Maywell Technology (MRVL.US) Q2 conference call: The custom chip business has accelerated markedly since FY2029, and the growth rate of large-scale optics has far exceeded expectations

The Zhitong Finance App learned that Mewell Technology (MRVL.US) announced financial results for the second quarter of fiscal year 2027 and held a performance conference call.

Overall, Q2 results exceeded expectations. Revenue of $2,739 million (up 37% year over year and 13% month on month) and non-GAAP earnings per share of $0.94, all exceeding the guideline median value. The company raised its revenue forecast for the full year of fiscal year 2027 from about US$11.5 billion to about US$12 billion (up about 45% year on year), and raised its revenue forecast for fiscal year 2028 sharply from US$16.5 billion to about US$18 billion (up about 50% year over year). The data center business is an absolute growth engine. Q2 contributed $2.17 billion (+46%) to revenue, accounting for 79% of total revenue. Q3 is expected to increase by more than 20% month-on-month, increase by about 75% year-on-year, and increase by about 60% for the full year of fiscal year 2027. Communications and other businesses generated revenue of US$568 million (+10%) in Q2, but Q3 is expected to drop 10%-15% month-on-month. New CFO Dan Durn emphasized that the company is at a “unique turning point,” with the goal of reaching the long-term target range of 38%-40% non-GAAP operating margin in Q4 and reaching the upper limit of the range in FY2028.

The core focus is on the company's partnership with Google for custom chips. According to J.P. Morgan Harlan Sur, the cumulative revenue of the six-year commercial agreement signed in late July will reach 120 billion US dollars, and the annual revenue of Google XPU additional custom ASIC alone will reach about 18.5 billion US dollars. He asked if the revenue target for the custom business in FY2029 would rise sharply from the $10-11 billion previously set. CEO Matt Murphy responded that the agreement covers a wide range of product lines such as inference accelerators, memory controllers, network cards, and memory interface controllers. Most of the revenue impact will be felt in FY2029 and beyond. The custom business “will more than double year-on-year” in FY2028, and “will accelerate significantly” in FY2029. Murphy made it clear that “there is a lot of room for growth in customized business revenue data for FY2029 and beyond,” and promised to provide detailed analysis on October 6th Investor Day. Vivek Arya (Bank of America) inquired about the progress of another large-scale XPU project. Murphy said, “We are making continuous progress every quarter, and the situation is only getting better.”

Scale Optics (Scale Optics) became the second most discussed topic. Wells Fargo Aaron Rakers inquired about developments in CXL and expandable optics technology. Murphy emphasized that CXL technology is “ideal for memory expansion and inference,” and has won new design orders over the past few quarters. In terms of large-scale optics, he gave a key update: the growth rate of the entire large-scale optics sector, including CPO and NPO, “far exceeded expectations three months ago,” and the $15 billion financing plan for the 2028 fiscal year “is largely due to this sector.” NPO opportunities are developing “much faster than CPO”, and Marvell is deeply involved through self-developed light engine solutions and silicon germanium broadband analog products (TIA and drivers). Morgan Stanley Joe Moore asked about the large-scale transition path of copper cabling to fiber. Murphy responded that Marvell's Teralynx switching architecture has achieved great success in the Ethernet field and can be “seamlessly transformed into vertical expansion.” He emphasized that customers needed “complete solutions from end to end” and that “fragmented solutions alone cannot complete the task.” Cantor Fitzgerald CJ Muse cuts from Nvidia's views on computing, networking, and memory interchangeability. Murphy stated that Marvell “must have the ability to quickly perform custom and semi-custom designs,” as examples of which are CXL and memory extensions. Goldman Sachs Jim Schneider inquired about the growth rate of connectivity. Murphy confirmed that connectivity is the “biggest driving force” of the $15 billion financing plan, covering a full line of products such as optical DSPs, horizontal expansion transceivers, and switches.

Conference content

presenter

Good afternoon, and welcome to Marvell Technology Inc. Q2 FY2027 earnings conference call. [OPERATION INSTRUCTIONS] Please note that this meeting is being recorded.

I will now entrust the meeting to Mr. Ross Seymour, Senior Vice President of Investor Relations. Thank you. You're ready to go.

Ross Seymour, Senior Vice President of Investor Relations

Thank you all, and good afternoon. Welcome to Marvell's Q2 FY2027 earnings conference call. I was in attendance today with Marvell Chairman and CEO Matt Murphy, Chief Financial Officer Dan Durn, President and Chief Operating Officer Chris Koopmans, and Data Center Group President Sandeep Bharathi. I would like to remind everyone that today's statement contains forward-looking statements that are subject to significant risks and uncertainties that may cause our actual performance to differ materially from management's current expectations.

Please review the cautionary statements and risk factors in the earnings press release we filed with the US Securities and Exchange Commission (SEC) today and posted on our website, as well as the latest 8-K, 10-K, 10-Q, and other documents we have filed with the SEC from time to time. We are not planning to update this forward-looking statement. In today's conference call, we'll be referring to some non-GAAP (non-GAAP) financial measures. A reconciliation of generally accepted accounting standards (GAAP) and non-GAAP financial measures is provided in our earnings press release.

Now I'm handing over the phone to Matt and ask him to comment on how things have been going this quarter. Matt?

Matthew Murphy, CEO and Chairman

Thanks Ross and good afternoon everyone. Before discussing our results and outlook, I'd like to briefly review the two management changes that occurred last quarter. First, Willem Meintjes (Willem Meintjes) stepped down as Marvell's chief financial officer in mid-June. I am very grateful to William for his steady leadership, outstanding leadership, and tireless efforts over the past 10 years for Marvell. He is committed to transforming the company, and I also really respect his desire to spend more time with his family. To ensure a smooth transition, we are also welcoming Dan Durn as our new Chief Financial Officer. Dan has over 30 years of senior financial management experience in semiconductor and enterprise technology companies. He recently served on Marvell's board of directors, so he has a deep understanding of our business and strategy, and a unique insight into Marvell's huge future growth opportunities.

Second, in July, we launched an investor relations leadership transition. After 8 years at Marvell, Ashish Saran will retire from the company in April 2027. On behalf of Marvell, I would like to sincerely thank Ashish for his leadership, cooperation, and his numerous contributions to Marvell. I am also very grateful to him for his thorough consideration and careful arrangements for this transition, including helping us find the right successor. I'm very happy to welcome Ross Seymore to us here. Ross Seymore previously worked at Deutsche Bank and has over 25 years of experience in the semiconductor industry. A warm welcome to Dan and Ross!

Now let's take a look at our performance and outlook. In the second quarter of fiscal year 2027, Marvell achieved a record revenue of US$2,739 billion, up 13% month-on-month and 37% year-over-year. Revenue and non-GAAP earnings of $0.94 per share both exceeded previous median expectations. In our last earnings call, we raised our month-on-month revenue growth forecast for the third and fourth quarters of this fiscal year from the previous high single-digit growth rate to a double-digit percentage. Since then, our expectations have continued to grow, and we now expect revenue growth to accelerate further in the second half of the year. This strong momentum is reflected in our performance guidance for the third quarter of fiscal year 2027. We expect the company's median total revenue of US$3.15 billion, an increase of 15% month-on-month and an increase of more than 50% year-on-year.

We expect the fourth quarter's month-on-month and year-over-year growth to accelerate further. As a result, we now expect Marvell's total revenue for the 2027 fiscal year to grow approximately 45% year over year to approximately $12 billion, up from our forecast of approximately $11.5 billion for the previous quarter. The increase in revenue expectations is mainly due to growth in the data center business. We expect this business to grow by about 60% this fiscal year, higher than the previous forecast of about 50%. Notably, this growth is still all-round. The interconnect business continues to lead the way, while the custom business is expected to grow significantly in the second half of the year. The growth trajectory of communications and other terminal market businesses has generally been in line with expectations. Although these businesses generally fluctuate from quarter to quarter, we currently expect the 2027 fiscal year growth rate to approach the 10% target.

Looking ahead to FY2028, overall demand continues to grow at an accelerated pace, and our operations team did an excellent job of securing additional supply despite the industry's prevailing supply constraints. As a result, we now expect Marvell's data center business revenue to grow by more than 60% year over year in the 2028 fiscal year, mainly due to strong growth in all of our key data center businesses. As we noted last quarter, revenue from the custom business more than doubled. We look forward to exploring in greater depth the specific factors driving our long-term growth at Investor Day in New York on October 6, but today's key message is clear: the strong performance of our data center business continues to exceed our previous expectations.

Combining all factors, we now expect revenue for the 2028 fiscal year to be around $18 billion, an increase of $1.5 billion over our forecast of $16.5 billion for the previous quarter. More importantly, while our revenue base has expanded significantly, the pace of growth is accelerating. We now expect revenue growth of around 50% year over year for the 2028 fiscal year, compared to the previous forecast of around 45%.

Next, I'll go into more detail about our current business, starting with our data center business. In the data center terminal market, our second-quarter revenue hit a record high of $2.17 billion, up 18% month-on-month and 46% year-on-year. Compared with the first fiscal quarter, both the month-on-month and year-over-year growth rates accelerated. Data center revenue in the first fiscal quarter increased 11% month-on-month and 27% year-on-year.

Looking ahead to the third fiscal quarter, we expect this accelerated growth trend to continue, with data center revenue expected to increase by more than 20% month-on-month and approximately 75% year-on-year. There are still a wide range of factors driving this growth, and demand for artificial intelligence products continues to grow. We've seen that all data center businesses, including interconnect, switching, and customization, benefit greatly. Strong demand for interconnect and exchange products continues to drive connectivity, making it a key factor in improving the performance of artificial intelligence. To date, the biggest driver of these business growth has been scale-out applications.

Here, Marvell's market-leading position in fiber DSPs, switches, and broadband analog components continues to be in strong demand. In terms of fiber DSP, demand for 800G is still strong, and our 1.6T business is growing rapidly, and this trend is expected to accelerate further in the 2028 fiscal year. In the field of scale-out switches, thanks to the strong growth of 51.2T products in an expanding customer base, our business is expected to double this year. In the field of broadband simulation, demand for our industry-leading TIA and drivers continues to exceed expectations.

Today, we are moving beyond horizontal expansion and expect this connectivity advantage to be further enhanced as hyperscale data centers build larger AI clusters (which increasingly span multiple data centers). As we discussed in our previous conference call, this kind of aggregated bandwidth requirement across the network scale is expected to be more than 10 times higher than current front-end DCI networks. Marvell is uniquely positioned to seize this opportunity with its leading position in the coherent DSP field — 1.6t ZR and ZR+ DCI modules supporting coherent DSPs.

Finally, we still anticipate that Marvell will usher in a huge new market size as the horizontal expansion of networks and the spread of artificial intelligence infrastructure. The horizontal extension domain is expected to expand significantly, requiring high-bandwidth interconnects and high-cardinality, low-latency switches. Currently, customers are initially deploying copper interconnects to achieve scale-out networks, but as clusters grow in size, copper transmission distance and bandwidth limitations are increasingly driving customers to switch to optical interconnects and dedicated UALink ESUN and NVLink scale-out switches. Therefore, we are investing heavily to lead the industry in the next generation of horizontally expanding optical interconnection and switching technology.

In terms of interconnection, pluggable modules are still the main form of packaging for scale-out networks, and we don't expect this to change. However, scaling the network vertically requires much higher bandwidth density, and the best solution is to move the optical module closer to the expansion processor (XPU) and switch. Although the transition from copper to optical cable and the deployment of vertically extended networks are expected to take several years, and the two technologies will coexist, customers are actively planning the deployment of vertically extended optical networks, which may begin as early as next year.

As we are still in the early stages of transformation, customers are evaluating a range of potential solutions and considering various technologies, including NPO and CPO packaging solutions. Both solutions utilize advanced silicon photonics technology and three different modulator technologies: MZM, EAM, and MRM. Each choice has different considerations in terms of cost, power consumption, and technical maturity. Importantly, with our broad optical product portfolio covering multiple modulation techniques and full support for our broadband analog TIA and drivers, Marvell is uniquely positioned to help customers transition to optical architectures that best meet their needs. The full range of solutions developed by Marvell is reflected in our accelerated design activities with many customers. In addition to our continued success in the CPO field, we have also seen many customers strongly support our NPO solutions. As a result, compared with previous expectations, our revenue expectations for large-scale optical products in FY2028 have increased significantly, making Marvell one of the biggest drivers of NPO applications in the field of artificial intelligence infrastructure.

We are also seeing similar growth in the field of expanding exchanges. With its self-developed UAL and eSUN switches and Nvidia's partnership with NVIDIA on NVLink Fusion, Marvell has a unique advantage in supporting all three proprietary expansion protocols. Our extension switches combine decades of experience in developing large lithography chips with our self-developed, industry-leading high-performance SerDes technology. Given our market-leading position in optical devices and switching technology, the close integration of optical devices and switching devices in the extended network provides Marvell with another important differentiating advantage. This allows us to provide highly optimized expansion solutions designed to provide customers with superior performance and reliability while speeding time to market.

As a result, we are in in-depth discussions with a number of Tier 1 customers about our extended switch product portfolio. Given the expected size of the extended switch market, each negotiation has multi-billion dollar lifetime revenue opportunities. All in all, as the demand for performance of artificial intelligence architectures continues to rise in horizontal expansion, horizontal parallel expansion, and vertical expansion, we are extremely excited about the continued acceleration and expansion of network demand.

Now let's talk about the custom business in the data center terminal market. As I mentioned before, demand for customization increased significantly in the second half of this year. We remain confident that this business will more than double its year-on-year growth in FY2028 and achieve significant acceleration in FY2029. We expect this growth to be driven by a combination of our XPU and XPU add-on products. On the XPU side, we have partnered with a number of hyperscale data center operators to continue to make strong progress in existing and next generation projects.

In terms of XPU add-on services, we are benefiting from increased demand for CXL and custom network cards. Looking ahead, we expect the custom business to continue to experience strong revenue growth as we continue to have strong design partnerships with hyperscale customers. The latest example of this growing momentum is our 8-K filing last week, which revealed an extended commercial agreement and related warrants with a major hyperscale data center operator (and one of the largest users of custom chips). The warrant agreement covers ongoing custom projects awarded to Marvell over the past few years, newly won design projects, and potential future projects.

The share option structure reflects the scale and long-term potential of the partnership between the two parties, and further strengthens common interests as our cooperation expands. We have collaborated on a wide range of projects, covering many custom chip projects, including AI inference accelerators, memory controllers, network interface controllers, memory interface controllers, and near-memory computing related to the TPU ecosystem. We look forward to continuing to work closely with this customer to build the next generation AI infrastructure.

The range of additional products covered by this partnership has continued to expand, and the scale has significantly enhanced the market position of the XPU add-on product category pioneered by Marvell. We are pleased to see that the application scenarios and application penetration rate of such products continue to grow. These products are based on Marvell's rich intellectual property portfolio and have specialized features designed to dramatically improve the performance and efficiency of the entire AI data center. This once again strongly confirms Marvell's leadership in connectivity, computing, and memory technologies, and the increasingly important role these technologies play in helping customers build more advanced AI infrastructures.

As far as the impact of the new warrants agreement is concerned, revenue from projects covered by the agreement up to fiscal year 2028 has been reflected in the overall customized business revenue targets we previously provided. Looking ahead to FY2029 and beyond, this agreement, along with several other projects, gives us more confidence to expand our custom business to a larger scale at that time. We look forward to sharing more details on the long-term development plan for the customized business on the upcoming Investor Day.

Next, let's talk about our communications and other terminal markets. Revenue for the second quarter was US$568 million, down 3% from the previous quarter and up 10% from the previous year. Looking ahead, we expect quarterly revenue to fluctuate slightly due to the diverse composition of this business segment. In the third fiscal quarter, we expect revenue to drop around 10% to 15% month-on-month and year-over-year, followed by a strong month-on-month recovery in the fourth quarter.

All in all, the growth momentum of our various businesses is still strong. In the short term, this strong momentum is reflected in our sharp increase in performance expectations. Compared with the forecast a quarter ago, we raised our 2027 revenue forecast by about 500 million US dollars, and raised the 2028 revenue forecast by about 1.5 billion US dollars. The core of this increase in performance expectations is our data center business. The growth rate of this business continues to accelerate, and the scope of business continues to expand. We raised our data center business revenue growth forecast for FY2027 from 50% to about 60%, and we anticipate that the business has the potential to accelerate further growth in FY2028.

In the connectivity sector, we continue to see strong growth in fields such as optical DSP, while broadband analog TIA and drivers, horizontal expansion DCI modules, and horizontal expansion switches have also achieved significant growth. All three businesses have met or surpassed the $1 billion annualized revenue target we emphasized last quarter. The opportunities for horizontal expansion are still huge, and far from being realized. Marvell is uniquely positioned to transition to NPO and CPO optical interconnect and use of dedicated scale-out switches.

Our custom business, including XPU and XPU add-ons, is also on a strong growth trajectory, both in the short and long term. Our 8-K submission last week highlights the expansion of our partnership with a major hyperscale data center operator and further strengthens our confidence that custom chips will play an increasingly important role in the ongoing construction of AI infrastructure. We look forward to sharing more of our insights on Marvell's long-term growth opportunities at Investor Day in New York on October 6, and hope to see you all there.

Next, I'll hand over the phone to Dan to give him a detailed account of our recent performance and outlook.

Daniel Dunne, Chief Financial Officer, Executive Vice President and Chief Financial Officer

Thanks Matt. Good afternoon everyone. Before I dive into financial data, since this is my first earnings call as Marvell's Chief Financial Officer (CFO), I'd like to talk about three things: Why did I join Marvell? What goals do I want to achieve as CFO? And what do I think this ultimately means for our shareholders? First, I joined Marvell because I believe the company has huge potential for growth. I think Marvell is at a unique turning point, with a broad and diverse range of capabilities that have been carefully constructed and perfectly fit the future direction of AI infrastructure development.

Second, as CFO, my goal is to help Marvell scale up efficiently and effectively and seize the huge opportunities I anticipate. The company has always been known for its rigorous execution, and maintaining this rigor is critical in strategically allocating capital to drive growth. Third, I expect the company's growth to create significant value for shareholders. I know that future growth opportunities must match the quality of growth, so I plan to continue to make every effort to improve operating margins, increase cash flow, and drive significant returns to shareholders.

Next, I'll introduce our financial results for the second quarter of fiscal year 2027. Revenue was US$2,739 million, up 37% year over year and 13% month on month. Data centers are our largest end market, contributing 79% of total revenue, up 46% year over year. GAAP gross margin was 53.1%. The non-GAAP gross margin was 58.9%, which is slightly higher than our previous median forecast.

Next up are operating expenses. Operating expenses in accordance with US GAAP were $996 million, which included equity incentives, amortization of intangible assets acquired, restructuring costs, and acquisition-related costs. Operating expenses in non-GAAP terms were $611 million, slightly higher than our expectations. The US GAAP operating margin was 16.8%, while the non-GAAP operating margin was 36.6%. Non-GAAP operating margins increased 180 basis points year over year and 160 basis points month over month, reflecting the significant operating leverage effect in our model. For the second quarter, the diluted earnings per share were $0.33 in accordance with US GAAP. Non-GAAP earnings per diluted share were $0.94, 1 cent higher than our median expectations, and a 40% year-over-year increase.

Now let's look at the cash flow and balance sheet. Cash flow from operating activities in the second quarter was $606 million, a slight decrease from month to month, mainly due to higher capacity advance payments to suppliers to support Marvell's future growth. Inventory at the end of the second quarter was $1.36 billion, a slight decrease from the previous quarter. This quarter, we repurchased $200 million of common stock through an ongoing capital return program and returned $54 million to shareholders through cash dividends. At the end of the second quarter, total debt was US$4.96 billion. The ratio of total debt to EBITDA was 1.32 times, and the ratio of net debt to EBITDA was 0.27 times.

Next, let's talk about our earnings outlook for the third quarter of fiscal year 2027. We forecast revenue of $3.15 billion, fluctuating 5% up and down. We expect GAAP gross margins to be between 52.9% and 53.9%. We expect non-GAAP gross margins to be between 57.5% and 58.5%. Revenue levels and product mix remain the key determinants of gross margin each quarter, and the expected growth in our customized business will provide room for month-on-month growth in the third fiscal quarter — and also create resistance to month-on-month growth. We currently expect gross margin for the fourth fiscal quarter to remain within this range.

We expect GAAP operating expenses of approximately $1,015 million and non-GAAP operating expenses of approximately $655 million for the fiscal third quarter. We expect GAAP other income and expenses (including interest on debt) to be approximately $86 million. Non-GAAP other income and expenses (including interest on debt) are approximately $36 million. We expect the non-GAAP rate to be 11%. Looking ahead, we expect a non-GAAP tax rate of approximately 13% for the 2028 fiscal year, given the significant increase in our revenue and profit expectations. We expect the basic weighted average tradable shares to be about 900 million shares and the diluted weighted average tradable shares to be about 921 million shares, all roughly the same as in the second quarter.

We expect GAAP diluted earnings of $0.48 to $0.58 per share. We expect non-GAAP diluted earnings of $1.05 to $1.15 per share. Looking ahead, we plan to continue investing in the business to drive significant revenue growth over the next few years while leveraging operations. For fiscal year 2027, we expect non-GAAP operating expenses of approximately $2.55 billion, slightly higher than the previous forecast of $2.45 billion, which reflects the greater revenue growth opportunities we are currently seeing. Importantly, we expect revenue to continue to grow much faster than operating expenses. As a result, we expect the operating leverage effect to increase significantly, and the non-GAAP operating margin is expected to reach our long-term target range of 38% to 40% in the fourth quarter of this fiscal year.

Looking ahead to the 2028 fiscal year. We currently expect non-GAAP operating expenses to grow at about half of the revenue growth rate (as a percentage). This reflects our continued investment to seize expanding opportunities while maintaining operational leverage and striving to achieve 38% to 40% of our non-GAAP operating target during the year. Regarding the use of cash. Based on the design orders we have received and our confidence in the continued needs of our customers, we are actively acquiring additional production capacity to support business growth. We are still expected to pay approximately $1 billion in advance capacity payments to suppliers in fiscal year 2027. As a reminder, in line with the guidance we provided last quarter, these prepayments will be used for future material purchases and will be funded through our robust balance sheet and strong operating cash flow. Meanwhile, we plan to continue to buy back shares to control share dilution.

Now let me go back to my original topic. I joined Marvell because I believe the company has excellent opportunities to grow. Now that I'm Chief Financial Officer, I'm more convinced of our strong technology portfolio, deep customer relationships, and huge long-term growth potential. Our next task is to rigorously and efficiently expand the size of the company, seize this opportunity, and ensure that growth translates into increased profit margins, strong cash flow, and rich returns to shareholders. I'm honored to serve as CFO at such an important time in Marvell's development and look forward to working with Matt and the rest of the team to seize the opportunities that lie ahead.

OK, now let's get to the Q&A session. Operator, please connect the line and announce instructions.

Q & A session

presenter

[Moderator's Note] Our first question is Tom O'Malley from Barclays.

Thomas O'Malle, Barclays Research Division

Thank you very much, the results were great. I'd like to start by asking about the collaboration with Google. Obviously, you've had a very strong partnership over the years. I'm curious to know what factors contribute to these revenues? You obviously have inference accelerators, but you've also mentioned the XPU add-on many times, and you're arguably setting an industry standard in terms of definition. So what are some of the XPU's additional features you're excited about? Can you tell me how much revenue each of the inference accelerators and some of the additional features you've seen contribute?

Matthew Murphy, CEO and Chairman

Yes, that's great. Thanks, Tom. Glad to hear from you. I think you've summed it up very well. This is a very exciting time for Marvell. This collaboration and authorization is significant. This is evident from the scale of the opportunities we face. As you pointed out, I would like to confirm a few points. First, it covers a very wide range. It involves so many products and product lines, which is very exciting. As you mentioned, it includes inference accelerators, as well as memory controllers, network cards, memory interface controllers, near-memory computing, etc., and there are many types. You're right, we defined what we call the XPU add-on category a few years ago, and we explained it in great detail at the June 2025 Custom Chip Conference.

I think all of our predictions to date have underestimated actual potential, which means opportunities are becoming more important. Tom, the range of products covered here is very wide, and the range of fields and contributions we can participate in is also very wide. I'm not going to specifically mention any product at the moment, but overall, if you look at the whole opportunity, you'll find that it's significant for Marvell, and it is expected to achieve peak performance within the next 6 to 6.5 years, thereby completely changing the industry landscape.

presenter

The next question comes from Harlan Suhl of J.P. Morgan Chase.

Harlan Suhl, J.P. Morgan Chase Research Division

Welcome Dan and Ross to the team. Matt, given the Google commercial project you signed in late July, the cumulative revenue of the project over six years will reach 120 billion US dollars. If you can reach all the milestones, then the annual revenue for Google XPU with additional custom ASIC chips alone will reach about 18.5 billion US dollars, right? I thought this portion of revenue would start showing in the 2028 fiscal year or 2027, but according to your forecast, it seems that the revenue of custom chips will remain around 5 billion to 6 billion US dollars in 2027.

But perhaps some of the larger projects associated with this commercial agreement are more focused on late-stage development. So perhaps the more pertinent question is: the team had previously set a goal of achieving $10 billion to $11 billion in customized revenue in fiscal year 2029 or calendar year 2028. What would that number be now based on a commercial agreement with Google and some related new projects?

Matthew Murphy, CEO and Chairman

Yes, yes. Good question, Haran. Thank you. First, you're right. Looking ahead to next year, we have taken into account the revenue from this authorization in our current financial data. This is because some of these projects have apparently already started or are being implemented, but those that are being executed or have just begun production will contribute more revenue in the 2029 fiscal year. By the way, about next year, I only said before that the order volume will more than double. I haven't given a specific upper limit or a specific figure, but it will definitely double more than once next year.

Then—next, let's look ahead to fiscal year 2029. On my last conference call, I mentioned that the revenue from the custom business will exceed $10 billion in fiscal year 2029, which, incidentally, is not a new figure. As early as June 2025, we proposed this target, and even in April 2024, we talked about custom business revenue of between $8 billion and $10 billion. So we have always maintained a consistent view of the future of the bespoke business. The opportunity to work with this customer has greatly increased our revenue potential in the bespoke business, with a portion of the revenue likely to start in 2029. I'm not going to give specific numbers on the conference call today, but considering our upcoming Investor Day, Harlan, Marvell team, we always prepare well for Investor Day — that is, Analyst/Investor Day.

Therefore, you can look forward to a very detailed analytical report detailing how we can gradually achieve our revenue goals, not only covering fiscal year 2029, but also looking forward to the next ten years. We are excited to present this report, which will help us better understand the size of the warrants we have just signed and other exciting projects we are undertaking in the field of customization. Please be patient, there will be more to come. However, it is clear that there is a lot of room for growth in customized business revenue data for fiscal year 2029 and beyond.

Harlan Suhl, J.P. Morgan Chase Research Division

Yes, of course. I'm really looking forward to it.

presenter

The next question comes from Vivek Ayah of Bank of America Securities.

Vivek Arya, Bank of America Securities Research

I wish Dan and Ross all the best. Matt, I'd like you to tell us about another major XPU project you plan to launch next year. I think next year's budget is quite conservative, about 600 million to 700 million US dollars, so I'd like you to update on the latest developments in this project. Furthermore, what are the long-term development prospects of this project? Since this hyperscale data center operator, like their rivals, is just getting involved in the XPU field, I think they definitely want to grow this XPU project to a much larger scale than it is now. So could you please tell us about the current state of progress? Does this project really make sense to you in the long run?

Matthew Murphy, CEO and Chairman

Yes, thanks, Vivek. Yes, we are very encouraged by the customized plan for next year's double growth plan. This new project is clearly part of it. I think we've been evaluating it with a very conservative and pragmatic attitude as we've progressed, and we've been improving every quarter, not only in terms of design execution, but also in terms of supplying commercial channels and figuring out the scope of this opportunity. Things are only getting better. So, although it's not quantifiable, what I want to say is that judging from the last conference call, if you look at the $1.5 billion we raised next year, from a macro perspective, it covers a wide range of areas.

So, part of the growth will come from next year's customization business. Also -- in fact, I have another question about this -- the most important part of next year's growth may actually come from an increase in the scale of optical device switching in other fields, but the customization business will definitely also account for a large part, Vivek. So even when we evaluated this opportunity (which dates back a few years), we always thought it might be one of Marvell's biggest revenue opportunities, and it still is. We are therefore very encouraged by this project—the future of this project—. There will be more progress in the future, but it is definitely progressing well so far, and we are confident that it will develop next year and the next few years.

presenter

The next question comes from Wells Fargo's Aaron Rakers.

Aaron Rakers, Wells Fargo Securities LLC Research Division

Welcome Dan and Ross. Perhaps the previous question could be extended a bit. Matt, can you talk about the CXL trend you see from an architectural perspective and how it's evolving as Structera products evolve? Clearly, the ecosystem around Structera products is constantly expanding. Also from an architectural perspective, are there any recent developments with regard to the expandable optics you just mentioned, and the Celestial AI data you've outlined earlier? How do these developments compare to your initial goals?

Matthew Murphy, CEO and Chairman

Of course. So I'll start with CXL and then discuss expandable optics. When it comes to CXL, I mean that part of the technology we've been investing in over the years has grown and ultimately been a huge success for a few reasons. As you may recall, it started out as a traditional computational memory architecture centered around servers. All of our investments proved to be perfect for memory expansion and reasoning.

As a result, we are seeing this technology being deployed in many hyperscale data centers with various architectures, and on a very large scale. On the one hand, there is a need for reasoning ability, and on the other hand, it is due to the scarcity of memory resources. Customers are adjusting their plans to make greater use of this type of technology. As a result, this technology continues to maintain a strong growth momentum. Over the past few quarters, we've won a few more design orders in this area. We'll cover these in more detail on Investor Day. But Aaron, this opportunity is really huge.

I think the strategic adjustments we made about five years ago — vigorously promoting internal organic investment in a number of new technologies — are now beginning to pay off. This is very exciting, and you'll see a very comprehensive memory expansion section in our Investor Day presentation.

As far as scaled optics is concerned, I think this is becoming one of our most exciting fields. Last quarter, we focused on Celestial AI's CPO photonic structure solution. According to reference, the program's budget for next year is around US$150 million. We also mentioned at the time that the budget for the entire large-scale field of optics, including CPO (and NPO, by the way) is about 300 million US dollars. As I said in my prepared statement, all of this combined, the field of large-scale optics is growing faster than the numbers I gave last quarter. So Aaron, our $1.5 billion funding plan for next year is largely due to this sector. And it's not a single factor. I want to emphasize that the CPO is definitely a key component of this. This is true not only next year, but the year after. However, if I look at the current opportunities to grow in parallel with our existing CPO programs, I'll see that NPO opportunities are growing much faster.

In NPO, we are involved through various design projects won by us and our partners. Some of these projects involve our silicon germanium broadband analog products, including TIA and drivers, in which we have significant market share and content, and others involve our self-developed NPO solution, which we call the Light Engine. We have been developing it for several years and have been showcasing it at OFC for several years.

Aaron, so all of these fall into one category, which is exactly how we think about business. Certainly, Celestial and Photonic Fabrics are key parts of it. But I want to emphasize that this is not about choosing one of the two, but rather a combination of the two. It's a bit like people used to think that with an 800G DSP, they would upgrade to 1.6T, and everything would be seamless. But things are different now. Our customers are adopting different architectures. They implement many different solutions simultaneously. So we think it's a combination of the two, rather than choosing one of the two. I think the scale of our opportunities for scaling up optical devices next year will far exceed our expectations three months ago. The next year, I think it will be far larger than I could have imagined when I visited Celestial last year.

If you take a look at the solutions we provide (as you've heard from my last review), we're fully committed to advancing our switch roadmap and combining these fiber solutions with our switches (including UAL and ESUN), and being able to seamlessly integrate and architect our customers in terms of XPU/GPU.

In summary, we have complete end-to-end solutions, as well as scale advantages and ten years of experience in various fields such as manufacturing, R&D, packaging, modulator technology, and silicon photonics. This is a huge opportunity, and Marvell is involved with its unique strengths. We are very happy with the Celestial team, they are well integrated and are an important part of the project. But for Marvell, this will be a larger overall business.

presenter

The next question comes from Morgan Stanley's Joe Moore.

Joseph Moore, Morgan Stanley Research Division

Similarly, can you talk about progress in scaling up copper cabling? You mentioned that all three agreements were successful. What do you think of the large-scale transition of copper cabling to fiber? Did these early successes lay the foundation for your work in the field of fiber optics? Or are these two separate decisions?

Matthew Murphy, CEO and Chairman

Yes, thanks, Joe. I think you're talking about it from a switch perspective, right? Once we have implemented these cable/copper solutions, how can these successful experiences be applied to the field of fiber optics? There are a few points that need clarification. First, in terms of Ethernet, our Teralynx architecture was a huge success, thanks to our acquisition of Innovium in 2021. This business is growing strongly; we've expanded our customer base, achieved significant revenue, and proven to the market that we can deliver these solutions in bulk.

So, Joe, this could very seamlessly translate into horizontal scaling. In fact, we've seen that even at 100T scale, our scale-out solution is just as suitable for vertical scaling. Therefore, our drive to achieve this goal has nothing to do with whether the optical module is compatible. The same goes for UAL. We are investing heavily in UAL exchange, and the development of this product is progressing smoothly, and we can of course replace any solution based on electricity or copper cabling. What customers really want to see, and the reason I put in my time, is: It's all great. Currently, they are confident that we can deliver these very complex exchange products because we have done it, which, incidentally, is no easy feat. Many companies have tried to build such large-scale, extremely complex Ethernet switches, but have failed, and as SerDes performance and speed have improved, it will only become more difficult.

But Joe, to go further, you need to show a clear and compelling roadmap to prove that you can effectively implement fiber, including NPO (non-fiber access) and ultimately full integration with CPO (copper access). So while there will definitely be development opportunities for us in copper cabling in the next few years, I think the reason we can really win in the long term is because we can convince our customers that we can provide a complete end-to-end solution. This is what customers ultimately really need, especially during large-scale deployments. They need to ensure reliability at both ends of the link and trust both ends of the link and everything in between. They found a partner who could do the job.

We believe that piecemeal solutions alone cannot currently accomplish the task. You must have an end-to-end solution portfolio. Incidentally, this is also the feedback we've heard from our customers. Therefore, the performance of existing chips will be good, or even excellent, but I am looking forward to the development of optical interface technology in the next few years, as it can effectively nearly double the interface capacity (SAM) of switches. Thank you.

presenter

Your next question comes from Ben Reitzes from Melius Research.

Benjamin Retzmelly, Ussuri Research LLC

Matt, I'd like to talk more about Google deals. I think everyone is struggling with this. I know you're hosting a great Analyst Day event soon. So... don't blame me. However, as far as the fiscal year is concerned, if you spread this 120 billion dollars over several years, that is about 18 billion US dollars per year, then divide 120 billion US dollars by 6.5 billion dollars, which is equivalent to adding one Marvell project for fiscal year 2028 every year. I know we don't know how many of these are new, and how many have already been mentioned in your guide.

However, we said, “Will FY2029 and FY2030 see significant growth as a result?” It's because these numbers are huge. I know you'll be announcing it on Analyst Day but I think we're all trying to figure out how high should we aim for the next few years? And is the profit margin comparable to previous years? It would be great if you could take our perspective and provide some more detailed information, such as how we should respond to Analyst Day, or simply wait.

Matthew Murphy, CEO and Chairman

Yes, Ben, thanks. BTW, your question is totally reasonable. I mean, when you consider the scale of this thing, you're not wrong with your calculations. The conclusion I can draw from what I just said is that since most of the impact will already be felt next year, the real impact will occur in 2029 and beyond. So — if you take all the factors and opportunities into account, then you're right, that's simply — it's an astronomical number.

What I'm saying is — so we do need Analyst Day, and I think you all understand that we need to analyze the actual situation and show some possible ranges of results. But you should assume that during that time period, the custom business figures will be much larger than anyone's currently predicting. I mean, for years, people have been wondering if we can reach the $8 billion to $10 billion target. I think this should reassure investors that we've received quite a few projects, not only in this field, but across our broad customer base.

Overall, this is a bespoke business, and we have a financial model specifically for a bespoke business. We also have financial models for standard products and merchant products. The growth of the custom business is largely in line with this model, but the impact is huge. I can't give specific quantitative data today, but you can assume that starting in fiscal year 2029, the growth of the custom business will definitely exceed the growth predicted by your previous model based on existing data. As for how big the increase is, we're happy to show you all the possible outcomes.

But as far as Marvell was at the time, Ben, I mean, we were an $8 billion company last year. We just surpassed $12 billion this year. This year isn't over yet; next year it will reach $18 billion. At a rough estimate, if all projects had the best results, the company would take an amazing leap in size. But I think it helps to show the overall scale of opportunities Marvell is involved in. Because we have collaborated so much with so many customers. This situation is quite special because it involves warrants, and given their size and equity aspects, we need to disclose relevant information.

But I think you've all seen over the past few years that we've established very important strategic partnerships with other giants in the AI market. Some of them we've announced publicly, and others we've done privately. So, I think the most important point right now is that it strongly validates Marvell's current position in the market, and we're very honored to be part of the TPU ecosystem. Let's wait and see. But that's all I want to say today. I'd like to reserve some content for my Investor Day.

Benjamin Retzmelly, Ussuri Research LLC

I am grateful and congratulate you on closing this deal. See you later.

presenter

Your next question is from CJ Muse by Cantor Fitzgerald.

Christopher Muse, Cantor Fitzgerald Company Research Division

I think one of the most interesting points in last night's Nvidia report was Jensen's views on the interchangeability of computing, networking, and memory. In an environment where costs are limited, the three can work together to improve the performance of AI data centers. Considering that you have extensive experience and expertise in all three areas, how do you think this background will support your new design plan at current memory price levels? Do these solutions really focus on memory controllers and custom HBM to improve performance? Or is there an improvement on the XPU side as well? I'm looking forward to hearing from you.

Matthew Murphy, CEO and Chairman

Yes, thanks, CJ. No — first, I agree with you. I totally understood what they said. I think at the broadest level, this requirement for interchangeability is critical given the dynamic changes in the market and some of the limitations we face. Some of the restrictions are supply restrictions, some are power restrictions, and some are architectural restrictions. Marvell must have the ability to quickly execute custom and semi-custom designs, modify products, or adapt to changing architectures. It's one of the company's key competencies, and it's the flexibility the company has, because we've built our business around the ability to quickly customize, which has even been in our DNA since Investor Day 2021. We mentioned at the time that every part of the network will eventually be customized to a certain extent, not just an accelerator. Five years have passed, and we have done it.

CXL and memory extensions are an example, and customers quickly adapted to these technologies. Additionally, there are other optimization opportunities, such as improving reasoning skills. This is mentioned in warrants, and it is also one of the market trends. Artificial intelligence inference accelerators are another market trend. So I think all companies are looking at how to maximize performance and how to maximize cost and performance relative to the tokens they generate.

We are in an age of commercialization. So these techniques are really important. It has rapidly evolved from the training stage to the inference stage, and now companies are starting to use these techniques to profit. As a result, we are seeing a significant increase in activity in related fields. And it's not just about customization. What I mean is that the technologies we've discussed before, such as CPO, NPO, optics, and copper cabling, are evolving rapidly because companies are designing architectures at an unprecedented speed to ensure they have the most competitive solutions.

So, CJ, Marvell fits your needs exactly in many ways. This is reflected not only in the revenue from our “custom” business, but also in the underlying solutions we provide, including storage and memory in our networking, storage, and computing sectors.

presenter

The next question comes from Jim Schneider at Goldman Sachs.

James Schneider, Goldman Sachs Research Division

I'd like to ask you to give us a brief overview of your expectations for the growth rate of connectivity. Matt, I remember last quarter you already outlined the growth rates for this fiscal year and the next fiscal year. Can you update your expectations for these growth rates given the current strong growth momentum? Sounds like most of the current growth is coming from this segment?

Matthew Murphy, CEO and Chairman

Yes, thanks Jim. As I said, this year's $500 million financing and next year's $1.5 billion financing plan are all-round. Connectivity is clearly one of the driving factors; I'm talking about connectivity at a broader level because we've previously discussed expanding optical devices, which include our NPO, CPO, and SiGe products. Furthermore, the transceiver market for horizontal expansion and optical DSPs entering this field are also growing at a higher rate than we previously discussed. So, these $500 million and $1.5 billion financing plans all cover this part of the business. The business continues to be hot, and we are doing a good job in this area.

Furthermore, in the field of connectivity and more broadly, our switching business has high expectations for this year, and according to our predictions for the previous quarter, it will be the same next year. Therefore, these three businesses, plus what I think is a customized business, have surpassed our previous expectations. Jim, we're really seeing strong growth overall. I think almost every product line shows this trend as we look to the future, but these are just a few of the aspects I want to focus on. You're right, the connectivity business is larger, and it is probably the biggest driver of this $1.5 billion funding. If I just look at all of the businesses I've mentioned below, such as horizontal expansion, vertical expansion, and switch business, I'll see that they all perform well.

presenter

Today's last question comes from Needham & Company's Quinn Bolton.

Quinn Bolton, Needham & Company, LLC Research Division

I think, Matt, I just wanted to talk more about recent performance guidelines. In the data center sector, I remember your month-on-month growth forecast of more than 20%. However, since profit margins fell 90 basis points month-on-month, did the fastest growing business segment within data centers — customized data centers — cause profit margins to decline? Or are there other factors within the data center that caused profit margins to drop by 90 basis points — or more accurately, the reason for the October quarterly results guide being lowered by 90 basis points?

Matthew Murphy, CEO and Chairman

OK, thank you. Let me kick things off first, but I want Dan to speak because you guys have been so forgiving of him during this conference call. I mean, he's the new CFO, and you should have asked him questions, but it turned out to be asking me all the questions. A joke is a joke, you can tell from the data. The stronger the third-quarter earnings forecast, the better the fourth-quarter forecast. We did say that the number of customers increased significantly in the second half of the year. I said it, and so did Dan. Dan, maybe you can talk about profit margins, how you think profit margins will change, and next year and how our business will grow.

Daniel Dunne, Chief Financial Officer, Executive Vice President and Chief Financial Officer

Of course. Well, let's get straight to the point. Obviously, the company's performance is good. You can see accelerated growth in the second quarter, continued to accelerate in the third quarter, and so did the fourth quarter. As you pointed out, the accelerated growth in the third quarter was partly due to the product portfolio being the main driver. Our bespoke business is growing strongly. So you can see this from the profit margin analysis. This is no surprise. We have already predicted that the growth momentum of the custom business will continue for some time. Looking ahead to the fourth quarter, we expect this strong momentum to bring even greater growth. From a revenue perspective, if we consider the annual target of 12 billion US dollars, the increase will be very significant.

Growth in the fourth quarter was all-round. Growth can be seen not only in the customized business, but also in the connected business. There was also a rebound in communications and other businesses. As a result, the overall performance was good. All in all, we expect gross margin for the fourth quarter to remain within the same target range as the third quarter. If we take the next six months as a starting point and look forward to the 2028 fiscal year, this all-round strong momentum will continue. We continued this year's positive momentum into next year, and achieved growth in all aspects of the customization business and connectivity business. All business segments have improved.

The gross margin depends on the final product mix. My initial opinion is that next year's gross margin will be in a similar range to the gross margin at the end of this year, that is, the same as the gross margin range for fiscal year 2028. Therefore, we are confident in the company's performance. We will continue to drive the company's growth and achieve this goal with rigor while ensuring strong profit margins. However, as far as profit margins are concerned, operating profit margins still have a considerable leverage effect. You'll see an increase in operating margins in the third quarter. You'll see us enter the long-term target model range, which is 38% to 40% by the end of this year. As the 2028 fiscal year progresses, you'll see us reach the upper limit of this range. We'll be readjusting this long-term target model on Analyst Day in the next few weeks. Therefore, we are confident in the company's performance and overall strength.

presenter

Thank you all. Ladies and gentlemen, thank you all for participating. That concludes today's conference call. You can disconnect and have a great day.