Marvell Technology

Marvell Technology reported good July-quarter results and raised its guidance for the next two years. Sales rose 37% year on year to $2.74 billion, and fiscal 2028 guidance went up nearly 10% to $18 billion. Management previewed further strong growth past fiscal 2028.

Why it matters: Marvell is flexing its differentiated portfolio of silicon across the data center and in artificial intelligence. We like the broad-based strengthening of demand across custom chips, interconnect, and switching. The recently announced long-term agreement with Google adds further upside to long-term growth. We expect growth to accelerate through calendar 2028, led by rising share of custom silicon for AI customers. Marvell now holds three blue-chip custom ASIC customers in Amazon, Microsoft, and Google, and XPU attach is a meaningful contributor to growth that reflects its broad portfolio. Marvell is positioned well for future technology transitions, including a secular shift to customized chips for AI, higher bandwidth needs, and the proliferation of optics. It is highly leveraged to AI spending, where we see it gaining material share over the medium term.

The bottom line: We raise our fair value estimate for narrow-moat Marvell to $300 per share, from $270, behind a higher medium-term growth forecast. Despite meaningful raises to guidance, the market looks unimpressed with a 6% selloff after hours. We see good value for long-term shareholders here. Shares have nearly tripled year to date but remain well below the June 2026 high. Even after the rise, we believe the market is overdiscounting Marvell's long-term growth past 2028, especially with the $120 billion revenue opportunity from Google.

Coming up: We see Marvell's Oct. 6 investor day as a potential catalyst to the stock. We expect longer-term targets through calendar 2030 and more detail on the Google ramp. There could be further upside to our model, as we don't model in the full $120 billion from Google through 2032.

BLANK PAGESpecifically, management raised its fiscal 2027 (calendar 2026) guidance to $12 billion in revenue, from $11.5 billion, implying nearly 50% growth. For fiscal 2028, $18 billion revenue guidance rose from $16.5 billion previously and implies another year of 50% growth. We model upside to both figures. Both are underpinned by 60% expected data center revenue growth, which we expect to be led by custom silicon sales and interconnect. We model custom revenue nearing $11 billion in calendar 2028, which would be more than 5 times the level we expect in calendar 2026 and is above management's "more than $10 billion" outlook. Starting next year, we expect scale-up connectivity sales to be a material growth driver, rising to $3 billion in calendar 2030 from zero today.

This was the first call with new CFO Dan Durn, with whom we expect continuity. Durn highlighted focuses on scaling efficiently and expanding operating margin and cash flow as well as supplying further returns to shareholders. We like Durn's steep experience in semiconductor and technology finance, with an illustrious background, including positions at Applied Materials and Adobe. We expect strong expansion to both margin and cash flow over the medium term under his tenure.

Our fair value estimate for Marvell Technology is $300 per share. Our valuation implies a fiscal 2027 price/ adjusted earnings of 70 times and an enterprise value to sales of 21 times, along with a 1% free cash flow yield. Against our estimates for fiscal 2028 and 2029 adjusted earnings, our valuation implies multiples of 41 times and 24 times, respectively. The primary driver of our valuation is growth in data center revenue.

We expect 44% sales growth for Marvell over the next five years, primarily from organic share gains as it flexes its portfolio for growing applications in data centers and artificial intelligence. We expect it to take organic share across most of its end markets, but for its largest growth to come from custom compute chips and interconnect chips for optical connectivity.

We anticipate the data center to remain Marvell’s largest end market, and it is the primary driver of our forecast with a 51% compound annual growth rate through fiscal 2031. We estimate Marvell will gain significant wallet share at customers as cloud customers adopt higher levels of custom chip and optical content to advance to higher speeds and higher-performing AI models efficiently. We think Marvell will be a significant beneficiary of AI spending, given its leadership in optical chips and custom chip design. AI contributes meaningfully to our growth forecast, and we forecast it to be the majority of data center revenue going forward.

We see custom chips as one of the firm’s largest growth drivers, with revenue expected to grow at a 70% clip over the next five years. We see this growth as a function of rising custom chip volumes and increasing customer counts. We observe strong penetration in Amazon Web Services’ Trainium chips, along with significant future revenue from Microsoft’s Maia chips starting in fiscal 2028, as the main contributors to this growth. In the longer term, we expect Marvell’s penetration of Google’s TPU ecosystem to maintain high growth. We forecast optical connectivity chips to grow at a 45% annualized rate over our five-year forecast. In optical, we believe Marvell’s leadership position in scale-out connectivity will benefit from increasing investments in networking to enable AI model performance. We also expect a strong ramp for co-packaged optics revenue, supported by Marvell’s 2026 acquisition of Celestial AI, to pad growth over the next five years.

Non-data center revenues become a small minority of total sales over our forecast, and are led by merchant networking chips sold into enterprises and wireless processors sold into carrier networks. We believe both businesses are recovering nicely from severe cyclical downturns in fiscal 2025, and we expect them to be back to midcycle by the end of fiscal 2027. From there, we expect mid-single-digit growth, aligned with broader on-premises IT spending and carrier investment.

We estimate Marvell’s end markets to have gross margin profiles tightly grouped around the corporate average. The standout is the custom chip business, which we estimate to be closer to a 50% gross margin. We expect non-GAAP gross margins to remain in the high-50% range in the long run. A rising mix of custom silicon sales for AI will weigh on gross margins versus historical levels, but we note that these are actually operating margin-accretive via strong co-investment from customers that registers as contra-operating expenses.

We forecast strong reinvestment in research and development for Marvell, but see meaningful potential for operating leverage against significant revenue growth and a relatively concentrated customer base for custom chips. Marvell has guided to close to 40% non-GAAP operating margin in fiscal 2028, and we see non-GAAP operating margins expanding further, reaching the high-40% range in fiscal 2031, well above the 35% level from fiscal 2025.

We assign Marvell Technology a narrow economic moat rating. In our view, Marvell holds intangible assets in networking chip design that enable it to compete at the cutting edge and defend its competitive position from well-capitalized competition, and also benefits from switching costs. We expect Marvell to earn excess returns on invested capital, more likely than not, over the next 10 years.

To us, intangible assets in networking chip design come in the form of engineering expertise, both in terms of silicon design and integration with complementary hardware and customer networking topologies, which results from decades of development, R&D expense, and engraved customer relationships. In our view, Marvell’s billions of dollars of cumulative R&D over the past decade have created a portfolio of differentiated intellectual property from which it can draw to build custom and semi-custom designs for myriad applications and customers.

We contend that Marvell commands pricing power for its differentiated technology and design. We believe that customers pay for performance and features in their networking chips, and Marvell’s pricing power shows in its healthy gross margins as it aggressively adds new sockets and expands within existing customers. Marvell’s customers have shown a commitment to build greater portions of their networks on its chips, reflected in strong cross-selling and upselling activity. Marvell’s $100-plus million customer count has risen quickly from 13 in 2020 to 19 today. These customers include the likes of Microsoft, which sole-sources Marvell’s optical chips for its North American regional data centers.

New chip generations with higher speeds allow networking equipment providers and end customers to improve network speed and performance, and transfer, store, and process more data. Beyond speed, performance as defined by a customer is affected by a wide variety of features like energy efficiency, memory, security, and specific network design characteristics. Marvell’s products are either tailored to a specific application within an end market, or semi-custom chips fine-tuned for an end customer’s networking architecture to allow easier software integration and desired feature sets. Marvell’s customers build networks according to its silicon feature sets, which creates an added level of stickiness, and design wins can last three to five years. As an incumbent, Marvell is in constant conversation with customers about roadmaps for their next technology generation and how to fine-tune a chip to future networking needs. Marvell’s most cutting-edge designs involve co-design from its large customers, which we see as a virtuous feedback loop of investment and IP that would be difficult for a new entrant to overcome.

Going forward, we believe networking technologies like switching, optical, and processing will become more intertwined. We see one example of this being switch chips that feature co-packaged optics technology and enable higher speeds. In our view, Marvell is one of a small handful of suppliers that has strong technology on each front and the breadth with which to serve several distinct end markets and applications—a position it has taken more than a decade and billions in R&D to develop. Though we see technological strength and differentiation in the firm’s switches, optical chips, and processors alone, we see even greater differentiation when combined into a tailored solution for an end market or specific customer. We believe higher levels of customization will elicit switching costs at customers, and believe Marvell may even generate R&D synergies in the future as technologies become intertwined. We expect Marvell’s organic investment to stay high over the long-term, keeping it at the forefront of design while pricing power maintains healthy profitability.

We see room for multiple moats in this market with established existing players, but we see significant barriers to entry for new entrants—in terms of investment, technological breadth, and customer relationships. We view moats in networking silicon as being carved by technological depth and breadth. Broadcom, Marvell, Intel, and Nvidia have all devoted years and billions of dollars toward organic development of technology and customer relationships, as well as consistent acquisition activity to build out broader portfolios. Broadcom dominates this market but has been outpaced in growth by the likes of Marvell, Intel, and Nvidia behind higher investment and penetration of processing and optical technologies. We surmise that Marvell’s optical, switching, and processing products are among the market leaders, and we also suspect that Marvell’s investments in networking chips even surpasses Intel and Nvidia today.

We believe new entrants would struggle to develop a sufficiently deep and wide portfolio to compete for high-value design wins, especially without existing customer relationships. Even a well-capitalized networking vendor in Cisco has struggled to make material market share gains with its Silicon One chips, with what share it does have coming primarily from integration into its walled garden ecosystem. Cisco showed that heavy dollar investment could lead to parity in terms of raw speed, but that wasn’t enough to generate meaningful share gains without broader features and complementary technologies with which to accompany it. A new entrant without Cisco’s budget and existing relationships within networking would have an even steeper hill to climb, in our view, with the years of expense and development required not threatening Marvell’s moat.

Marvell’s strong position in optical chips and its burgeoning custom chip business offer a strong foothold into generative AI infrastructure, which should fuel high growth.

Marvell’s strong non-GAAP profitability reflects moaty pricing power, in our view.

We believe Marvell’s wide portfolio of switches, processors, and optical chips gives it ample opportunity for cross-selling and share gains.

Marvell is significantly smaller than competitors like Broadcom and Nvidia, which risks the firm being unable to invest adequately to keep up with these massive peers.

Marvell is highly concentrated in data centers, which makes it highly sensitive to the rate of data center and AI spending. Deceleration or corrections in the rate of spending would mean downside to growth and valuation.

Marvell is a frequent acquirer, which risks the firm overextending itself or overpaying in the future and diluting shareholder value.