-quarter. Those are company descriptions, not independent market-share estimates.
- Choose Broadcom as the larger disclosed AI-semiconductor platform. Broadcom reported $16.7 billion of AI-semiconductor revenue in fiscal 2026 third-quarter results, up 221% year over year, attributing the growth to custom AI accelerators and networking. Broadcom also provides technology and intellectual property for customer ASICs used in AI and high-performance computing, networking, and storage.
- Do not compare the headline revenue amounts as if they were equivalent. Marvell’s number is a data-center end-market figure for one quarter; Broadcom’s number is an AI-semiconductor category for a different quarter. The categories overlap but are not interchangeable.
- Partnership announcements indicate ecosystem activity, not guaranteed sales. Marvell announced a collaboration with NVIDIA around NVLink Fusion. Broadcom announced accelerator and Ethernet networking work with OpenAI and an extended partnership supporting Meta’s MTIA custom silicon. None of these announcements establishes exclusivity, realized revenue, or technical superiority.
For a technology or investment comparison, the most useful question is therefore not “Which company wins?” but “Which disclosed exposure best matches the specific AI infrastructure layer and time horizon I am analyzing?”
What Marvell reports about AI data-center exposure
Data-center revenue and business mix
Marvell’s fiscal 2026 fourth-quarter presentation, published March 5, 2026, reported $1.651 billion in data-center revenue, a 21% year-over-year increase. Marvell said sequential growth came from optical interconnects, custom silicon, switching, and storage.
#1 Best Overall
Its fiscal 2026 proxy statement, published May 13, 2026, characterized optical interconnect as roughly half of data-center revenue and custom silicon as approximately one-quarter. These percentages are approximate descriptions of Marvell’s business mix rather than separately audited segment results. They are useful for understanding emphasis, but they should not be treated as a precise market-share calculation or as a forecast.
Connectivity is central to the Marvell thesis
Marvell’s disclosed mix makes connectivity a prominent part of its AI-data-center story. Optical interconnect products move data between accelerators, switches, servers, and other infrastructure over fiber. In large AI clusters, the ability to move data at high bandwidth and with acceptable power and signal integrity can constrain overall system performance just as much as compute capacity.
Marvell also reports switching and storage within its data-center business. The available disclosures do not break out revenue for each of those lines, so it is not possible to calculate how much of the $1.651 billion came from optical, custom silicon, switching, or storage individually.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Scan for outdated or missing drivers - takes under a minute3Clear out junk files and repair common Windows errorsMarvell’s 1.6T optical roadmap
On March 12, 2026, Marvell announced an expanded 1.6T optical digital-signal-processor portfolio. The announcement named the Ara, Alaska, and Nova DSP platforms, along with Ethernet PHY, silicon-photonics, and linear-pluggable-optics transmitter/impedance-amplifier (LPO TIA) and laser-driver chipsets.
A 1.6T announcement demonstrates product positioning and roadmap breadth. It does not, by itself, prove that every listed component is shipping at volume, that every platform has been adopted by customers, or that the products will generate a particular amount of revenue. Those questions require shipment, design-win, qualification, and customer-deployment evidence that the announcement does not quantify.
What Broadcom reports about AI data-center exposure
AI-semiconductor revenue combines accelerators and networking
Broadcom’s fiscal 2026 third-quarter results, published September 2, 2026, reported $16.7 billion of AI-semiconductor revenue, up 221% year over year. Broadcom attributed that growth to demand for custom AI accelerators and networking.
Broadcom’s reported category is broader than a single chip type. It includes the company’s role in custom accelerator programs as well as AI networking. The figure is therefore a useful indicator of Broadcom’s disclosed AI-semiconductor scale, but it cannot be mapped one-for-one to Marvell’s data-center end-market revenue.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Rank #2
- Ideal for Gifting
- Ideal for a bookworm
- Compact for travelling
Custom ASIC technology and intellectual property
Broadcom’s fiscal 2025 Form 10-K says its custom-silicon solutions provide technology and intellectual property for customer ASICs in AI and high-performance computing, networking, and storage. This describes a business model in which Broadcom helps customers develop specialized chips rather than selling only standardized merchant components.
Customer ASIC programs can involve architecture, design tools, intellectual property blocks, physical implementation, high-speed interfaces, and manufacturing coordination. The filing establishes Broadcom’s stated scope, but it does not disclose the economics, volume, or timing of each customer program.
Broadcom’s forward guidance is not a reported result
Broadcom expected $21.7 billion of AI-semiconductor revenue in fiscal 2026 fourth quarter. That number is management guidance, not realized revenue. It should be kept separate from the $16.7 billion third-quarter result when assessing growth, valuation, or execution.
Why the headline figures are not a clean head-to-head comparison
| Company | Reported measure | Period and publication | Reported result | Year-over-year change |
|---|---|---|---|---|
| Marvell | Data-center end-market revenue | Fiscal 2026 fourth quarter; March 5, 2026 presentation | $1.651 billion | 21% increase |
| Broadcom | AI-semiconductor revenue | Fiscal 2026 third quarter; September 2, 2026 results | $16.7 billion | 221% increase |
| Broadcom | AI-semiconductor revenue guidance | Fiscal 2026 fourth quarter guidance | $21.7 billion expected | Guidance, not a reported result |
Three differences matter:
- Category: Marvell’s figure is tied to a data-center end market. Broadcom’s figure is tied to AI semiconductors. Broadcom’s category can include products and programs that would not appear in Marvell’s data-center line on the same basis.
- Reporting period: Marvell’s cited quarter is fiscal 2026 fourth quarter; Broadcom’s cited actual result is fiscal 2026 third quarter.
- Scope: Neither company’s disclosed figure isolates exactly the same combination of optical components, switches, custom accelerators, ASIC services, and related products.
The figures can show that both companies are reporting AI-related growth, but they cannot establish that Broadcom has a specific multiple of Marvell’s AI business or that one has a larger market share.
Networking: optical interconnect, Ethernet, and switching
Marvell’s disclosed networking angle
Marvell provides the clearest public emphasis on optical interconnect in the cited material. Its proxy’s approximate mix description and its 1.6T DSP announcement point to a strategy built around moving data between elements of AI infrastructure.
The 1.6T portfolio spans more than a single DSP. Marvell named DSP platforms, Ethernet PHY, silicon-photonics components, LPO TIAs, and laser drivers. A portfolio covering these functions can give a supplier several points of participation in an optical link, but the announcement does not show how customers will combine the parts or which designs will reach production.
Broadcom’s disclosed networking angle
Broadcom attributes part of its AI-semiconductor growth to networking and has announced Ethernet networking collaboration alongside accelerator programs. Its public disclosures therefore support a networking role, but the cited material does not provide a product-by-product optical roadmap comparable to Marvell’s 1.6T announcement.
That does not mean Broadcom lacks optical or switching capability. It means the available announcements support different levels of detail: Marvell’s cited announcement names specific optical components, while Broadcom’s cited financial release describes networking as a driver of AI-semiconductor demand.
PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchScale-up and scale-out are different evaluation questions
Scale-up connects processors within a tightly coupled AI system, where latency, bandwidth, and coherence can be critical. Scale-out connects many systems across a cluster, where Ethernet, switching, optical modules, and congestion management become central. A supplier can participate in one layer, several layers, or a customer-specific combination.
The cited disclosures do not provide a complete, apples-to-apples map of Marvell and Broadcom products across every scale-up and scale-out layer. When evaluating a design, identify the exact link type, protocol, distance, speed, power envelope, and qualification status rather than inferring capability from a company-wide AI revenue number.
Custom silicon: how the companies’ disclosed positions differ
Marvell’s custom-silicon position
Marvell lists custom silicon as a major part of its data-center business, approximately one-quarter according to its fiscal 2026 proxy description. Its custom-silicon activity is presented alongside optical interconnect, switching, and storage rather than as a standalone reported revenue segment.
This combination suggests a model in which Marvell can address both the customized compute or infrastructure chip and the connectivity needed to attach it. The available figures do not disclose the number of active programs, customer concentration, per-program revenue, or time from design start to production.
Broadcom’s custom-accelerator position
Broadcom’s AI-semiconductor results explicitly attribute growth to custom AI accelerators and networking. Its Form 10-K describes customer ASIC technology and intellectual property spanning AI/HPC, networking, and storage.
Broadcom’s disclosed AI figure is much larger than Marvell’s cited data-center figure, but the difference cannot be interpreted as a pure custom-silicon comparison. Broadcom’s number includes networking and uses a different category definition and quarter.
Rank #4
What a customer or analyst should request
Public company disclosures are not enough to evaluate a specific custom-silicon engagement. A serious comparison should seek:
- Whether the program is in architecture, design, tape-out, qualification, or volume production.
- The customer’s intended deployment and expected ramp timing.
- Which functions are supplied by the vendor: compute die, interconnect, SerDes, optical components, software, packaging, or manufacturing support.
- Whether the relationship is exclusive, preferred, or one of several suppliers.
- How much revenue is recognized at each stage and how concentrated the program is.
Partnerships and ecosystem signals
Marvell and NVIDIA
On March 31, 2026, Marvell and NVIDIA announced a collaboration connecting Marvell custom-silicon and networking capabilities with NVIDIA’s AI ecosystem through NVLink Fusion. The announcement shows that Marvell is positioning its custom and connectivity technologies to work with a major AI-platform ecosystem.
It does not establish that the relationship is exclusive, that a specific product has entered volume production, or that it will produce a stated amount of revenue. Treat it as an ecosystem and product-alignment signal.
Broadcom and OpenAI
On October 13, 2025, Broadcom announced accelerator and Ethernet-networking collaboration with OpenAI. The stated scope links custom accelerator work with Ethernet networking, matching the two drivers Broadcom later cited for AI-semiconductor growth.
The announcement does not provide a realized-revenue figure, shipment schedule, exclusivity term, or independent performance comparison.
Broadcom and Meta
On April 14, 2026, Broadcom announced an extended partnership supporting Meta’s MTIA custom silicon. This is evidence of a continuing customer relationship around a named custom-silicon program. It does not disclose the program’s revenue contribution, production volume, or margin.
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →Roadmap, revenue, and customer adoption are separate evidence types
Readers often combine these signals as if they measured the same thing. They do not.
Best Value
- It can be a gift option
- Comes with secure packaging
- Helpful in various ways
- Reported results are recognized revenue for a defined period, such as Marvell’s $1.651 billion data-center figure or Broadcom’s $16.7 billion AI-semiconductor figure.
- Business-mix descriptions such as Marvell’s approximate half optical and quarter custom-silicon split explain emphasis but are not independent segment audits.
- Product announcements such as Marvell’s 1.6T portfolio show intended product breadth, not shipment scale.
- Partnership announcements show stated collaboration scope, not guaranteed revenue or exclusivity.
- Guidance such as Broadcom’s $21.7 billion fourth-quarter expectation is management’s outlook and can differ from the eventual result.
A disciplined comparison keeps each evidence type in its own column or paragraph instead of treating a roadmap announcement as revenue or a partnership as a booked design win.
Which company fits different evaluation goals?
If your focus is optical connectivity
Marvell has the more explicit cited evidence: optical interconnect is described as roughly half of its data-center revenue, and its 1.6T announcement names DSP, PHY, silicon-photonics, TIA, and laser-driver products. You still need shipment and qualification data before concluding that the roadmap translates into market adoption.
If your focus is custom AI accelerators
Broadcom reports AI-semiconductor growth driven by custom AI accelerators and networking, and its filing describes customer ASIC support across AI/HPC, networking, and storage. Marvell also reports a substantial custom-silicon role, but the cited disclosures provide less absolute revenue detail for that activity.
The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →If your focus is a named ecosystem relationship
Marvell’s cited relationship is with NVIDIA through NVLink Fusion. Broadcom’s cited relationships are with OpenAI for accelerator and Ethernet networking collaboration and with Meta for MTIA custom silicon. The relevant choice depends on which ecosystem, architecture, and customer deployment you are analyzing.
If your focus is reported AI growth
Broadcom’s $16.7 billion AI-semiconductor result and 221% year-over-year growth are the larger disclosed numbers, but they cover a different category and period from Marvell’s $1.651 billion data-center result and 21% growth. Use the figures to understand each company’s reporting narrative, not to calculate relative market share.
Financial-analysis cautions for individual readers
This comparison is about business and technology disclosures, not a recommendation to buy or sell either company’s stock. Individual investors should avoid treating one quarter’s growth rate as a durable trend without checking subsequent filings, customer concentration, gross-margin effects, capital requirements, and the timing of design ramps.
Pay particular attention to the distinction between revenue already recognized and management expectations. Broadcom’s $21.7 billion fourth-quarter AI-semiconductor figure is guidance. Marvell’s optical and custom-silicon percentages are approximate business descriptions. Neither disclosure supplies independent verification of market share or customer satisfaction.
Free tools Windows power users keep installed
One-click scans. No signup required.
Because the companies use different category definitions, a spreadsheet comparison should preserve the original label, fiscal period, currency, and qualification beside every number. Reclassifying Marvell data-center revenue as AI revenue or Broadcom AI-semiconductor revenue as pure custom silicon would create a misleading series.
Risks and unanswered questions
- Category overlap: Networking, optical, custom silicon, storage, and accelerators can be reported in different categories by different companies.
- Roadmap execution: A 1.6T product family must still pass customer qualification and production-ramp milestones.
- Customer concentration: Large custom programs can make results sensitive to a small number of hyperscaler decisions.
- Partnership interpretation: An announced collaboration may involve evaluation, co-development, or future deployment; the cited announcements do not specify realized revenue.
- Guidance risk: Broadcom’s fourth-quarter AI figure is an expectation and can differ from the eventual result.
- Missing independent benchmarks: The cited material does not quantify comparative market share, relevant-category margins, or independent customer satisfaction.
A practical checklist for updating the comparison
- Record each company’s fiscal quarter and the exact revenue label before comparing growth.
- Separate actual revenue from guidance and from approximate business-mix commentary.
- For optical products, check whether the specific DSP, PHY, photonics, TIA, or laser-driver part is sampling, qualified, or in volume production.
- For custom silicon, identify the customer, program stage, expected ramp, and functions supplied.
- Map the product to scale-up or scale-out networking and record the protocol, bandwidth, distance, and power requirements.
- Read partnership language for scope and timing; do not assume exclusivity or guaranteed revenue.
- Look for later filings that confirm or revise the stated figures and guidance.
The Bottom Line
Bottom line: Marvell offers the clearer disclosed optical-connectivity and custom-silicon mix, while Broadcom reports the larger AI-semiconductor platform spanning custom accelerators and networking. The evidence supports choosing a comparison criterion, not naming a universal winner; the reported figures, fiscal periods, product disclosures, and partnership announcements are not directly interchangeable.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
