Marvell (Minutes): Still has not responded positively to Alchip's challenge
The following are the minutes of Marvell Technology's Q2 FY2026 earnings call. For an interpretation of the earnings report, please refer to "Marvell Technology: Why Didn't ASIC Benefit from the Generous Spending of Cloud Giants?"
I. $Marvell Tech(MRVL.US) Key Financial Information Review
1. Capital Deployment and Investment
The company has completed the divestiture of its automotive Ethernet business, providing $2.5 billion in cash, which offers flexibility to continue its stock repurchase program and further invest in technological capabilities to strengthen its technology platform. The company strategically shifts its investment focus to data centers to seize the significant AI opportunity.
2. Cash Flow
In Q2 FY2026, the company achieved $462 million in operating cash flow, a significant increase from $333 million in Q1.
3. Shareholder Return Plan
- In the first half of the fiscal year, the company repurchased $540 million in stock, with approximately $2 billion remaining in the authorized amount.
- In the second quarter, the company returned $52 million to shareholders through cash dividends and repurchased $200 million in stock.
- The $2.5 billion in cash from the divestiture of the automotive Ethernet business will continue to support the stock repurchase program.
4. Personnel Adjustment Plan
- The company has strengthened its leadership structure by promoting two leaders: Chris Koopmans has been promoted to President and Chief Operating Officer, responsible for sales, non-data center businesses, corporate development, end-to-end revenue execution, market strategy, customer engagement, operations, and long-term strategic planning.
- Sandeep Bharati has been promoted to President of the Data Center Business Unit, fully responsible for the data center business and continuing to lead data center engineering and central engineering, covering technology platforms, IP, roadmaps, customer engagement, product definition, and chip development.
5. Performance Guidance
1) Overall Outlook for Q3 FY2026:
- Total revenue is expected to be $2.06 billion at the midpoint, representing a year-over-year growth of approximately 36%.
- Excluding automotive Ethernet business revenue, the implied revenue growth rate for continuing operations is expected to be close to 40%.
- Non-GAAP diluted EPS is expected to be between $0.69 and $0.79, representing a sequential growth of approximately 10%.
- Non-GAAP gross margin is expected to be between 59.5% and 60%.
2) End Market Guidance:
- Data Center End Market: Expected to be flat sequentially but maintain strong year-over-year growth in the mid-30% range. The double-digit growth in electro-optical interconnect products will offset the short-term decline in custom ASIC business, with electro-optical product revenue expected to achieve consecutive double-digit percentage growth. Although the custom ASIC business will decline sequentially in Q3, it will rebound in Q4, and the overall performance in the second half will still be better than the first half.
- Enterprise Networking and Carrier Infrastructure End Market: Total revenue is expected to grow approximately 30% sequentially. These two end markets are expected to generate approximately $2 billion in annual revenue combined.
- Consumer End Market: Revenue is expected to decline by a low single-digit percentage sequentially. Annual revenue is expected to be approximately $300 million.
- Automotive and Industrial End Market: Total revenue is expected to be approximately $35 million, including a mid-single-digit million-dollar contribution from the automotive Ethernet business before divestiture. After the divestiture of the automotive Ethernet business, the industrial end market is expected to generate approximately $100 million in annual revenue.
3) Long-term Goals: The company is committed to achieving strong revenue growth and moving towards its long-term goal of expanding operating profit margins.
II. Detailed Content of Marvell Technology's Earnings Call
2.1 Key Information from Executive Statements
1. End Market Focus and Strategy
1) Data Center End Market
a. Market Drivers and Focus: Business strength is mainly due to custom XPU, XPU-attached products, and electro-optical interconnect products, with AI and cloud contributing over 90% of data center revenue;
b. Custom Chip Business: Secured adoption of 18+ multi-generational slot designs, some in mass production, with over 50 new potential opportunities, estimated to have a $75 billion lifecycle revenue potential. The goal is to increase our data center market share from 13% of a $33 billion TAM in 2024 to 20% of a $94 billion TAM in 2028.
[Note: Marvell has deployed 12 slots (3 XPUs and 9 Attach chips) among the four major cloud providers, with an additional 6 slots in emerging hyperscale providers, totaling 18 AI acceleration-related chips]
c. Horizontal Expansion in Networking and Interconnect: Marvell leads in AI interconnect, benefiting from strong demand for 800G PAM DSP and has shipped the next-generation 200G/channel 1.6T PAM DSP in volume. The company demonstrated 400G/channel PAM technology, a key innovation for achieving 3.2T optical interconnect.
d. Storage, Switching, and Security: Data center storage revenue has significantly improved. 12.8T switching products continue to ship in volume, with the next-generation 51.2T switch accelerating mass production, expected to be the main driver of switch revenue growth in the next fiscal year. In security, expanded collaboration with Microsoft Azure on hardware security modules.
2) Enterprise Networking and Carrier Infrastructure End Market
Recovery Drivers: Recovery is driven by customer inventory normalization and strong adoption of updated product portfolios (migrated to advanced process nodes). These two end markets are expected to generate approximately $2 billion in annual revenue combined.
3) Consumer End Market
Main Drivers: This business is mainly driven by gaming demand and its seasonal impact. Annual revenue is expected to be approximately $300 million.
4) Automotive and Industrial End Market
Business Divestiture and Focus: The company has completed the divestiture of its automotive Ethernet business, obtaining $2.5 billion in cash, aiming to shift investment towards data centers to seize AI opportunities. After the divestiture, the industrial end market is expected to generate approximately $100 million in annual revenue.
2. Product and Technology Innovation
1) Custom Chips and Advanced Processes: Successfully delivered multiple highly complex custom XPU and XPU-attached product projects, achieving mass production. Played a key role in propelling Marvell to a leadership position in 5nm process technology.
2) Optical Interconnect Frontiers: Demonstrated 400G/channel PAM technology, an important step towards achieving 3.2T optical interconnect.
3) Switch Technology Progress: 12.8T products are shipping in volume, with the next-generation 51.2T switch accelerating mass production.
4) Product Portfolio Update: Enterprise networking and carrier infrastructure products have migrated to advanced process nodes.
3. Organizational Structure and Market Outlook
a. Leadership Restructuring: Promoted Chris Koopmans to President and Chief Operating Officer, and Sandeep Bharati to President of the Data Center Business Unit, to further seize significant opportunities in the AI and cloud markets.
b. Future Reporting Structure: Starting from the next quarter (Q3 FY2026), non-data center end markets will be consolidated into a new "Communications and Other End Markets," while the data center end market composition will remain unchanged.
c. Market Position and Opportunities: The data center has contributed three-quarters of the company's total revenue (74% in the second quarter), solidifying its position as a leading supplier of data infrastructure semiconductors. The company is optimistic about short-term and long-term growth prospects, with custom AI design projects at an all-time high, and the rise of Scale-Up networks providing strong momentum.
2.2 Q&A Session
Q: Could you elaborate on the guidance for the custom ASIC business? Specifically, what headwinds are faced in the third quarter? Also, what is your outlook on the growth confidence and magnitude for the custom business in the fourth quarter?
A: The custom business indeed faces uncertainties, especially in the early stages of large hyperscale data center projects, which is a common phenomenon. The third-quarter business is affected by the product digestion period, mainly requiring some time for adjustment after project launch. However, we have strong confidence in the custom business for the fourth quarter. The optical business is expected to achieve double-digit growth. Meanwhile, demand for custom business is picking up. Overall, the custom ASIC business is expected to perform better in the second half than in the first half, leading to strong growth in the fourth quarter.
Q: Could you provide more details on the current design adoption situation, and how much of the custom product revenue in the second half is expected to come from these new projects versus existing design adoption projects?
A: The data center custom chip business is in an exciting period. Design activity is at an unprecedented level, covering XPU, XPU Attach, and both emerging and existing hyperscale data centers. Since June, the lifecycle design adoption project value for XPU Attach opportunities has increased to several billion dollars. These new design adoption projects are significant, enhancing our confidence in achieving the 20% market share target in a rapidly growing market. Regarding revenue composition, the company has not explicitly divided the contribution between new and existing projects but emphasized the strong momentum of overall design activity.
Q: Is the company facing any supply constraints in the supply chain? Also, have tariff policies impacted the business?
A: The supply chain is indeed very tight, requiring close coordination with customers and strong team execution. Despite the tight situation, we have the capability to meet all customer demands and are confident in our future response capabilities. Regarding tariffs, while the environment is dynamic and ever-changing, we have not observed any significant impact on the company's business so far. The company is closely monitoring the tariff situation but has not found any significant impact in various end markets.
Q: Considering the uncertainty in the custom XPU business, how do you view the concentration of major customers? How will these additional design adoptions integrate into XPU revenue over the next 6 to 12 months?
A: The initial core AI projects we started a few years ago have begun mass production, despite short-term uncertainties. Additionally, the 18 slots mentioned at AI Day will gradually be put into use over the next 18 to 24 months. The company has received additional orders, bringing the total to over 18. This development path aims to increase market share from 10% in FY2023 to 13% the following year, and ultimately to 20%. The unprecedented design activity recently has enhanced the company's confidence in achieving these goals.
Q: NVIDIA mentioned the concept of "horizontal scaling" or "cross-network" this week. How does Marvell view the opportunity of "cross-data center interconnect" (DCI)? What growth potential could this opportunity bring to Marvell?
A: Marvell is committed to leveraging its networking and connectivity technology to address the emerging Scale-up opportunities. In addition to proprietary architectures from major GPU manufacturers, there is significant demand for dedicated architectures like Ethernet and UALINK. The company is heavily investing in bringing related switches to market, expecting strong momentum in the coming years. By leveraging the most advanced low-latency switch IP portfolio (partially derived from the Innovium acquisition), Marvell is confident that this business will become a key growth driver.
Q: Do you think data center growth in the fourth quarter can accelerate year-over-year from the third quarter level?
A: We typically only provide quarterly performance guidance, not annual guidance. The custom business is expected to perform strongly in the fourth quarter. The overall performance of the custom business in the second half is expected to be better than the first half. Additionally, the optical business performed strongly in the second and third quarters, particularly achieving double-digit growth in the third quarter. The core business of enterprise networking and carriers also shows strong recovery, with annualized revenue rebounding from approximately $900 million to approximately $1.7 billion in the third quarter guidance. These factors will provide favorable support for the data center business in the fourth quarter.
Q: When looking ahead to 2026, a competitor in the XPU space mentioned their business could grow by 60%, and Jensen also mentioned around 50%. Does Marvell currently have enough visibility on the timing and scale of large projects to align with industry expectations for growth, or are there other pros and cons to consider?
A: The company typically does not provide annual guidance, only in very few cases and with higher visibility at the end of the year. Nonetheless, the overall momentum of the custom business has been strong for several quarters. The custom business in the second half is expected to be better than the first half, with the optical business achieving double-digit growth in the third quarter. Additionally, the core business of enterprise networking and carriers is experiencing a strong recovery, with annualized revenue rebounding from approximately $900 million to approximately $1.7 billion. These positive factors collectively suggest that the company is likely to achieve growth in line with industry expectations in 2026.
Q: Regarding the "digestion period" for the custom business in the October quarter, does this mean one project is shrinking while another restarts in the fourth quarter? Is this just a temporary pause, or is it related to product transition?
A: The current "digestion period" mainly involves existing projects, not project shrinkage. It is essentially a matter of product delivery timing and customer build and demand timing. We are still in the early stages of the custom business, which is our first significant year, and the initial few slots will gradually expand. As more projects start, the diversity of this business within the company will significantly increase. It is merely a matter of timing between quarters and is not directly related to product transition.
Q: Regarding the optical business, you mentioned achieving double-digit growth in the October quarter. Are there any issues with mass production or supply chain, such as supply constraints in lasers?
A: Over the past few years, we have successfully scaled up the optical business. The company has established deep partnerships with upstream and downstream supply chain and key module companies to plan the business together. While there are always some "noise" and issues systematically, we have consistently been able to effectively address and overcome challenges, achieving significant growth. Therefore, despite potential supply constraints or component issues in the industry, we have maintained relatively unaffected product mass production through strong partnerships and excellent execution.
Q: You expect data center to be flat while optical business grows double digits. What is the proportion of optical business? Will AI revenue reach half of the company's total revenue in Q3 FY2026? Do optical and custom businesses account for 50% of the company's total revenue?
A: In the fourth quarter, the optical business accounts for about half of the company's total revenue, the custom business about a quarter, and other businesses about 25%. Although both optical and custom businesses have grown since then, the company has not updated the exact composition ratio. While unable to provide precise quarterly update numbers, considering the custom business performs better in the second half than the first half, and the optical business is strong, the direction of AI revenue reaching half of the company's total revenue has not changed.
Q: The company mentioned having secured 18 design adoptions, and then seemed to indicate a few more were added. Among all the custom chips and attached chips Marvell is developing, how many are generating revenue? What stage are we currently at?
A: Currently, multiple products are in production and have been contributing revenue since the end of last year. Among the 18 projects mentioned by the company, some are in production, and some will start this year or next. We see new batches of such projects going into production almost every quarter. This revenue contribution is expected to continue growing over time. The company is committed to achieving a 20% share in a $94 billion market by 2028.
Q: Are most or all of Marvell's orders proceeding as planned? Because there is a lot of debate when we communicate with customers and investors.
A: There is always debate in the market. The main motivation for the company's AI Day was to elaborate on business development direction, technological differentiation, and opportunity sets, and for the first time, to break down the relative scale of hyperscale and emerging, XPU and XPU Attach opportunities in detail. Given the strong momentum of current design adoptions, the company is continuously gaining new incremental business from both traditional large hyperscale data centers and the emerging generation. Therefore, despite the debate, the company is proceeding as planned and continues to gain new business. (Still not clearly responding)
Q: Regarding capital allocation, could you discuss the expected use of proceeds from the sale of the automotive Ethernet business? Is there a preference for supplementary acquisitions to accelerate the AI strategy, or for stock repurchases? More broadly, is the company willing to sell other parts of the business at the right price?
A: Since August 2016, the company has implemented a strategic process to prioritize R&D funding in the data center and AI fields. Currently, over 80% of R&D spending is invested in AI and data center areas, up from 60% a few years ago. The deployment of proceeds from the automotive business divestiture has not been decided but remains targeted at the AI field or related acquisitions. The company has also increased repurchase levels through ongoing free cash flow and plans to continue advancing. At this historic moment in the AI market, the company will use this additional capital for opportunistic repurchases and leverage it when finding supplementary acquisitions that can accelerate the roadmap. Currently, the company does not comment on whether it is willing to sell other business units.
Q: Regarding the follow-up 3nm XPU project for major customers, despite rumors from Asia, how is Marvell's 3nm XPU follow-up project progressing? Is it still on schedule, and how confident are you in achieving growth next year? How is the progress of the third 3nm XPU customer order?
A: We understand the various rumors in the market about the 3nm XPU project. The company's initial projects and design-adopted orders are gradually going into production. We have significantly increased the opportunity set from the initial few slots to over 18 and are working towards future market share goals. Given the high attention and sensitivity in this field, commenting on a single slot at this time may add market noise. The company's core focus is on winning core designs, executing existing projects, driving business development, and ultimately capturing 20% of the potential $90 billion market in the future. (Still no clear response on the progress of new products and Alchip's challenge)
Q: Regarding the scale-up switching structure opportunity, it seems to be a larger part of the XPU attach market. When will Marvell's first product generate revenue? Will this be in the 2026 calendar year, or is it more inclined towards UALink, or in the 2027 calendar year?
A: The core advantage of scale-up lies in the excellent integration of Marvell's multiple key intellectual properties (IP), especially low-latency switch IP, serializer/deserializer (SerDes) technology, coupled with the ecosystem built around XPU (heterogeneous computing units) — this business itself belongs to the critical XPU attach (XPU Attach) area, essentially almost equivalent to a strategic layout at the chipset level.
The company is actively developing products designed for scale-up scenarios, for UA links and Ethernet, to be launched within the next two years. Additionally, the company is actively participating in the active cable/optical cable markets such as AEC and AOC in interconnect.
Q: You have a large business in DSP-based optical modules. Some peers have noticed that three hyperscale data centers are promoting linear direct-drive optical modules (LPO modules). Could you describe whether you see a significant amount of LPO development emerging? Are they niche applications? What is the penetration rate of LPO in the overall optical transceiver market? Will it remain at a low single-digit percentage, or will it become larger in the coming years?
A: The development and application of LPO modules are proceeding on a smaller scale, and Marvell is involved, with active orders and production. However, given the large market size of DSP-based pluggable modules, LPO modules are currently just a very small number, more like a niche use case. If customers indeed require LPO and effectively implement it in production, it can bring certain benefits. But for now and the foreseeable future, the vast majority of the market will still be pluggable modules, and LPO penetration is likely to remain at a low level.
<End Here>
Risk Disclosure and Statement of this Article:Dolphin Research Disclaimer and General Disclosure
