Maivel conference call: ASIC accounts for 25% of data center business revenue, AI revenue will significantly exceed the $2.5 billion target in fiscal year 2026

On March 5th, Eastern Time, after the US stock market closed, Marvell released its fourth-quarter financial report, with revenue increasing by 27% year-on-year to $1.82 billion, with data centers being the largest growth point. However, a merely adequate financial report is not enough to meet investors' high expectations for AI chip hot stocks, and Marvell Technology saw a sharp decline in after-hours trading. For the first quarter of the fiscal year ending in April, Marvell expects sales of approximately $1.88 billion. Although this is in line with analysts' average expectations, some optimistic forecasts go as high as $2 billion. After the financial report was released, Marvell's stock price fell by as much as 16% in after-hours trading, and has accumulated a decline of 18% so far this year. Here are the key points from Marvell's earnings call: The company made significant progress in fiscal year 2025, with substantial revenue growth, operational leverage in its business model, GAAP profitability in the fourth quarter, and strong performance in the data center standard cloud infrastructure business, with a recovery in sequential growth in multi-market businesses such as enterprise and carrier services. In the fourth quarter, ASIC accounted for 25% of data center business revenue, and AI revenue for fiscal year 2026 is expected to significantly exceed the $2.5 billion target. Inventory increased by 20% quarter-on-quarter, mainly to support the growth of custom projects and strong development in the optical business, with inventory days remaining flat quarter-on-quarter. The company is steadily moving towards its goal of capturing 20% of the global data center market share, while the total market size is also developing towards $75 billion, with current growth rates exceeding expectations. The company's optical business is strong, with a significant increase in orders in the second half of last year and robust demand this year, with healthy demand for 800 Gigabit products transitioning to 1.6 Terabits, and 3-nanometer products gaining attention due to energy-saving advantages. To further optimize AI interconnect performance, the company will accelerate the launch pace of next-generation products, introducing the industry's first 3-nanometer 1.6T DSP, featuring 200 Gigabit electrical and optical interfaces per channel, which can reduce the power consumption of 1.6T optical modules by over 20% compared to previous generations. With the accelerated adoption of 1.6T technology, mass production is expected in the second half of this year. Custom HBM is the company's key intellectual property and a core part of the customer roadmap. The company is actively collaborating with major memory partners and large-scale data center customers to achieve significant advantages, including reducing logic chip area, enhancing computing integration, and increasing HBM and computing throughput, providing higher density memory solutions. The company has secured follow-up AI XPU projects, which are large-scale projects and extensions of existing business. Due to confidentiality, specific details such as timing and processes cannot be disclosed, but the company is confident in successfully advancing projects with customers. Below is the full transcript of the call (translated by AI): Host: Good afternoon, and welcome to the Marvell Technology, Inc. fiscal year 2025 fourth quarter and full-year earnings call. All participants will be in listen-only mode. After today's presentation, there will be an opportunity for questions. Please note that this event is being recorded. Now, I would like to turn the meeting over to Mr. Ashish Saran, Senior Vice President of Investor Relations Senior Vice President of Investor Relations Ashish Saran: Thank you, good afternoon everyone. Welcome to the conference call for Marvell Technology's Q4 and full-year fiscal 2025 earnings report. Joining me today are Marvell Technology's Chairman and CEO Matt Murphy, as well as our Chief Financial Officer Willem Meintjes. I would like to remind everyone that some comments today contain forward-looking statements, which involve significant risks and uncertainties that could cause our actual results to differ materially from management's current expectations. Please refer to the cautionary statements and risk factors included in our earnings press release, which we submitted to the U.S. Securities and Exchange Commission (SEC) today and published on our website, and can also be found in our latest 10-K and 10-Q filings. We do not intend to update our forward-looking statements. In today's conference call, we will mention certain non-GAAP financial metrics. Our earnings press release also provides reconciliation information between GAAP and non-GAAP financial metrics. Now, I will turn the call over to Matt to discuss his views on the quarter. Matt? Chairman and CEO Matt Murphy: Thank you, Ashish, good afternoon everyone. In Q4 of fiscal 2025, Marvell Technology achieved record revenues of $1.817 billion, exceeding the midpoint of our guidance, with a quarter-over-quarter growth of 20% and a year-over-year growth of 27%. Our data center end market was the primary growth driver, benefiting from strong demand and execution in artificial intelligence. Additionally, we are seeing a continued recovery in demand across multiple market segments, including carriers, enterprise, networking, as well as automotive and industrial sectors. I am pleased to report that we achieved profitability under GAAP in Q4 and expect this to continue into fiscal 2026. Our record non-GAAP earnings per share of $0.60 exceeded the midpoint of our guidance, with a quarter-over-quarter growth of 40%. This earnings growth rate is double our revenue growth rate, highlighting the strong operating leverage in our business model. For the entire fiscal 2025, our total revenue reached $5.77 billion, with data center revenue growing 88% year-over-year. By the end of this fiscal year, our AI revenue significantly exceeded the $1.5 billion target set on AI Day in April 2024. Moreover, we expect to substantially exceed the $2.5 billion target in fiscal 2026. Driven by the advancement of customized chip projects and the continued strong growth of our optoelectronics business, our overall revenue growth accelerated in the second half of the year. In fiscal 2025, we also achieved record operating cash flow of $1.68 billion and significantly increased capital returns to shareholders through stock buybacks and dividends, totaling $933 million. I am extremely satisfied with our performance in fiscal 2025 and even more excited about the prospects for strong year-over-year revenue growth in fiscal 2026 We are well-prepared for a strong start to the new year, expecting first-quarter revenue to grow more than 60% year-over-year based on the midpoint of our guidance. Now, let me discuss the performance and expectations of our various end markets. In the data center end market, we achieved a record revenue of $1.37 billion in the fourth quarter, a year-over-year increase of 78% and a quarter-over-quarter increase of 24%. These outstanding results were driven by the large-scale production of our customized artificial intelligence chip projects. Additionally, we benefited from strong shipments of optoelectronic products and Teralynx Ethernet switches, both of which achieved double-digit percentage growth in revenue quarter-over-quarter. In our optoelectronics business, we continue to see strong demand in the market for our leading 800 Gigabit Pulse Amplitude Modulation (PAM) products and 400ZR Data Center Interconnect (DCI) products. We have also begun shipping the industry's first 1.6T PAM Digital Signal Processor (DSP) and products using 5-nanometer process technology. To further optimize artificial intelligence interconnect performance, we have accelerated the launch of next-generation products, introducing the industry's first 1.6T DSP using 3-nanometer process technology, featuring electrical and optical interfaces of 200 Gigabits per channel. This new Maywell Technology DSP enables customers to reduce the power consumption of 1.6T optical modules by more than 20% compared to previous generations. With the accelerated adoption of 1.6T technology, we expect to achieve mass production in the second half of this year. While there has been significant attention on Graphics Processing Units (GPUs) and Accelerated Processing Units (XPUs), the distributed nature of computing and artificial intelligence makes connectivity as crucial as individual processors. Therefore, the design of next-generation accelerated infrastructure is closely related to the ingress and egress of data on accelerators and efficient transmission across the entire cluster. The enhanced role of high-speed networking in artificial intelligence data centers perfectly aligns with our advantages as an industry leader. Similar to the computing field, hyperscale data center operators are also customizing their networks, and we see similar momentum in flash-based storage, High Bandwidth Memory (HBM), and pooled Dynamic Random Access Memory (DRAM) based on Compute Express Link (CXL). In October last year, we announced that we won the design order for Meta's customized Network Interface Cards (NICs) at the Open Compute Project (OCP), and we see hyperscale data center operators adopting similar strategies more broadly. This is reflected in our recently won design orders, which include multiple customized NICs and subsequent customized CXL memory solutions. We are also enabling new interconnect technologies for scalable architectures, such as Co-Packaged Optics (CPO) and Linear Photonics (LPO), as well as coherent optical DSPs for emerging campus-scale large-scale artificial intelligence data centers. Earlier this year, we announced Maywell Technology's groundbreaking co-packaged optical architecture for customized XPUs, allowing customers to integrate optical components into future customized accelerators. The core of the Maywell Technology CPO platform, showcased for the first time at the 2024 Optical Fiber Communication Conference and Exhibition (OFC), is our 6.4T three-dimensional silicon photonic engine It integrates hundreds of active and passive components into a unified device, built on the foundation of multiple generations of silicon photonic innovation technologies that we have shipped in large quantities in DCI modules over the years. Co-packaged optics can scale up the size and dimensions of artificial intelligence servers that currently rely on passive copper interconnects. We expect the transition from copper interconnects to optical interconnects to significantly expand Marvell Technology's interconnect business revenue and market opportunities. We are working with customers to evaluate this advanced technology and anticipate a multi-year trial system development phase before large-scale industry applications of CPO. Now let me talk about our current custom chip projects. Marvell Technology has successfully advanced highly complex XPUs and CPUs, with over 100 billion transistors, from initial samples to large-scale production after a successful tape-out. As customers increasingly rely on Marvell Technology to help them achieve their custom chip goals, our custom business continues to gain momentum. As I mentioned, our two leading artificial intelligence custom projects have achieved large-scale production. We expect growth to continue. One of them is a custom ARM CPU, which we anticipate will see broader application in customers' data centers. The second project is a custom artificial intelligence XPU, which has also performed exceptionally well and is expected to see significant mass production in the future. Meanwhile, we are collaborating comprehensively with this customer on the next-generation product of this XPU and plan to ramp up mass production after completing the sample and validation cycles. Therefore, we expect custom XPU revenue from this customer to grow not only in fiscal year 2026 (this year) but also to continue growing in fiscal year 2027 and beyond. Additionally, we have made significant progress on the new design order for a custom artificial intelligence XPU for another U.S. hyperscale data center operator announced during our AI Day in April 2024. Marvell Technology's engineering team has worked closely with the customer, successfully completing multiple key technical milestones during the joint development process. Therefore, we are confident that we are fully on track to begin mass production in the calendar year 2026 according to the customer's expected timeline. This collaboration is also multi-generational, and we expect it to bring substantial incremental revenue to Marvell Technology in the coming years. I am very pleased with our achievements in custom business revenue for fiscal year 2025, thanks to the advancement of multiple projects. This success, along with the strong progress of upcoming custom projects, gives us greater confidence in achieving our long-term market share goals for custom business revenue. We continue to invest in various aspects of our technology platform, including advanced process nodes, electrical and optical serializers/deserializers (SerDes), high-speed chip-to-chip interconnects, embedded memory, custom HBM, 2.5D and 3D packaging, and silicon photonic technology. This week, we announced the demonstration of the industry's first 2-nanometer silicon intellectual property (IP) for next-generation artificial intelligence and cloud infrastructure. This usable silicon chip, produced on TSMC's 2-nanometer process, is a key part of Marvell Technology's development of custom XPUs, CPUs, switches, and other technology platforms critical to next-generation accelerator workloads Now let me talk about our outlook for the data center market in the first quarter of fiscal year 2026. We expect the cloud and artificial intelligence segments within this end market to continue driving double-digit revenue growth on a sequential basis. For the local deployment segment of our data center end market, we anticipate a seasonal sequential decline in revenue, which will partially offset the growth driven by cloud and artificial intelligence. Therefore, we expect overall data center revenue to grow in the mid-single-digit percentage range sequentially. Now let me discuss the enterprise networking and carrier infrastructure end markets of Marvell Technology. In the fourth quarter, enterprise networking revenue was $171 million, and carrier infrastructure revenue totaled $106 million. In the fourth quarter, we saw continued recovery in demand for both end markets, with combined revenue growing 18% sequentially. Looking ahead to the first quarter of fiscal year 2026, we expect total revenue from enterprise networking and carrier infrastructure to grow approximately 10% sequentially. We are pleased with the continued recovery in these two end markets, although this forecast still anticipates that Marvell Technology's product shipments will be below the consumption levels of the end markets. In the consumer end market, fourth-quarter revenue was $89 million, a sequential decline of 8%. In the first quarter of fiscal year 2026, consistent with our previous comments, we expect seasonal factors and gaming demand to lead to a sequential decline in consumer end market revenue of approximately 35%. Over the next few years, we expect annual revenue in the consumer end market to be around $300 million. Turning to our automotive and industrial end markets. Fourth-quarter revenue was $86 million, a sequential increase of 3%, as we continue to see moderate recovery in this end market. Looking ahead to the first quarter of fiscal year 2026, we expect the automotive end market to continue to grow sequentially. However, we anticipate this will be offset by a decline in revenue from the industrial end market, as order patterns in the industrial end market can be unstable in any given quarter. Therefore, we expect overall revenue from the automotive and industrial end markets to decline sequentially by a high single-digit percentage. In summary, we achieved significant revenue growth throughout fiscal year 2025, with the company scaling up substantially, and annualized revenue growing from $4.6 billion in the first quarter to over $7.2 billion in the fourth quarter. Building on this expansion, we expect strong year-over-year revenue growth in fiscal year 2026. Our AI-driven data center end market is expected to remain a key contributor, while the continued recovery of our multi-market business will provide further support. In the fourth quarter, Marvell Technology's data center end market revenue accounted for 75% of consolidated revenue. Reflecting this rapid shift in fiscal year 2025, we have purposefully redirected investments towards data centers relative to other end markets to fully capitalize on the tremendous opportunities presented by artificial intelligence. We recently adjusted our organizational structure to fully realize this strategic transformation. All products aimed at hyperscale customers are now managed by a single cloud data center team led by our President of Products and Technology, Raghib Hussain We have consolidated the remaining end markets into a single multi-market business group led by Chief Operating Officer Chris Koopmans. Marvell Technology has solidified its position as a leading supplier of data infrastructure semiconductors, with a unique business model that covers solutions ranging from fully customized to fully commercial. We see that both existing hyperscale data center operators and numerous well-funded new market entrants are accelerating significant investments in infrastructure, competing to build 1 million XPU training clusters. These customers are highly motivated to increasingly utilize customized infrastructure to enhance their commercial solutions. Recent developments in the artificial intelligence market, such as the emergence of inference models, are also expected to continue driving strong demand for computing, networking, and storage semiconductors. Therefore, we remain very optimistic about the growth prospects in both the short and long term, as well as our role in driving accelerated infrastructure. We look forward to introducing our business model and the significant opportunities ahead of us to investors at the Investor Day event on June 10th in New York. At this point, I will hand the call over to William, who will provide more details on our recent performance and outlook. Chief Financial Officer William Meaney: Thank you, Matt, and good afternoon, everyone. Let me first summarize our performance for the entire fiscal year. In fiscal year 2025, Marvell Technology achieved revenue of $5.767 billion, with a strong 37% growth in the second half compared to the first half. This growth was primarily driven by AI projects in our data center end market and the continued recovery in other end markets. For the full year, our gross margin was 41.3% under U.S. GAAP. The operating margin was negative 12.5%, with a diluted loss per share of $1.02. Under non-GAAP, our gross margin was 61%. The operating margin was 28.9%, with a diluted earnings per share of $1.57. In fiscal year 2025, our non-GAAP operating margin improved by over 1000 basis points, rising from 23.3% in the first quarter to 33.7% in the fourth quarter. We also achieved a record $1.68 billion in operating cash flow and significantly increased capital returns to shareholders, returning $933 million through dividends and buybacks. Next, let's look at our financial performance for the fourth quarter of fiscal year 2025. Fourth-quarter revenue was $1.817 billion, exceeding the midpoint of our guidance, representing a year-over-year growth of 27% and a quarter-over-quarter growth of 20%. The data center is our largest end market, accounting for 75% of total revenue. The gross margin under U.S. GAAP was 50.5%. The gross margin under non-GAAP was 60.1%. Now, looking at operating expenses. Operating expenses under U.S. GAAP were $682 million, including stock-based compensation, amortization of acquired intangible assets, restructuring costs, and acquisition-related costs. Operating expenses under non-GAAP were $479 million, in line with our guidance. Our operating margin under U.S. GAAP was 12.9%, while the operating margin under non-GAAP was 33.7% In the fourth quarter, the diluted earnings per share under U.S. Generally Accepted Accounting Principles (GAAP) was $0.23. The diluted earnings per share under non-GAAP was $0.60, a quarter-over-quarter increase of 40%, which is twice the revenue growth rate, reflecting significant operating leverage in our business model. Now, let's look at our cash flow and balance sheet. The operating cash flow for the fourth quarter was $514 million. At the end of the fourth quarter, our inventory was $1.03 billion, an increase of $170 million from the previous quarter, to support our strong business growth. Our Days Sales Outstanding (DSO) was 51 days, a decrease of 9 days from the previous quarter. We returned $52 million to shareholders through cash dividends. Additionally, we repurchased $200 million of stock in the fourth quarter. Our total debt was $4.06 billion, with a total debt to EBITDA ratio of 2.06 times, and a net debt to EBITDA ratio of 1.58 times. As we drive EBITDA growth in fiscal year 2025, our debt ratios continue to improve. We are pleased to have received an upgrade to our investment-grade credit rating from Fitch in January, which noted their positive outlook on Marvell Technology's strong operational momentum, robust data center demand, structurally improved leverage metrics, strong market position, and enhanced cash flow situation. As of the end of the fourth fiscal quarter, our cash and cash equivalents were $948 million, an increase of $80 million from the previous quarter. Turning to our guidance for the first quarter of fiscal year 2026, we expect revenue to be around $1.875 billion, plus or minus 5%. We expect a gross margin of approximately 50.5% under U.S. GAAP. We anticipate a gross margin of about 60% under non-GAAP. Looking ahead, we expect that the overall level of revenue and product mix will remain key determinants of our gross margin in any given quarter. For the first quarter, we expect operating expenses under U.S. GAAP to be approximately $712 million. We expect operating expenses under non-GAAP to be about $490 million, a slight increase of around 2% from the previous quarter. It is worth noting that the increase in operating expenses this time is lower than our usual expectations due to typical seasonal factors in the first quarter, payroll taxes, and employee salary performance growth. This is because we expect to gain more leverage from the custom model and anticipate a quarter-over-quarter increase in one-time engineering expenses (NRE). Just a reminder, NRE is considered a reduction item in operating expenses. For the first quarter, we expect other income and expenses, including interest on our debt, to be about $43 million. We expect a non-GAAP tax rate of 10% for the first quarter. We anticipate a basic weighted average shares outstanding of 867 million shares and a diluted weighted average shares outstanding of 818 million shares. We expect diluted earnings per share under U.S. GAAP to be between $0.14 and $0.24. We expect diluted earnings per share under non-GAAP to be between $0.56 and $0.66 I am very satisfied with the performance of Marvell Technology in all key financial metrics for the fiscal year 2025. As we enter fiscal year 2026, we intend to continue driving strong operational leverage and expect to make significant progress toward achieving our long-term non-GAAP operating margin target of 38% to 40%. We will also continue to focus on generating strong cash flow and returning capital to shareholders. I am very excited about our future prospects and look forward to achieving strong year-over-year revenue growth in fiscal year 2026. At this point, we are ready to begin the Q&A session. Moderator, please open the question lines and announce the Q&A instructions. Thank you. Q&A Session Moderator: We will now begin the Q&A session. Your first question comes from Ross Seymore of Deutsche Bank. Your line is open. Ross Seymore: Hi, everyone. Thank you for letting me ask a question. Matt, I want to go back to your comments about confidence in growing alongside major XPU customers, not just this year but next year as well. I believe you also understand that there is a lot of controversy around issues like competitors taking market share. How do you think we should reconcile your confidence with theirs? It seems like both sides can't be completely right. Chairman and CEO Matt Murphy: Yes. Hey, Ross. Thank you for your question. I fully understand the controversies that exist in the market, and I appreciate the strong interest from the analyst community and our investors in this topic. So, under the constraints of customer confidentiality and their concerns about the confidentiality of their own projects, let me provide you with as much useful information as I can. Okay. We are very pleased with the progress of our current major XPU projects. We successfully delivered products on our first tape-out. We see that the current generation of products is about to achieve mass production, and you can already see the growth trend from our fourth-quarter data. Meanwhile, we have been working closely with this customer on the next-generation AI XPU. I mentioned this in my prepared remarks, but I will reiterate. We plan to ramp up production after completing the sample and validation cycles. So, to be clear, we do expect revenue from customized XPUs from this customer to grow not only in fiscal year 2026 (this year) but also in fiscal year 2027 and beyond. I would also like to point out that I am not including revenue from other products, such as networking or connectivity products, which are additional revenues for the customized business. And I believe that, overall, we have a good layout in this regard relative to the opportunities presented by this customer, in fact, in both aspects. Then, regarding the question of whether our customers might collaborate with other companies outside of Marvell Technology to develop the next-generation XPUs, we cannot comment on that. What I can comment on is the visibility of the products we are building for our customers. As I mentioned, this extends to the next-generation products of our current AI XPU projects Ross Seymour: Thank you. Chairman and CEO Matt Murphy: Okay. Host: Your next question comes from Vivek Arya of Bank of America. Your line is open. Vivek Arya: Thank you for taking my question. Matt, I’d like you to clarify if you could help quantify the proportion of AI versus non-AI business in the data center for the fourth and first quarters, is it roughly two-thirds and one-third? So, any quantifiable information in this regard would be very helpful. Then a more general question, when we look at the data center and AI business for the fourth and first quarters, the year-over-year growth is very strong, but it seems somewhat moderate compared to the strong spending from your largest customers. So, any comments on this would be helpful. Is this a supply issue? Or is it still in the early stages? So, any comments on this would be helpful. Thank you. Chairman and CEO Matt Murphy: Okay. Thank you, Vivek. Maybe I’ll start with the second question regarding the overall situation. If you look at it, we’ve seen very strong growth over the past few quarters. In fact, I believe our growth has indeed outpaced the overall market. I think from the second quarter to the third quarter, data center revenue grew 25% quarter-over-quarter, and from the third quarter to the fourth quarter, it was also 25%. As I mentioned in my prepared remarks, the revenue from AI and cloud business in our data center grew double digits quarter-over-quarter from the fourth quarter to the first quarter. By the way, these are all quarter-over-quarter figures. Of course, year-over-year, if you look at the fourth quarter, our data center business grew by about 77%. I think the growth in the first quarter is also similar. So, in terms of quarter-over-quarter and year-over-year growth rates, we are very satisfied with the trajectory of this business. So I think progress is going very well. In terms of AI, this is how we think about it; I can give you a rough reference point as of the end of last year, but it’s also a good overall indicator. Currently, the revenue share from AI business has exceeded half. So now it has become the majority, surpassing this proportion. And I believe this proportion will continue to increase over time. So, since over 75% of Marvell Technology's overall revenue comes from the data center business, from this perspective, this will serve as a very good reference point. To elaborate further, if you look at the entire data center business, about half of the revenue comes from the optoelectronics business, while the custom business has now grown to account for about 25% of data center revenue, a little over a quarter. The remaining revenue comes from other areas. So, this is again a very good growth trajectory. The growth of the AI business has outpaced the overall business growth, and it is the main driving force behind business development. The data center business now occupies a significant proportion of Marvell Technology's overall business. And with the real growth of the custom business, the business mix is also continuing to develop well. So, I hope these data points are helpful Overall, we are very satisfied with the trajectory of our business development. Vivek Arya: Thank you, Matt. Chairman and CEO Matt Murphy: Okay. Senior Vice President of Investor Relations Ashish Saran: Can you please proceed to the next question? Chairman and CEO Matt Murphy: Host? Senior Vice President of Investor Relations Ashish Saran: I'm confirming. Everyone, please hold on while we get the host back online. Host: Your next question comes from Timothy Arcuri of UBS Group. Your line is open. Timothy Arcuri: Thank you very much. Matt, I want to clarify your earlier response regarding the composition of AI revenue. You mentioned that over half of the data center revenue is from AI business, and then you said that half of the data center business is from optoelectronics. Are you saying that half of the AI business revenue is from optoelectronics? So, the composition of AI business revenue is basically about half optoelectronics and half custom application-specific integrated circuits (ASICs), is that correct? Chairman and CEO Matt Murphy: Yes, Tim, thank you for the clarification. To be precise, the numbers I gave you were about the entire data center business. Timothy Arcuri: Okay. Then you previously projected AI business revenue to reach $2.5 billion this year. I think you mentioned that in fiscal year 2026 you would "significantly" exceed that target. What does that mean? I think most sell-side analysts believe it’s around $3.5 billion, so that would certainly count as "significantly" exceeding it. When you say "significantly," does that mean exceeding by like $1 billion? Is that what you mean? Thank you. Chairman and CEO Matt Murphy: Yes, Tim. Hey, thanks for your question. Yes, we really have not set limits on what we can achieve this year. Last year, we talked about a $1.5 billion target. We easily exceeded that target. This year, we also expect to significantly exceed that target. I don’t want to give specific numbers right now. I think in terms of the momentum of the business and the opportunities in front of us, there is a lot of potential. So currently, we are mainly relying on the last update from last year's AI Day, and then we will find the right time in the future to announce specific numbers. But right now, our growth trend is very good. You can really see that from the data. As I mentioned earlier, at some point, data center business revenue now accounts for 75% of Marvell Technology's overall revenue, and I have also told you the proportion of AI business, so you can start using this as a very good reference point to understand the direction of our business. Timothy Arcuri: Okay. Matt, thank you. Chairman and CEO Matt Murphy: Okay. Host: Your next question comes from Harish Kumar of Piper Sandler. Your line is open Harish Kumar: Yes. Hi. Thank you for letting me ask a question. Matt, I think you mentioned that you have three custom ASIC customers. I want to talk about — I want to understand some information about the stickiness of these customers. What factors might lead these customers to look for other design partners? Is it price factors, or if the product cannot be produced, will it be 100% based on performance considerations to seek other partners? Then the second part of the question is, how likely are you to connect your network products with these custom ASIC customers? Or if you have already done so, can you help us understand the current cooperation situation? Chairman and CEO Matt Murphy: Okay. Thank you for your question. First of all, we actually have custom business collaborations with all four major hyperscale data center operators. To be precise, two of them have collaborations in the computing field, and a third one is about to join, just to clarify. Moreover, these chips are very complex, and our view is consistent with what we have always said, which is that when you look at the total addressable market (TAM), the data center TAM we estimated for Marvell Technology last April was about $75 billion, and this number is likely to only grow, with a significant portion coming from custom business. The reality is that the partners that the customer base is truly looking for have several key characteristics. First is technological leadership, technology investment, and intellectual property. For example, just a few days ago, we announced our available 2-nanometer platform. We have very competitive products in custom HBM, chip interconnects, advanced packaging, and so on, which is part of it. The second is manufacturing scale, the ability to adopt the right design methodology to achieve first silicon success, and the capability to bring these highly complex chips with over 100 billion transistors into production. They are also looking for partners with good relationships with suppliers; for Marvell Technology, we have invested heavily in our supply chain. Finally, partners with flexibility in their business models, able to adopt different approaches for different chips, from traditional ASIC services to fully customized solutions, we can provide all of that. So ultimately, it’s these different factors. When you look at all these factors together, we still believe that only we and one other very strong competitor can serve this market. So these customer relationships are very solid. At the same time, every generation of products requires bidding and competition, but since we started winning these large custom chip projects in 2021, securing orders, and now bringing them into production while looking at the next generation of products, we have been very successful. We are very confident in our position. So customer stickiness is high, and the barriers to entry are also high. But at the same time, we must win the favor of customers every time. Finally, I want to say that from the perspective of design opportunities, this business channel is continuously expanding, and in the process, we continue to achieve very important design collaborations, as was the case in the recent fourth quarter. So we still see these dynamics. Competition is fierce, and everyone is in a race But to be frank, in terms of technology, platform, and our ability to execute around what I believe is the biggest opportunity in the semiconductor industry for a long time — namely, the artificial intelligence supercycle that is still ahead of us — we feel we are in the best position ever. Harish Kumar: Thank you, Matt. My follow-up question is, I'm curious if you could elaborate on your statement that "the growth of the data center cloud and AI business will reach double digits, while on-premises deployment business will decline." What do you mean by double-digit growth? Is it in the teens or higher? It's just that investors are very sensitive to your growth rates in cloud and AI business. Chairman and CEO Matt Murphy: Okay. I mean, we can be a bit more precise. Of course, this is our expected guidance. Double-digit growth means double digits, not 20% or any other number. I mean, it's above 10%. You should understand it that way. Obviously, we will look at the actuals for this quarter. But what we see at the moment is just that. You also asked me about the collaboration on networking products. So I want to quickly address that question, and then we can move on to the next one. Yes, it has proven that customer stickiness in this area is very high. And this may be another aspect that complements custom design, which I mentioned in my prepared remarks. The value of solutions now lies not just in who can make the best individual compute chip. I mean, while that is critical, you must have the right performance, leverage Moore's Law, etc., but ultimately to achieve hyperscale performance in solutions, you must combine it with networking and connectivity. I think from scaling to scalable output, to the various different input-output (IO) interconnect solutions we have, this has proven to be a very compelling reason for customers to choose us because we can design these synergies. In fact, we can do this in the opportunity for custom compute chips as well. By the way, when we do this, especially when we engage in projects like accelerators or CPUs, we can gain insights and perspectives on customer architectures very early on. Similarly, it's not just about the compute chip part, but all the necessary IO, and then we develop our solutions around those. So, in terms of how we approach our technology platform, everything is meticulously designed, but compared to some of our competitors who may only have one aspect of this, it presents a more attractive value proposition for our customers. Harish Kumar: Very helpful, Matt. Thank you. Chairman and CEO Matt Murphy: Okay. Thank you. Host: Your next question comes from Aaron Rakers of Wells Fargo. Your line is open. Aaron Rakers: Yes. Thank you for taking my question. I also want to continue asking about AI. Matt, you mentioned that you have collaborations with four hyperscale data center operators, and you also mentioned that two of them have collaborations in computing, with a third one coming on board soon I want to clarify, is the fourth customer also an opportunity in the computing field? Or does it refer to the custom NIC project you are currently working on? Then, in the same context, how do you view the custom logic opportunities surrounding HBM4, and what is the timeline for its impact on Maywell Technology? Thank you. Chairman and CEO Matt Murphy: Okay. Thank you, Aaron, those are two great questions. So, for the first question, I think you understood it correctly. In fact, you are right. And this opportunity is not limited to that product, just as an example, but yes, for the fourth customer, we are already producing related products, and we have publicly announced the custom NIC project in that field. Then regarding custom HBM, yes, this is still our key intellectual property, and I think it is a key part of the customer roadmap we are seeing. So we are currently very actively working with our main memory partners, collaborating with these partners alongside our hyperscale data center customers, and the benefits we can propose are very significant, such as reducing chip area on logic chips, enabling our customers to integrate more computing power on that chip, and also improving the throughput between HBM and computing, and actually providing higher density memory solutions for our customers' computing chips. So all of this is still very important, and I think it will be a significant industry trend in the next generation of high-end accelerator products. Thank you. Host: Your next question comes from Tom O'Malley of Barclays Bank. Your line is open. Tom O'Malley: Hey, everyone. Thanks for taking my question. I really appreciate it. So I want to shift slightly to a technical question. Matt, in your remarks, you mentioned that optical connections are increasingly replacing areas currently dominated by electrical connections. So I’d like to hear your thoughts on that. When you look at the optical field overall, especially Co-Packaged Optics (CPO), do you think it will first be realized inside the rack, or will it be in the aggregation layer? And what does this mean for your DSP business? Clearly, your share in the optical field is different from your share in the switching field. I would love to hear your thoughts on where CPO will first appear, and in terms of your product portfolio, where you think you have advantages? Chairman and CEO Matt Murphy: Okay. Thank you, Tom. So, perhaps I can provide some perspective. First, I think from the historical context of CPO, the ideal goal for over a decade, even close to twenty years, before the huge transformation brought by accelerated computing growth in recent years, was that the aggregation layer or extended output would eventually adopt CPO technology. There have been many proof of concept (POC) projects in this area. There are also many different efforts within the industry. I mean, I attended the OFC conference last year, and at that time, many expectations fell short. I think it’s not just us; our large cloud customers have also indicated that this is a very long-term opportunity. By the way, this is also our view, and we have been keeping an eye on this area and have made some investments in this regard However, in the past — by the way, linear optical (LPO) has also experienced a similar situation. In 2023, there was a lot of discussion about whether LPO could replace DSP-based pluggable modules. But that did not happen. However, as I mentioned in my prepared remarks, there are many opportunities to transform connections from passive to active and accelerated, and to adopt optical connections in terms of expansion, connections within racks, between accelerators, and within clusters. This can be achieved in several different ways. Co-packaged optics is a very interesting and attractive implementation method. We showcased our 6.4T product at the OFC exhibition last year. We are definitely investing in this area. And relative to the solutions we proposed as part of the ASIC platform, we have made significant progress in this regard. The same goes for LPO. I believe it may find its place at some connection points within racks. But these will be important applications, especially in closed systems. There are still many technologies to be developed, and a lot of work to be done. But we place great importance on investment in this area, and we will compete in this field with first-class technology. The question is, when will it be adopted? What is the certification process like? What are all the manufacturing and yield issues we will have to deal with? So it's complex, Tom, but this is not an area we are going to give up on. In fact, we believe this is a key part of our long-term technology investment, not only to have first-class DSPs and pluggable modules, first-class Automatic Equalizers (AEC), such as retimers, and a full suite of connectivity solutions, but also to invest in next-generation technologies that will enable us to achieve significant breakthroughs in power and performance. We just need to strike a balance between this and reality, as it is very complex, novel, and not traditional complementary metal-oxide-semiconductor (CMOS) technology. These technologies are more complex. So you will see more progress from us in this area at the OFC exhibition, and of course, this is a field worth paying attention to. Tom O'Malley: Very helpful. So in the short term, it sounds like things are looking better in the optical field, but like many things related to artificial intelligence today, there are also some noises, such as the inventory of 800 gigabit products possibly facing some resistance, or the price dynamics during the transition to 1.6 terabits. Can you talk about the situation at these two nodes, and your views on the growth trajectory and pricing conditions in the 800 gigabit space this year? Because there has been a lot of discussion on this topic. Thank you, Matt. Chairman and CEO Matt Murphy: Sure. We feel very good. The optical business has always been very strong for us. In the second half of last year, our order volume increased significantly. We have been fulfilling these orders. We see very strong demand this year, especially for 800 gigabit products, where demand is very healthy. We do see a trend towards the transition to 1.6 terabits, but frankly, 800 gigabit products will still be the main products We are preparing to support our 5-nanometer 1.6 terabit devices, but we are very excited about our 3-nanometer products. Due to their energy-saving effects, they have received widespread attention. I mean, each link can save 20% of energy consumption. So, when you consider the enormous scale of these large-scale clusters being built and the number of interconnection points, these energy savings add up to a significant amount. So, Tom, everything looks very optimistic. Of course, there will always be various noises in the market. This is a very hot field. I mean in the entire artificial intelligence sector, but from the perspective of Maiwei Technology, based on what we see, we observe strong and healthy demand continuing. Host: Your next question comes from Harlan Sur of JP Morgan. Your line is open. Harlan Sur: Hey, good afternoon, everyone. Thank you for taking my question. It's great to see that after your current XPU projects with major cloud services and hyperscale data center customers, you have also secured follow-up AI XPU projects. Matt, I want to confirm, is the new follow-up XPU project also for training XPU? Will it start ramping up production in the 2026 calendar year? Is it using 5-nanometer technology or 3-nanometer technology? Any additional information would be appreciated. Then, William, your inventory increased by 20% quarter-over-quarter, which usually indicates strong growth in the future in an environment of strong demand and good product cycles. However, compared to the expected guidance of a 3% quarter-over-quarter increase in total revenue in April, there seems to be some disconnect. So, can it be understood that the 20% quarter-over-quarter increase in inventory better reflects the team's strong growth momentum in the AI business in the future? Chairman and CEO Matt Murphy: Okay, I'll answer first, Harlan. There are a few points to clarify. Since we need to adhere to confidentiality agreements, I can only say this. First, you should assume this is a large-scale project and a continuation of our current work. Generally speaking, for each generation of next-generation products, you can expect technological advancements and changes, so you can understand it this way. Additionally, from a timing perspective, I can only say that we will be ready to start ramping up production when the time is right, and we will manage this transition process properly. We are very confident in successfully completing this transition with our customers, but we can only determine the specific timing when the time is right, and we will arrange it based on actual conditions. Moreover, I really cannot comment on the customers' plans and these details; they do not like that, and I completely understand. I'll let William answer the question about inventory. Thank you. Harlan Sur: Okay. Thank you. CFO William Meaney: Hey, Harlan. Regarding inventory, this is actually to support the continued growth of our custom projects this year, as well as the strong growth of our optical business. So, in fact, these two factors drove the changes in inventory. But if we look at inventory days, it is actually flat quarter-over-quarter Harlan Suer: Understood. Thank you for your answers, everyone. Chairman and CEO Matt Murphy: Alright. Host: Your next question comes from Mark Lipacis of Evercore ISI. Your line is open. Mark Lipacis: Great. Thank you for answering my question. Matt, I think there has been ongoing concern in the market about a slowdown in spending or even the risk of a digestion period. Yet you seem very confident about the outlook for your business this year. I wonder if you could tell us what indicators you use to assess whether there is a risk of a digestion period? Additionally, as part of this, I have always viewed Marvell Technology as a communications integrated circuit (com IC) company. Now your processor business is growing very rapidly. I would like to know, if you were to dissect this risk assessment, how do you view these two businesses? Chairman and CEO Matt Murphy: Okay, Mark. I will answer in two parts. As a company, we have done a lot of comprehensive analysis when forecasting our data center business. We do consider a lot of top-down data, such as capital expenditure (CapEx) trends, among other aspects. But ultimately, due to the nature of these projects and the critical nature of our products, we are closely connected with our customers, especially when our products are exclusively supplied and become a potential key link for them. We are very cautious in planning the supply chain together with our customers. I think the growth of our business over the past two years with the emergence of ChatGPT speaks to this. It’s really remarkable to think about where we used to be. If we look back at this call two years ago, the significant breakthrough was our expectation of achieving $200 million in AI business revenue in the 2023 calendar year, reaching $400 million in 2024. And now we just wrapped up 2024, and we are talking about breaking through $1.5 billion and achieving over $2.5 billion in revenue. So, by the way, that’s really impressive. But we have indeed grown alongside our customers and have generally been able to exceed expectations, meet their needs, and achieve business growth. So, as we look forward to the remainder of this year, and frankly, from a multi-year perspective, the outlook is very optimistic. Some of this is due to our specific factors, such as new projects in our custom business that are bringing incremental revenue. We also have additional business opportunities. We have customers undergoing transformations from a technology perspective, such as in interconnect technology. So we have many positive driving factors. Now we are closely monitoring the supply chain, which does require constant review, but currently, we again see that in fiscal years 2026 and 2027, the advancement of custom projects, the continued growth and strengthening of the connectivity business, and the recovery of higher-level networking and storage businesses are all laying a very solid foundation for strong growth. So at the moment, everything appears to be positive Mark Lipacis: Great. Thank you. Very helpful. Chairman and CEO Matt Murphy: Okay. Host: Your next question comes from Ben Reitzes of Melius Research. Your line is open. Ben Reitzes: Hey everyone. Thanks for taking my question and for the break in between. It was a nice break opportunity. My question is about sequential growth, Matt, and a long-term question. When Tim asked earlier, did you explicitly state that AI business revenue would continue to grow sequentially this year? If so, where does your confidence come from? Also, Matt, I’d like you to take a macro view, as there’s a lot of speculation in the market regarding that customer and its related content. But looking back, you set a goal for the data center business to reach $15 billion in long-term revenue by the 2028 calendar year. Do you think your custom business has made progress toward achieving that goal? Because currently, no one in the market is close to that target. Thank you. Chairman and CEO Matt Murphy: Okay, Ben. Thank you for your question. Regarding expectations for this year, I think I mentioned earlier that in the first quarter, we expect the part of our data center business, excluding on-premise business, to achieve double-digit growth on a sequential basis across the entire data center business, such as a 25% sequential growth each quarter. So I haven’t commented on the full year, but you can reasonably infer. That’s not an unreasonable assumption, right? Or rather, given the strong momentum of the business and the development trends we’re seeing, it’s natural to think that this growth will continue from a year-over-year perspective. In the long term, I believe we are steadily moving toward our goal of a 20% market share. When we look back at the 2023 calendar year, 2024, and 2025, and compare these with the targets we set for AI Day, I think the outlook is very optimistic. We certainly gained more market share from 2023 to 2024, and we will definitely gain more share from 2024 to 2025 as well. So, to achieve that revenue target, we need to work on two fronts. On one hand, you need to increase market share from about 10% to 20%. On the other hand, the market size clearly needs to grow as well. You need to reach a total addressable market (TAM) of about $75 billion. But both aspects are moving in a very positive direction. In fact, looking at some rough numbers for 2024 and even 2025, if we compare the market and our growth situation with the compound growth rate needed from 2023 to 2028, our current growth rate is actually exceeding that target. So we are in a growth phase now, and that growth will continue So, from the perspective of market share, I believe our progress is very smooth. By the way, if the market size is larger—remember our viewpoint from last April. Since then, the world has changed, especially in terms of actual capital expenditures. Particularly in recent months, there have been various projects from key players, as well as the potential of new entrants, sovereign projects, government projects, and so on. So frankly, we see that the total market size and opportunities for Marvell Technology are much larger than what we assessed about a year ago. So all of this makes me very optimistic about the development of market size and our progress in market share. You are right; if we can achieve this goal, it will be a huge success, which is also the goal that my team and I strive for every day in the company—to drive market share growth, help expand the market, achieve growth in total market size, and execute these goals with full commitment and focus. This is also the reason why we reorganized the company and established dedicated data center, engineering, and business teams to drive business development. So I think the current situation is very good. Thank you. Ben Reitzes: Thank you, Matt. Host: Your last question comes from Christopher Rolland of Susquehanna Financial Group. Your line is open. Christopher Rolland: Hey. Thanks for letting me ask a question. Some of this has been discussed before, but in terms of expansion, you have a lot of opportunities. Some of these may involve optical connections, some may be co-packaged optics, and some may be active copper connections. I think you might even have some alternatives similar to Unified Architecture Link (UAL) or NVIDIA Link (NVLink), as well as digital signal processors throughout the system. I wonder, as we see changes in GPU or XPU architectures, where do you see your biggest opportunities? When might we see results from you in these areas? Thank you, Matt. Chairman and CEO Matt Murphy: Good question, Chris. This is a great question to end this Q&A session. First, you did your homework well. The key message I want to emphasize is that we plan to participate in and drive the development of all these areas you mentioned. We have ongoing projects and development work in all these different fields. I want to emphasize to you and the investors on the call that this represents additional Total Addressable Market (TAM) growth opportunities. The opportunities for expansion, as we have pointed out multiple times, are the right areas for achieving total market size expansion. And it looks like this will become a reality at some point in the future. So these are all additional growth opportunities. This does not mean that if these opportunities materialize, Marvell Technology's digital signal processor business revenue will suddenly decline or anything like that. Expansion and this connection technology represent opportunities for revenue growth and market growth So, we are very excited about this opportunity. I think you will definitely hear more from us about this at the Optical Fiber Communication Exhibition (OFC) and our Investor Day events. There is no doubt that this is a huge opportunity. Christopher Roland: Thank you, Matt. Chairman and CEO Matt Murphy: Okay. Thank you, Chris. Host: Now I will hand the conference call over to Mr. Murphy for closing remarks. Chairman and CEO Matt Murphy: Okay. Thank you, host. First, I want to thank everyone for joining this conference call. I apologize for the interruptions that occurred in the middle. I hope everyone had a chance to relax with a cup of coffee during the brief break, and we will certainly find out what happened, but I still appreciate everyone’s attention to the company. I want to thank everyone for their attention and recap a few things. I think first, we made very significant progress in fiscal year 2025. Our annualized revenue for the first half of the year is approximately between $4 billion and $4.3 billion. And now we are providing revenue guidance of about $18.75 billion, which is far above the annualized revenue level of around $7.5 billion. So I think this is a very strong development trajectory for the company, a huge change. We have also gained greater leverage in our business model, including operational leverage, cash flow generation, and improvement in operating margins. You can even see this from the change in our operating expenses from the fourth quarter to the first quarter. As William mentioned, typically, the growth in operating expenses exceeds our expectations, but we are starting to see the benefits brought by these custom silicon chip projects and the contribution from non-recurring engineering (NRE) costs. So, we are seeing results not only in revenue growth but also in cost control. This is a very good situation. We achieved profitability in the fourth quarter according to Generally Accepted Accounting Principles (GAAP), which is fantastic. We will continue to drive this positive momentum. And it’s not just the artificial intelligence business; our standard cloud infrastructure business in data centers also had a very strong year. A year ago, we might have had concerns about some projects, such as whether these products could be mass-produced smoothly and certified. What would happen? But I think our team, and I want to thank the team at Maywell Technology, has excellently scaled these very complex devices into mass production and supported our customers. Tom previously asked about the optoelectronics business. Our optoelectronics business performed very well last year and is expected to be very strong next year as well. Additionally, the multi-market business that has been concerning over the past year, such as enterprise and operator businesses, is now also seeing very strong quarter-over-quarter growth. As these businesses recover, their quarter-over-quarter growth trend is good. At least for the enterprise and operator businesses, they are moving back towards an annualized revenue level of $2 billion. Moreover, we have raised our performance expectations. We expect revenue and profits to continue to grow in the first quarter. Customer engagement is very high, as I mentioned, last year we won many important design projects, and our sales team and business units executed very well in winning these key design projects, which will drive our future growth We are very optimistic about the future development direction. So, thank you all for your attention. I look forward to meeting everyone at various upcoming investor meetings and continuing our communication. Thank you all, I really appreciate it. Host: Ladies and gentlemen, today's conference call is now concluded. Thank you all for your participation. You may now disconnect the call. Risk Warning and Disclaimer The market has risks, and investment requires caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Investment based on this is at your own risk