Marvell Raises Long-Term Targets as AI Data Center Demand Surges

Marvell raised its long-term revenue outlook after reporting record quarterly sales, yet the Google News headline leaves out a striking conflict. Revenue climbed 37% from a year earlier, but investors still pushed the shares sharply lower after the results.

The reaction suggests that strong AI growth is no longer enough by itself. Marvell must now prove that its custom processors, optical components, and networking chips can support another acceleration through fiscal 2028.

That puts the company between two demanding benchmarks. Nvidia dominates AI accelerators and increasingly influences complete data center architectures. Broadcom has established a formidable custom silicon business with hyperscale cloud customers. Marvell is promising growth across the infrastructure connecting those systems, but expectations have risen alongside its opportunity.

Marvell Raised the Bar After a Record Quarter

The important change was not simply that Marvell beat its quarterly forecast. Management raised its longer-term expectations again while predicting faster growth ahead.

Marvell reported fiscal second-quarter 2027 revenue of $2.739 billion for the period ending August 1, 2026. That represented 37% year-over-year growth and came in $39 million above the midpoint of management’s previous guidance.

The company recorded GAAP net income of $308 million, equal to $0.33 per diluted share. Non-GAAP net income reached $865.9 million, while non-GAAP diluted earnings were $0.94 per share.

Operating cash flow totaled $605.5 million. GAAP gross margin was 53.1%, compared with a non-GAAP gross margin of 58.9%.

Those figures, disclosed in Marvell’s quarterly results, established a new second-quarter revenue record. They also continued a sequence of accelerating growth.

First-quarter revenue had reached $2.418 billion, up 28% year over year. The second quarter then added roughly $321 million sequentially, or 13%.

The strongest contribution came from the data center segment. Revenue there rose 46% year over year to approximately $2.17 billion, representing about 79% of Marvell’s total quarterly revenue.

Marvell attributed that expansion to AI-related demand across electro-optics, custom silicon, storage, and switching products. These categories serve different parts of an AI cluster rather than relying on one processor family.

Electro-optics converts electrical signals into optical ones so data can travel efficiently between servers and data center buildings. Custom silicon refers to chips designed for a particular customer’s workload or system architecture.

Switches move data among processors, memory, storage, and networks. These products become more important as AI clusters expand from thousands to hundreds of thousands of computing devices.

CEO Matt Murphy said AI-related bookings remained exceptionally strong. Management expects revenue growth to accelerate through the remainder of fiscal 2027.

For the third quarter, Marvell forecast revenue of $3.15 billion, plus or minus 5%. The midpoint implies growth of more than 50% from the comparable period.

Marvell also forecast non-GAAP diluted earnings of $1.10 per share, plus or minus $0.05. Non-GAAP gross margin is expected to fall between 57.5% and 58.5%.

The longer-term upgrade carries greater weight than the quarterly beat. Marvell now expects data center revenue to grow about 60% during fiscal 2027 and more than 60% in fiscal 2028.

Management had previously expected roughly 50% data center growth for fiscal 2027. Its fiscal 2028 expectation was also lower before the latest update.

The company reportedly expects total revenue of approximately $12 billion in fiscal 2027 and around $18 billion in fiscal 2028. These figures remain management projections, not contracted or recognized revenue.

That distinction matters. A booking can indicate customer demand, but revenue depends on final designs, production schedules, component availability, deployment timing, and customer acceptance.

Marvell plans to provide more detail during an investor event on October 6, 2026. That presentation is now an important test of how much support exists beneath the upgraded targets.

Why the Google News Headline Misses the Real Pressure

The Google News version frames this as a successful earnings report, while the market treated it as an execution test with little room for delay.

Marvell’s results exceeded its prior revenue midpoint and matched the central AI growth narrative. Even so, the shares fell 10.3% in the next trading session, according to a market recap.

That reaction did not make the quarter weak. It showed that investors had already assigned considerable value to future AI revenue.

Marvell’s shares had more than doubled following Nvidia’s investment and partnership announcement in March. Against that backdrop, a solid quarter could still disappoint anyone expecting a larger immediate increase in guidance.

Morgan Stanley analyst Joe Moore described the quarter and outlook as largely consistent with management’s previous expectations. He maintained a neutral rating while recognizing the company’s expanding set of AI growth drivers.

That assessment captures the pressure now facing Marvell. Management has elevated its targets, but the market wants evidence that current bookings will produce repeated upside beyond those targets.

Marvell is not selling one interchangeable AI chip. Its opportunity spans custom compute, high-speed optical links, switches, storage controllers, and connections among separate data center sites.

That breadth reduces dependence on a single product category. It also creates more execution points where qualification schedules or customer deployments can slip.

A custom chip can require years of joint design work before generating substantial production revenue. A design win establishes a place in a future system, but it does not guarantee the deployment’s timing or ultimate size.

Optical products face their own qualification cycles. Cloud operators must validate performance, power use, reliability, thermal behavior, and compatibility before deploying components throughout expensive AI infrastructure.

Marvell must therefore convert a broad portfolio into synchronized customer ramps. Its fiscal 2028 outlook assumes that several product groups expand together rather than one segment carrying the entire forecast.

The competitive standard is also changing. Hyperscalers increasingly want complete platforms that combine compute, memory, networking, software, and interconnect technologies.

Nvidia has responded by extending beyond graphics processors into networking, switches, systems, and software. Broadcom combines networking leadership with a large custom accelerator business.

Marvell occupies a valuable middle position. It can help customers build specialized infrastructure without requiring them to source every data movement component from one dominant platform vendor.

However, that position only becomes financially meaningful when customers move from development programs to high-volume deployment. The difference between a promising architecture and recognized revenue can span several reporting periods.

This is why the post-earnings decline matters. It represents a conflict between long-term enthusiasm and short-term proof, not a simple rejection of the AI thesis.

A headline about improved profit and revenue captures the reported quarter. It does not capture the expectation embedded in Marvell’s valuation or the difficulty of meeting a sharply higher fiscal 2028 target.

Marvell Is Building Around Custom Silicon and Connectivity

Marvell’s central bet is that AI infrastructure will require more specialized chips and far more connectivity, creating an opening beside Nvidia rather than directly replacing it.

The biggest AI systems cannot operate through accelerators alone. Thousands of processors must exchange data quickly enough to work as one machine.

This creates three related networking problems. Scale-up connections link processors inside a computing system. Scale-out networks connect many systems inside a data center. Scale-across technology moves information between separate facilities.

Marvell sells components for each layer. Its portfolio includes high-speed Ethernet switches, optical digital signal processors, data center interconnect products, storage technology, and custom processors.

The company strengthened that portfolio through several acquisitions. It completed its acquisition of Celestial AI on February 2, 2026, followed by XConn Technologies on February 10.

Celestial AI developed a photonic fabric designed for high-bandwidth connections inside AI systems. Photonics uses light to move data, which can reduce the electrical and power constraints of conventional links.

XConn brought switching products based on PCI Express and Compute Express Link. CXL is a standard that lets processors share memory and attached devices more efficiently.

Marvell says these additions expand its ability to support scale-up networks. They also add integration work, acquisition costs, and product road maps that management must coordinate.

The company’s regulatory filing says both businesses contributed to fiscal second-quarter results from their acquisition dates. The filing does not isolate how much revenue each acquisition produced.

Marvell also acquired Polariton Technologies, which develops electro-optic components. Together, these transactions support a strategy built around increasing data movement requirements.

That mechanism is easy to overlook when attention remains fixed on accelerator performance. A faster processor cannot deliver its full value if surrounding networks cannot feed it data or coordinate work across the cluster.

AI models are also expanding beyond the physical limits of one server rack. Training and inference increasingly distribute tasks across larger groups of accelerators, making latency and bandwidth central system constraints.

Marvell expects these constraints to increase demand for 800-gigabit and 1.6-terabit optical products. It also sees opportunities in 51.2-terabit Ethernet switches and newer scale-up optical designs.

These numbers describe data transmission capacity, not processor speed. A 1.6-terabit optical connection can move twice the data of an 800-gigabit link under comparable conditions.

Marvell’s custom silicon business addresses another change. Large cloud providers want processors designed around their own workloads, software, energy limits, and infrastructure plans.

A custom XPU is a specialized processing unit built for a customer rather than sold as a standard merchant chip. The label can cover accelerators and related computing devices designed for particular AI operations.

This approach gives cloud companies greater control over cost and system design. It can also reduce dependence on general-purpose accelerators for predictable, high-volume workloads.

Broadcom has already shown that custom AI silicon can become a large semiconductor business. Marvell must establish that it can win and deliver enough programs to become a durable second supplier.

Its partnership with Nvidia further complicates the competitive picture. Nvidia invested $2 billion in Marvell through convertible preferred stock on March 31, 2026.

The companies also announced plans to connect Marvell custom XPUs and compatible scale-up networking with Nvidia’s AI infrastructure environment. That arrangement positions Marvell as both an alternative supplier and a participant in Nvidia-centered systems.

This is not a straightforward Marvell-versus-Nvidia contest. Marvell can benefit when Nvidia’s platform expands, provided customers adopt Marvell connectivity or custom components around it.

The partnership also gives Nvidia influence over an important infrastructure supplier. Investors will need to watch whether interoperability broadens Marvell’s addressable market or makes its growth more dependent on Nvidia’s architecture.

Marvell reported the investment and partnership details in its latest filing. The relationship still needs product deployments and recognized revenue before its full commercial value becomes measurable.

Broadcom Sets the Custom Chip Benchmark

Broadcom is the primary competitive benchmark because Marvell’s largest upside depends on proving that another supplier can scale custom AI silicon across major cloud customers.

Custom processors appeal to cloud companies because AI infrastructure has become both strategic and expensive. A provider operating at hyperscale can justify designing chips around its own software and workloads.

The economics differ from those of merchant processors. The customer often helps define the product, while the chip supplier supplies engineering, intellectual property, packaging expertise, and production coordination.

This model can produce stable, high-volume programs after deployment. It also creates substantial exposure to individual customers and long product cycles.

Broadcom entered the current AI cycle with deep networking expertise and established custom silicon relationships. Its position raises expectations for any competitor claiming similar opportunities.

Marvell does not need to displace Broadcom across the entire market. It needs enough large, successful programs to demonstrate that hyperscalers want multiple capable design partners.

Supply diversity offers customers several advantages. It reduces dependence on one vendor, expands negotiating flexibility, and gives different internal teams access to specialized technologies.

Marvell can distinguish itself through its combination of custom compute and connectivity. A customer might work with the company on an accelerator, the switches around it, and the optical links connecting clusters.

That integrated capability forms the strongest argument behind Marvell’s targets. The company can capture more value from each AI deployment if multiple portfolio elements enter the same infrastructure program.

Yet integration does not automatically translate into wins. Hyperscalers can divide projects among vendors, develop more technology internally, or change architectures before production begins.

Marvell’s latest guidance indicates that management expects a significant acceleration in its custom business during the second half of fiscal 2027. The following quarters should reveal whether that ramp is broad or concentrated.

The first-quarter outlook had already identified custom XPUs, XPU attachment products, switches, and optics as growth drivers. The second-quarter update raised expectations again.

Repeated increases can signal genuine demand visibility. They can also compress the margin for future upside because each revised forecast establishes a higher baseline.

Broadcom remains the most useful comparison because it has converted custom silicon relationships into a major AI business. Nvidia is a broader platform rival, partner, and customer influence rather than a clean direct comparison.

Marvell’s challenge is therefore specific. It must execute several customer-specific programs while scaling connectivity products that serve the wider AI market.

A successful result would not mean that Marvell replaces either competitor. It would show that AI infrastructure spending supports multiple large semiconductor suppliers with differentiated roles.

Failure would look different. A delayed custom program, slower optical adoption, or reduced hyperscaler spending could leave Marvell with high development expenses before expected revenue arrives.

The contest will be decided through production schedules and customer adoption, not announcement counts. That makes quarterly revenue composition more useful than a running list of design wins.

What the Raised Targets Do Not Prove

Marvell’s outlook supports a strong demand case, but it does not eliminate customer concentration, supply constraints, acquisition risk, or dependence on sustained AI spending.

The company’s data center segment generated about four-fifths of total quarterly revenue. That concentration allows Marvell to benefit quickly from AI infrastructure investment, but it also magnifies any slowdown.

Marvell explicitly warns that data center sales can fluctuate significantly. Customers can reduce purchases, change deployment timing, or alter their technology strategies.

Large cloud customers also account for a substantial share of the business. Four customers represented 72% of gross accounts receivable at the end of the second quarter.

That does not mean four customers produced 72% of revenue. Accounts receivable measures unpaid customer balances at a particular date, so it can shift with shipment and payment timing.

Still, the figure illustrates concentration. Losing a major design or encountering a delayed deployment can materially affect quarterly results.

The company also depends on a limited group of manufacturing partners. Marvell is fabless, meaning it designs chips while outside suppliers manufacture them.

Its filing identifies access to advanced TSMC wafers and other critical components as a risk. Capacity constraints can affect production even when end-customer demand remains strong.

Marvell entered agreements to secure long-term wafer and substrate capacity. Such commitments help protect supply, but they can create costs if demand later arrives below expectations.

Acquisitions introduce another uncertainty. Celestial AI, XConn, and Polariton expand the technical portfolio, but acquired teams and products still need integration.

Management must align engineering schedules, sales relationships, manufacturing plans, and product interfaces. Any delay could move expected revenue into a later fiscal period.

Margins also deserve attention. Marvell’s second-quarter GAAP gross margin was 53.1%, while its non-GAAP measure was 58.9%.

The difference reflects excluded costs such as stock-based compensation, acquisition-related expenses, and amortization of acquired intangible assets. Investors should evaluate both measures because acquisitions are central to the current strategy.

Third-quarter non-GAAP gross margin guidance is slightly lower at its midpoint. Product mix, early production ramps, and acquired operations can influence that measure.

The market reaction after earnings suggests that investors noticed these execution questions. A strong quarter was insufficient to produce another immediate revaluation.

That reaction should not be interpreted as proof that management’s forecast is wrong. Share prices incorporate interest rates, prior gains, positioning, and many expectations beyond one earnings release.

It does show that Marvell has moved beyond the stage where general AI optimism can carry the story. The company is now accountable for specific quarterly ramps.

There is also a wider industry risk. Cloud providers are spending heavily on data centers before the long-term returns from many AI services are fully established.

If those companies slow capital spending, semiconductor suppliers would feel the effect through delayed orders and altered deployment schedules. Connectivity vendors are not insulated from that cycle.

An independent AI spending analysis noted that strong technology earnings have not resolved questions about returns on the industry’s investment boom. Continued expansion depends on customers maintaining their buildout plans.

Marvell’s outlook assumes that AI systems will keep requiring more compute and much more data movement. The technical logic is credible, but the commercial timing remains exposed to customer budgets.

For readers arriving through Google News, the central caveat is straightforward. Reported revenue is historical and verified, while fiscal 2027 and 2028 targets remain forward-looking company estimates.

Three Signals to Watch After Marvell’s Google News Moment

The next test is whether Marvell turns broad AI demand into measurable custom silicon and connectivity revenue without sacrificing execution quality.

The first signal is the company’s October 6 investor event. Marvell has already announced that the presentation will focus on its long-term strategy and AI infrastructure growth drivers.

Management needs to explain the bridge between current quarterly revenue and its fiscal 2028 target. Useful disclosure would separate major contributions from custom compute, optics, switching, and other data center products.

Investors should also watch for details about timing. A forecast supported by several independent product ramps carries different risk from one dependent on a small number of customer programs.

The event’s published schedule confirms the date, but not the additional metrics management will provide. Clear segment detail would strengthen the raised-target narrative.

The second signal is Marvell’s third-quarter report. Management has forecast $3.15 billion in revenue at the midpoint, representing another substantial sequential increase.

Meeting that forecast would show that second-quarter momentum continued. Exceeding it through documented custom and connectivity growth would provide stronger evidence for the longer-term outlook.

A result near the low end would not automatically invalidate fiscal 2028 expectations. It would, however, increase scrutiny of program timing and the promised second-half acceleration.

Gross margin will matter alongside revenue. Rapid growth that requires an unfavorable product mix or elevated ramp costs can generate less operating leverage than the headline sales figure suggests.

The third signal is evidence of production-scale custom silicon adoption. Marvell has discussed strong bookings and customer design activity, but recognized revenue remains the clearest measure.

Readers should look for disclosures showing that multiple custom programs have progressed from engineering into volume manufacturing. Customer concentration should also be monitored as these programs expand.

Broadcom’s results offer another reference point. Continued strength there would support the broader custom silicon market while maintaining pressure on Marvell to establish its position.

Nvidia’s platform decisions also matter. Marvell benefits if interoperability lets its custom XPUs and networking products participate in more Nvidia-centered deployments.

A shift toward tighter proprietary integration could weaken that opportunity. More open interfaces and wider component qualification would strengthen it.

These three signals form a practical test of the original Google News story. Investor Day must clarify the target, the third quarter must sustain the ramp, and custom programs must reach production.

Marvell has already established that AI demand can lift current revenue and profit. Its second-quarter numbers provide clear evidence of that change.

The unresolved question is whether the company can turn a collection of promising technologies into a coordinated, durable growth engine. That requires delivering custom chips, optical links, and switches on customer schedules.

Watch the revenue mix rather than the number of announcements. If custom silicon and connectivity grow together across several customers, Marvell’s raised targets will gain credibility.

If disclosure remains broad while concentration rises, the market’s skeptical reaction will look more understandable. The next Google News headline should matter less than the operating evidence behind it.