邁威爾 (MRVL) 2027 財年第二季法說會：2028 財年展望上調至 180 億美元

邁威爾科技 (Marvell Technology, NASDAQ: MRVL) 2027 財年第二季法說會重點聚焦於加速成長的 AI 資料中心需求、調升的全年展望，以及顯著擴大 2028 財年的營收目標。

主要重點

• 2027 財年第二季營收達創紀錄的 27.39 億美元，季增 13%，年增 37%。Non-GAAP 稀釋後每股盈餘 (EPS) 為 0.94 美元，年增 40%。
• 資料中心營收升至創紀錄的 21.7 億美元，季增 18%，年增 46%。該業務部門占總營收的 79%。
• 管理層將 2027 財年的營收展望從原先的約 115 億美元調升至約 120 億美元，隱含約 45% 的年成長率。
• 邁威爾科技目前預計 2028 財年營收約為 180 億美元，高於先前預期的 165 億美元，代表約 50% 的成長率。
• 資料中心營收預計在 2027 財年成長約 60%，並在 2028 財年成長超過 60%。管理層指出，互連、交換器與客製化晶片領域皆展現廣泛需求。
• 邁威爾科技預計其客製化業務將在 2028 財年成長一倍以上。與擴大之超大型雲端業者 (Hyperscaler) 認股權協議相關的計畫已納入 2028 財年展望中，而管理層預計在 2029 財年及以後會有更顯著的貢獻。

核心財務數據

業務與營運績效

AI 資料中心需求持續擴大

邁威爾科技表示，資料中心業務成長正跨越互連、交換器與客製化晶片領域全面加速。針對 2027 財年第三季，管理層預計資料中心營收將季增 20% 以上，年增約 75%。

外擴式 (Scale-out) 連接能力仍為主要驅動力。800G 光學數字訊號處理器 (DSP) 需求持續強勁，公司預計其下一代業務將在 2028 財年更快速量產。邁威爾科技亦表示，在 51.2T 產品推廣至更廣泛客戶群的支援下，其 scale-out 交換器業務在 2027 財年仍按計畫將成長一倍以上。

用於跨阻抗放大器 (TIA) 與驅動器的寬頻類比需求持續超出管理層預期。隨著超大型雲端業者建置跨多個資料中心的 AI 集群，跨域 (Scale-across) 需求也隨之興起，支撐了邁威爾科技基於相干 DSP 的 1.6T ZR 與 ZR+ 資料中心互連模組。

擴增式 (Scale-up) 光學與交換技術

管理層將 scale-up 網路視為一項龐大的長期機會。雖然銅纜目前用於初期部署，但邁威爾科技預計隨著 AI 集群擴大，傳輸距離與頻寬限制將推動更多光學互連採購。該過渡預計需要數年時間，銅纜與光學技術將並存。

客戶正在評估 NPO 與 CPO 封裝、矽光子學以及多種調變器技術。邁威爾科技指出多家客戶對其 NPO 解決方案的採用率有所提高，並調升了 2028 財年 scale-up 光學營收展望（相比先前預期），但未提供具體新數字。

公司亦在為 UALink 和 ESUN 開發專用交換器，並與輝達 (NVIDIA) 就 NVLink Fusion 展開合作。管理層表示，邁威爾科技正與多家一線 (Tier 1) 客戶進行溝通，每次合作都代表潛在數十億美元的生命週期營收機會。不過，這些仍屬潛在機會而非已確定之營收預測。

客製化晶片與 XPU 附屬產品

邁威爾科技預計其客製化業務將在 2027 財年下半年顯著放量，在 2028 財年成長一倍以上，並在 2029 財年進一步加速。成長動力預計來自 XPU 及 XPU 附屬產品，包括 CXL 裝置和客製化網路介面控制器 (NIC)。

公司與某主要超大型雲端業者擴大簽署的協議涵蓋現有計畫、新獲取的設計定案 (design win) 以及未來潛在計畫。產品類別包括 AI 推理加速器、儲存控制器、網路介面控制器、記憶體介面控制器與近記憶體運算。

管理層強調，截至 2028 財年涵蓋計畫的營收已納入其現有的客製化營收目標中。該協議與額外計畫提高了管理層對 2029 財年起更龐大客製化業務的信心，但管理層將詳細預測延後至 10 月 6 日的投資人日公布。

CXL 與記憶體擴充

邁威爾科技表示，其 CXL 技術正於多家超大型雲端業者以不同架構大量部署。管理層將此機會歸因於 AI 推理的記憶體擴展需求，以及客戶對記憶體短缺的應對措施。公司在過去兩個季度中亦獲得了額外的 CXL 設計定案。

通訊及其他市場

通訊及其他部門第二財季營收為 5.68 億美元，季減 3%，但年增 10%。儘管各季度存在波動，管理層仍預計 2027 財年這些終端市場的成長率將接近 10%。

管理層財測

管理層預計第四財季營收成長在季增與年增方面均將進一步加速。預計第四財季 Non-GAAP 毛利率將保持在第三財季的區間內，且視產品組合而定，2028 財年預期亦將維持類似區間。

邁威爾科技預計其 Non-GAAP 營業利益率將在 2027 財年第四季進入 38% 至 40% 的長期目標區間，並隨著 2028 財年的推進達到該區間的上緣。2028 財年 Non-GAAP 營業費用的成長率預計約為營收成長率百分比的一半。

公司在 2027 財年支付約 10 億美元供應商產能預付款的進度仍符合預期。管理層表示，這些預付款將用於未來的材料採購，資金來源為邁威爾科技的資產負債表與營業現金流。

風險與關注焦點

• 全產業的供應瓶頸仍然普遍存在。邁威爾科技正透過包括大幅增加供應商預付款在內的方式獲取額外產能，以支援預期的成長。
• 客製化晶片加速量產對毛利率組合構成逆風。管理層預計第三財季 Non-GAAP 毛利率為 57.5% 至 58.5%，低於第二財季的 58.9%。
• 通訊及其他市場表現仍不穩定。預計第三財季營收將在季增與年增方面出現 10% 中低段的下滑，隨後在第四財季出現季增復甦。
• 擴增式 (scale-up) 網路中從銅纜轉向光學的過渡預計需耗時數年，且客戶仍在評估多種封裝與調變技術。
• 擴大之超大型雲端業者協議的完整營收潛力取決於計畫執行力與績效。管理層指出最大增量影響將發生在 2029 財年及以後，但未予以量化。

分析師問答亮點

• 擴大超大型雲端業者協議：管理層表示，該合作關係涵蓋多個客製化晶片類別，而非單一加速器。2028 財年營收已納入財測中，而 2029 財年及以後年份可能會帶來顯著更大的貢獻。
• 2028 財年展望調升：連接能力被認為是 2028 財年營收展望增加 15 億美元的最主要驅動力。擴增式 (scale-up) 光學、外擴式 (scale-out) 光學 DSP、交換器與客製化晶片皆有貢獻。
• 擴增式 (Scale-up) 光學：邁威爾科技先前的框架預估 2028 財年 scale-up 光學營收約為 3 億美元，其中包括來自 Celestial AI 相關 CPO 光子網路架構 (photonic fabric) 解決方案的約 1.5 億美元。管理層表示目前機會已顯著擴大，但並未提供最新數字。
• 客製化 XPU 計畫：管理層表示，另一項大型 XPU 計畫在設計執行、供應與商業安排方面持續推進。這仍是公司預期 2028 財年客製化業務營收翻倍以上的組成部分。
• 利潤率狀況：第三財季毛利率下降主要是由於客製化晶片放量導致的產品組合轉變。管理層仍預計營業槓桿效益將在財年底前將 Non-GAAP 營業利益率推升至 38% 至 40% 的目標區間。

法說會完整逐字稿

---

完整財報電話會議逐字稿

管理層陳述

Operator

Good afternoon, and welcome to Marvell Technology Inc. Second Quarter of Fiscal Year 2027 Earnings Conference Call. [Operator Instructions] Please note, this event is being recorded.

I will now turn the conference over to Mr. Ross Seymore, Senior Vice President of Investor Relations. Thank you. You may begin.

分析師問答

Ross Seymore

Thank you, and good afternoon, everyone. Welcome to Marvell's Second Fiscal Quarter 2027 Earnings Call. Joining me today are Matt Murphy, Marvell's Chairman and CEO; Dan Durn, CFO; Chris Koopmans, President and COO; and Sandeep Bharathi, President of our Data Center Group. Let me remind everyone that certain comments today -- made today include forward-looking statements, which are subject to significant risks and uncertainties that could cause our actual results to differ materially from management's current expectations.

Please review the cautionary statements and risk factors contained in our earnings press release, which we filed with the SEC today and posted on our website as well as our most recent 8-K, 10-K, 10-Q and other documents that we file from time to time with the SEC. We do not intend to update our forward-looking statements. During our call today, we will refer to certain non-GAAP financial measures. A reconciliation between our GAAP and non-GAAP financial measures is available in our earnings press release.

Let me now turn the call over to Matt for his comments on the quarter. Matt?

Matthew Murphy

Thanks, Ross, and good afternoon, everyone. Before I discuss our results and outlook, I want to briefly highlight 2 management transitions that occurred during our last quarter. First, Willem Meintjes stepped down as Marvell's CFO in mid-June. I deeply appreciate Willem's steady hand, leadership and tireless commitment to transforming Marvell over his decade with the company, and I greatly respect his desire to spend more time with his family. To ensure a seamless transition, we simultaneously welcome Dan Durn as our new CFO. Dan brings more than 3 decades of experience in senior finance roles across semiconductor and enterprise technology company.

And having most recently served on Marvell's Board of Directors, Dan comes into the role with a deep understanding of our business and strategy as well as a unique appreciation for the significant growth opportunities at Marvell ahead. Second, in July, we began a transition in our Investor Relations leadership. After 8 years with Marvell, Ashish Saran will retire from the company in April 2027. The I want to personally thank Ashish for his leadership, partnership and countless contributions to Marvell. I also appreciate the thoughtfulness and care he has brought to planning this transition, including helping us identify the right successor.

On that front, I'm very pleased to welcome Ross Seymore, who comes to us from Deutsche Bank, where he covered the semiconductor industry for more than 25 years. A warm welcome to you, Dan and Ross. Now let me move on to our results and outlook. For the second quarter of fiscal 2027, Marvell delivered record revenue of $2.739 billion, reflecting 13% sequential and 37% year-over-year growth. Revenue and non-GAAP earnings per share of $0.94 both exceeded the midpoint of guidance. On our last earnings call, we increased our sequential revenue growth expectation for the third and fourth quarters of this fiscal year to double-digit percentage growth, up from our prior outlook of high single-digit growth.

Since then, our outlook has continued to strengthen, and we now expect revenue growth to further accelerate in the second half. The strength is reflected in our guidance for the third quarter of fiscal 2027 where we expect total company revenue of $3.15 billion at the midpoint, representing growth of 15% sequentially and more than 50% year-over-year. We expect growth to further accelerate in the fourth quarter, both sequentially and year-over-year. As a result, we now expect overall Marvell revenue in fiscal 2027 to grow approximately 45% year-over-year to roughly $12 billion, up from our prior outlook of approximately $11.5 billion just 1 quarter ago.

The increase in our revenue outlook continues to be driven by our Data Center business, which we now expect to grow by approximately 60% this fiscal year, up from our prior expectation of approximately 50%. Importantly, this growth remains broad-based. Interconnect continues to lead the way, while our custom business is expected to ramp significantly in the second half. For our communications and other end markets, the trajectory remains largely as expected. Despite typical quarter-to-quarter lumpiness in these businesses, we currently expect fiscal 2027 growth to approach our 10% target.

Looking ahead to fiscal 2028, aggregate demand continues to accelerate, and our operations team is doing an outstanding job securing additional supply despite pervasive industry-wide constraints. As a result, we now expect Marvell's Data Center revenue to grow more than 60% year-over-year in fiscal 2028, driven by strong growth across all of our key data center businesses. This includes custom more than doubling as we indicated last quarter. We look forward to providing a deeper dive into the specific drivers of our longer-term growth at our Investor Day in New York City on October 6, but the key takeaway for today is clear.

The strength of our data center business continues to exceed our prior expectations. Putting it all together, we now expect fiscal 2028 revenue of approximately $18 billion, up $1.5 billion from the $16.5 billion outlook we provided just 1 quarter ago. And importantly, even as our revenue base becomes significantly larger, our growth rate is accelerating. We now expect fiscal 2028 revenue to grow approximately 50% year-over-year compared with approximately 45% in our prior outlook.

With that, let me provide color on our current business, beginning with Data Center. In our Data Center end market, we delivered record second quarter revenue of $2.17 billion, representing 18% sequential growth and 46% year-over-year growth. Both sequential and year-over-year growth accelerated from the first fiscal quarter when Data Center revenue increased 11% sequentially and 27% year-over-year.

Now looking ahead to the third fiscal quarter, we expect this acceleration to continue with data center revenue forecasted to grow more than 20% sequentially and roughly 75% year-over-year. The drivers of this growth remain very broad-based as AI demand for our products continues to rise. We are seeing strong tailwinds across each of our data center businesses, including interconnect, switching and custom.

Connectivity continues to be a critical enabler of AI performance driven by robust demand for both our interconnect and switching products. Thus far, the largest driver of growth for these businesses has been for scale-out applications. Here, Marvell's market-leading franchises in optical DSPs, switching and broadband analog components continue to see significant demand. On the [indiscernible] DSP side, 800 gig demand remains strong, while our [indiscernible] business is ramping rapidly, a trend we expect to accelerate further in fiscal 2028. Within scale-out switching, our business remains on track to more than double this year driven by a strong ramp in our 51.2 T products across a broadening array of customers. And within broadband analog, demand for our industry-leading TIAs and drivers continues to exceed expectations.

Now moving beyond scale out, we expect this connectivity strength to broaden as hyperscalers build significantly larger AI clusters that increasingly span multiple data centers. As we have discussed on prior calls, aggregate bandwidth requirements for these scale across networks are projected to be more than 10x greater than those of current front-end DI Networks. Marvell is ideally positioned to address this opportunity through our leadership in coherent enabled DSP -- coherent DSP enabled 1.6 T ZR and ZR+ DCI modules.

Finally, we continue to expect the adoption of scale-up networking and AI infrastructure to create a massive new TAM for Marvell. Scale-up domains are expected to become significantly larger, requiring high bandwidth interconnects closely coupled with high rates low-latency switches. Now while customers are initially deploying copper interconnect for scale-up networking, as cluster sizes grow, the reach and bandwidth limitations of copper are increasingly driving customers towards optical interconnects as well as purpose-built UALink, ESUN and NVLink scale-up switches. As a result, we are investing aggressively to lead the industry in next-generation scale of optical interconnect and switching technologies.

On the interconnect side, pluggable modules remain the primary form factor for scale-out networks, and we do not expect that to change. However, the significantly higher bandwidth density required by scale up networks is best served by bringing optics much closer to XPUs and switches. While the transition and scale-up networks from copper to optics is expected to take several years with both technologies coexisting, customers are aggressively planning scale-up optics deployment starting as early as next year.

Given how early we are in this transition, customers are evaluating a broad range of potential solutions with multiple technologies under consideration, including NPO and CPO packaging options, with both leveraging advanced silicon photonics as well as 3 different modulator technologies, MZM, EAM and MRM. Each of these choices has different considerations around cost, power and technology maturity. Importantly, given the breadth of our optical portfolio across modulation technologies fully supported by our broadband analog TIAs and drivers, Marvell is uniquely positioned to help customers move towards the optical scale of architecture that best meets their needs.

The full spectrum of Marvell developed solutions is reflected in accelerating design activity with a broad set of customers. In addition to our ongoing success in CPO, we are also seeing a strong adoption of our NPO solutions at multiple customers. As a result, our fiscal 2028 revenue outlook for scale-up optics has increased meaningfully compared to prior expectations, positioning Marvell to be one of the largest enabler of NPO in AI infrastructure.

Moving to scale-up switching, we are seeing similar momentum. Marvell is uniquely positioned to support all 3 purpose-built scale-up protocols through our internally developed UAL and ESUN switches as well as our partnership with NVIDIA around NVLink Fusion. Our scale up switches leverage decades of experience developing large reticle size switch silicon combined with our in-house best-in-class high-performance SerDes technology. the close coupling of optics and switching in scale-up networks provides another important differentiator for Marvell, given our market-leading positions in both technologies.

This allows us to deliver highly optimized scale-up solutions designed to provide customers with exceptional performance and reliability while accelerating time to market. As a result, we are engaged with in multiple deep discussions with Tier 1 customers across our scale-up switch portfolio, with each engagement representing a multibillion-dollar lifetime revenue opportunity given the expected size of the scale-up TAM. Taken together, we are extremely excited about the continued acceleration and broadening of networking demand as AI architectures require ever-increasing performance across scale out, scale across and scale up domains.

Okay. Now let's turn to the custom business within our data center end market. As I mentioned earlier, we are seeing a significant acceleration in custom demand in the second half of this year. We remain confident that this business will more than double year-over-year in fiscal 2028 and accelerate significantly in fiscal 2029. We expect this growth to be driven by both our XPU and XPU attached products. In XPUs, we continue to make strong progress across current and next-generation programs at multiple hyperscalers.

And the XPU attach, we are benefiting from increasing demand for both CXL and custom NIC. Looking further ahead, we expect our custom business to continue to deliver strong revenue growth as we see ongoing robust design activity with hyperscale customers. The most recent example of this momentum is the day we filed last week, disclosing an expanded commercial agreement and associated warrant with a key hyperscaler, one of the largest adopters of custom silicon. The warrant agreement encompasses custom programs already in execution that were awarded to Marvell over the past several years, new design wins and future potential programs.

The warrant structure reflects the scale and long-term potential of the relationship and further aligns common interest as our work together expands, expands a broad range of custom silicon programs, including those that attached to the TPU ecosystem such as AI inference accelerators, storage controllers, network interface controllers, memory interface controllers and near-memory compute. We look forward to continue working closely with this customer to enable the next generation of AI infrastructure.

This expanding range of attached products and the scale of this agreement provides significant validation of the XPU attached category that Marvell has pioneered. We are excited to see both the use cases and attach rates for this category of products continue to broaden. These products are built upon a rich portfolio of Marvell IP and perform specialized functions designed to significantly improve the performance and efficiency of the overall AI data center. This is another strong confirmation of Marvell's leadership in connectivity, compute and memory technologies and the increasingly important role they play in enabling our customers to design more advanced AI infrastructure.

In terms of the impact of the new warrant agreement, revenue from programs covered by the agreement through fiscal 2028 is already reflected in the overall custom revenue target we have previously provided. Looking at fiscal '29 and beyond, this agreement, along with several additional programs gives us even greater confidence in our ability to grow the custom business to a significantly larger scale in that time frame. We look forward to sharing more details on the long-term trajectory of our custom business at our upcoming Investor Day.

Turning to our communications and other end markets. We delivered second quarter revenue of $568 million, down 3% sequentially and up 10% year-over-year. Going forward, we expect to remain -- revenue to remain somewhat lumpy on a quarterly basis, given the mix of businesses in this segment. For the third fiscal quarter, we expect revenue to decline in the low to mid-teens percentage range both sequentially and year-over-year, a solid sequential recovery in the fourth quarter.

To summarize, the momentum across our business remains exceedingly strong. In the near term, that strength is reflected in the significant increases to our outlook. Compared with the expectations we provided just 1 quarter ago, we have increased our fiscal 2027 revenue outlook by approximately $500 million and our fiscal 2028 outlook by approximately $1.5 billion. The center of this higher outlook is our data center business, where growth continues to both accelerate and broaden.

We've increased our forecast for data center revenue growth in fiscal 2027 from our prior expectation of 50% to approximately 60%, and we see potential for further acceleration in fiscal 2028. Within connectivity, the [indiscernible] strength established in areas such as [indiscernible], while also seeing significant growth across broadband analog TIAs and drivers, scale across DCI modules and scale-out switching. Each of these 3 businesses is on or ahead of the trajectory towards the $1 billion annualized revenue run rate we highlighted last quarter.

Scale-up opportunity remains massive and is still largely ahead of us. Marvell is ideally positioned for the transition toward NPO and CPO optical interconnects as well as the adoption of purpose-built scale-up switches. Our custom business, including both XPU and XPU attach is also on a strong growth trajectory, both near term and longer term. The 8-K we filed last week highlights the expanded scope of our relationship with a key hyperscaler and reinforces our confidence in the increasingly important role custom silicon will play in the ongoing AI infrastructure build-out. We look forward to sharing more about the longer-term growth opportunities we see for Marvell at our Investor Day on October 6 in New York City, and we hope to see many of you there.

Now with that, I'll turn the call over to Dan for more details on our recent results and outlook.

Daniel Durn

Thank you, Matt. Good afternoon, everyone. Before I get into the financials, since this is my first earnings call as Marvell's CFO, I want to spend a moment on 3 things: why I joined Marvell? What I hope to accomplish as CFO? And what I believe that should ultimately mean for our stockholders? First, I joined Marvell because I believe the company has incredible growth potential. I believe Marvell is at a unique inflection point with a broad and diverse set of capabilities that have been intentionally built and are ideally aligned to enable the future of AI infrastructure.

Second, my goal as CFO is to help Marvell scale efficiently and effectively to capture what I expect to be a tremendous opportunity. The company has a strong history of disciplined execution and maintaining that rigor will remain paramount as we strategically allocate capital to fuel our growth.

And third, I expect that growth to translate into significant value creation for our stockholders. I recognize that the growth opportunity ahead must be matched by its quality, and I plan to remain intensely focused on expanding operating margins, growing cash flow and driving attractive stockholder returns.

With that, let me turn to our financial results for the second quarter of fiscal 2027. Revenue was $2.739 billion, growing 37% year-over-year and 13% sequentially. Data Center was our largest end market, contributing 79% of total revenue and growing 46% year-over-year. GAAP gross margin was 53.1%. Non-GAAP gross margin was 58.9%, slightly above the midpoint of our guidance. Moving to operating expenses. GAAP operating expenses were $996 million, including stock-based comp amortization of acquired intangible assets, restructuring costs and acquisition-related costs. Non-GAAP operating expenses were $611 million, slightly above our guidance.

GAAP operating margin was 16.8%, while non-GAAP operating margin was 36.6%. Non-GAAP operating margin expanded 180 basis points year-over-year and 160 basis points sequentially, demonstrating the significant operating leverage in our model. For the second quarter, GAAP earnings per diluted share were $0.33. Non-GAAP earnings per diluted share were $0.94 [indiscernible] above the midpoint of our guidance and up 40% year-over-year.

Now turning to cash flow and the balance sheet. Cash flow from operations was $606 million in the second quarter, down slightly quarter-over-quarter, primarily reflecting the higher capacity prepayments to suppliers in support of Marvell's future growth. Inventory at the end of the second quarter was $1.36 billion, down just slightly from the prior quarter. During the quarter, we purchased $200 million of our common stock through our ongoing capital return program and returned $54 million to stockholders through cash dividends.

At the end of the second quarter, total debt was $4.96 billion, with gross debt-to-EBITDA ratio of 1.32x and net debt-to-EBITDA ratio of 0.27x. Turning to our guidance for the third quarter of fiscal 2027. We are forecasting revenue of $3.15 billion, plus or minus 5%. We expect GAAP gross margin to be between 52.9% and 53.9%. We expect our non-GAAP gross margin to be between 57.5% and 58.5%. Revenue levels and product mix remain key determinants of gross margin in any given quarter with the forecasted acceleration of our custom business creating the sequential headroom in the fiscal third quarter -- the sequential headwind in the fiscal third quarter. We currently expect to maintain gross margin in this range in the fourth fiscal quarter.

We project GAAP operating expense of approximately $1.015 billion in fiscal Q3 and non-GAAP operating expenses of approximately $655 million. We expect GAAP other income and expense, including interest on our debt to be an expense of approximately $86 million. On a non-GAAP basis, we expect other income and expense, including interest on our debt to be an expense of approximately $36 million. We expect a non-GAAP tax rate of 11%, looking ahead, given the significant increase in our revenue and earnings outlook, we expect non-GAAP tax rate of approximately 13% in fiscal 2028.

We expect basic weighted average shares outstanding of approximately $900 million and diluted weighted average shares outstanding of approximately $921 million, both roughly flat with the second quarter. We anticipate GAAP earnings per diluted share of $0.48 to $0.58. We expect non-GAAP earnings per diluted share of $1.05 to $1.15. As we look ahead, we intend to continue investing in our business to drive substantial revenue growth in the years ahead while delivering operating leverage.

For fiscal 2027, we expect non-GAAP operating expenses of approximately $2.55 billion, slightly above our prior expectation of $2.45 billion, reflecting the significantly larger revenue opportunity we now see. Importantly, we expect revenue to continue growing substantially faster than operating expenses. As a result, we expect significant operating leverage with non-GAAP operating margin likely to enter our 38% to 40% long-term target range in Q4 of this fiscal year.

Looking ahead to fiscal 2028. We currently expect non-GAAP operating expenses to grow at roughly half the rate of revenue growth in percentage terms. This reflects continued investment against an expanding opportunity set while yielding continued operating leverage to achieve the upper end of our target non-GAAP operating model of 38% to 40% as we progress through the year. Moving to cash usage. Based on the design wins we've secured and our confidence in the sustained customer demand, we're aggressively securing additional capacity to support our growth. We remain on pace to make approximately $1 billion of capacity prepayments and to suppliers in fiscal 2027.

Consistent with the guidance we provided last quarter. As a reminder, these prepayments will be applied against future material purchases and will be funded through our strong balance sheet, and robust operating cash flow. In parallel, we intend to continue repurchasing shares to manage dilution. Now let me come back to where I started. I joined Marvell because I believe the company has an exceptional opportunity ahead of it.

Having now stepped into the CFO role, I'm even more convinced of the incredible strength of our technology portfolio our deep customer relationships and the substantial long-term growth potential. Our job from here is to execute with discipline as we efficiently scale the company to capture that opportunity. and ensure that our growth translates into expanding margins, strong cash flow and compelling returns for our stockholders. I'm honored to be Marvell's CFO at such an important point in the company's evolution and I look forward to working with Matt and the rest of the team to deliver on the incredible opportunity we see ahead.

With that, ready to start Q&A. Operator, please open the line and announce the instructions.

Operator

[Operator Instructions] And our first question Tom O'Malley with Barclays.

Thomas O'Malley

Appreciate it, good results. I wanted to ask first on the warrants with Google, obviously, a very robust relationship over a multiyear period of time. I love you could give a color on what's contributing to that revenue? Obviously, you have an inference accelerator, but you talked a lot about XPU attach, which you guys did kind of start the industry standard on in terms of defining. So maybe what XPU attach are you excited about there? And like any sort of color you can give us on the percentage of contribution between the inference accelerator and some of the attach that you're seeing there?

Matthew Murphy

Yes. Great. Thanks, Tom. Good to hear from you. So I think you captured it well. It's a very exciting time for Marvell. This engagement and warrant is significant. You can see that in the scale of the opportunity we have. And as you pointed out, a couple of things I would just validate. The first is it's very broad-based. It's a number of products and product lines, which is very exciting. It includes inference accelerators as you mentioned, also storage controllers, nicks, memory interface controllers, near-memory compute, a whole bunch of different products and you're right, we did define the -- what we call the XPU attach category a couple of years back, and actually we gave quite a detailed view of that in our June 2025 custom silicon event.

And I think all of our projections to date have been under called, meaning that opportunity continues to get more and more significant. So, it's a broad set of products, Tom, that are covered here, and it's very broad-based in terms of where we can engage and where we're going to contribute. And I wouldn't call out any one of them at the moment specifically, but all of them in total, you can see represent -- if you look at the total envelope of the opportunity, it's just massive for Marvell and game changing at the sort of peak performance of what could be achieved now over the next 6, 6.5 years.

Operator

And your next question comes from Harlan Sur with JPMorgan.

Harlan Sur

Welcome to the team, Dan and Ross. Matt, given the Google commercial program you signed back in the July, a $120 billion in cumulative revenue over 6 years, if you hit all of your milestones, right, that puts you if you just annualize it at about $18.5 billion per year in revenues just in Google XPU attached custom ASICs, right. I actually thought that, that would start to show up in fiscal '28 or your calendar '27, but given your guidance, it looks like custom is still going to be around $5 billion to $6 billion in calendar '27.

But maybe some of the big programs associated with this commercial agreement more back-end loaded. So maybe the better question is, the team has previously targeted $10 billion to $11 billion in custom revenues in fiscal '29 or [indiscernible] what does that number look like now based on the Google commercial agreement and some of the new programs associated with the agreement?

Matthew Murphy

Yes. Yes. Great question, Harlan. Thank you. So first, you're right. If you look out to next year, we have comprehended already revenue that would come as part of this warrant in our numbers. Now that's because some of these programs, obviously, are already in flight or have already started, but the programs ahead of us that are either in execution or just starting production, those will contribute much more significantly in fiscal '29 and for next year, all I've said, by the way, is that custom is going to over double. So I haven't capped it. I haven't sized it exactly, but it is going to overdouble next year.

And then -- so then you look out to fiscal '29 where in our last call, I talked about a $10 billion kind of plus number for custom in fiscal '29, which isn't a new number, by the way. That was something that we outlined back in June '25 and even back in April '24, we were talking about custom revenues in the $8 billion to $10 billion range. So we've had a consistent view of what that can look like. This opportunity with this customer greatly -- greatly increases the revenue opportunity for us in custom with some of that potentially starting in '29. I'm not going to size it on the call here today, but you should assume with our Investor Day coming up Harlan, the Marvell team, we always do a thorough job in our investor -- in our analyst/investor days.

And so you should expect a very robust detailed review of how we step through the revenue, not just through fiscal '29, but really out until the end of the decade. And so we're excited to present that to everybody, which will then really help us contextualize the scale of the warrant that we just signed, along with all the other exciting programs we have in custom. So if you don't mind, hold tight on that one, there will be more to come. But clearly, there's a lot of upside bias in those numbers in fiscal '29 and beyond in custom.

Harlan Sur

Yes, absolutely. Absolutely. Looking forward to it.

Operator

Your next question comes from Vivek Arya with Bank of America Securities.

Vivek Arya

Best wishes to Dan and Ross. Matt, I was hoping you could give us an update on the other large XPU program that you're planning to start in the next year. I think the assumptions there are quite modest for next year, $600 million or $700 million or so and I was hoping you could give us an update on that?

And then what is the opportunity for that program over time? Because that hyperscaler is just getting started on their XPU type like every one of their competitors, I imagine they want to make that XPU program much larger than what it is. So give us maybe an update on what the progress is? Can this be a really meaningful program right for you over time?

Matthew Murphy

Yes. Thanks, Vivek. So yes, we're very encouraged by our custom setup for next year on the double plus. That new program is clearly part of that. And we've been judging that, I think, in a very conservative and practical manner as we've made progress, but we make -- we continue to make progress every quarter, not only on design execution, but also supply commercials and figuring out the sort of the envelope of that opportunity. And that's only gotten better.

So without quantifying it, exactly what I would say is from the last call, if you look at the $1.5 billion raise we're doing for next year, just at a high level, it's broad-based. So some of that is from custom next year. There's also actually get another question on this one, probably the most meaningful part of the next year raise is actually from the scale-up optics switching in other areas, but customs definitely part of it, Vivek.

So even when we size that opportunity, which goes back to a couple of years, we've always said that, that is probably one of Marvell's largest revenue opportunity opportunities we have, and that's still the case. So we're very encouraged by the project -- the prospects of this project. And there'll be more to come, but it certainly is tracking, and we feel very good about next year and that layering in the year after and beyond.

Operator

Your next question comes from Aaron Rakers with Wells Fargo.

Aaron Rakers

And welcome Dan and Ross. Maybe kind of building on that last question a little bit. Matt, if you can maybe talk a little bit about what you're architecturally seeing in CXL and how that's evolving, obviously, with the [indiscernible] product and it seems to be a broadening ecosystem around that. And then similar on an architectural perspective, any update on -- you just touched on scale-up optics, the Celestial AI numbers that you've previously outlined, how have those progressed relative to your initial targets?

Matthew Murphy

Sure. So let me start with CXL and then we'll talk about scale-up optics. So with respect to CXL, I mean this investment we made organically over the years has really evolved and is turning out to be a home run for a couple of reasons. As you might recall, this started off as a server-centric traditional compute-centric memory architecture it turns out all that investment we've made is just ideal for memory expansion and inferencing.

And so we're seeing this technology now getting deployed at multiple hyperscalers with varying architectures by the way, in extremely high volumes. One is just the demand of inferencing require it. The other is what we're seeing is as a result of the scarcity that's out there in memory, customers are modifying and adjusting their plans to actually use more of this type of technology. So this has continued to have a strong upward bias. We've actually secured additional design wins in this area in the last couple of quarters. we're going to outline all of that at the Investor Day.

But that opportunity, Aaron, turns out to be massive. And I think it's a case where the pivot we made about 5 years ago where we really put the pedal to the metal on organic internal investment on a number of new technologies. This is one that's really bearing fruit. So very exciting, and you'll -- you should expect to see a very comprehensive memory expansion section in our Investor Day presentation.

On scale-up optics, this is turning out to be, I think, one of the most exciting areas for us. Last quarter, the way we framed it was we had the Celestial AI CPO photonic fabric solution, which for reference, was about $150 million for next year. We had actually said that overall scale-up optics inclusive of that was about $300 million as a category, that includes NPO, by the way. And as I said in my prepared remarks, all of that together, scale-up optics is accelerating even further from the number I gave you last quarter.

And so, a significant driver of the $1.5 billion raise we have for next year, Aaron, is coming from that area. And it's not just a one-trick pony. I think that's the thing I want to stress to you is the CPO is absolutely a key part of it. It will be for next year and certainly the year after. But if I look at the opportunity set that's happening now at a much faster pace in parallel with the CPO programs we have is for NPO.

And in NPO, we participate through a wide variety of design wins we have and partners, some of those are with our broadband analog products in silicon germanium, both TIAs and drivers, which we have a significant market share and content and also on our own organically developed solution, which call our light engine, which we've been in development on for several years, and we've shown it off at OFC for several years.

So all of that is a category, Aaron, is really how we think about the business. Certainly, Celestial and photonic fabric is a key part of that. But what I want to stress is that this is not an or it's an and. And it's a little bit analogous to when people thought, well, there's 800-gig DSPs and then they're going to move to 1.6T and it's all going to just cut over. That's not how this works anymore. Our customers have varying architectures that they're pursuing. They have a number of different solutions that they're going to implement simultaneously.

So we see this as an and, not an or. And I think the magnitude of our scale-up optics opportunity next year is much larger than we thought just a quarter ago. The year after is going to be, I think, way larger than I sort of could have ever comprehended back even last year when we were looking at Celestial. And if you look at the solution we're providing, which is here on my last comment, we are pedal to the metal on our switching road map and coupling those optical solutions with our switches, both UAL and ESUN as well as being able to seamlessly integrate and architect with our customers on the XPU/GPU side.

So all of that together and having the end-to-end length that we're able to provide in the entire solution and the scale in manufacturing, R&D, the diversity of all the technologies we have in terms of packaging, modulator technologies, silicon photonics, a decade of experience there. It's a massive opportunity and Marvell is very uniquely positioned to participate in it. And we couldn't be happier with the Celestial team. They've integrated well. They're a key part of the program. But this thing is going to be a much bigger overall business for Marvell.

Operator

Your next question comes from Joe Moore with Morgan Stanley.

Joseph Moore

On the same lines, can you talk about your progress in copper scale up, and you mentioned the sort of success across the 3 protocols. How do you think about copper scale-up as it transitions to optical? Is it those initial successes foundational to what you do in optical? Or are those sort of 2 separate decisions?

Matthew Murphy

Yes. Thanks, Joe. I think you're talking about from a switching perspective, right, as we implement those electrical/copper solutions and then how does that success translate when we move to optical? So a couple of things. One is, on the Ethernet side, we've seen great traction and success with our Teralinx architecture, which came from the acquisition we did of a company called Innovium back in 2021. That business has a significant head of steam on it. We've expanded the customer base. We're driving significant revenue there, and we've proven to the market that we can deliver these solutions in volume.

So that translates itself, Joe, very seamlessly to scale out. And in fact, we're seeing even at 100 T, our scale-out solutions be applicable to scale up as well. So we're driving that independently of the optics or not. Same thing on UAL. We're aggressively investing in UAL switching that product development is coming along nicely, and we certainly can intercept any electrical or copper-based solution there. What customers really want to see though, and that's why I'm spending time on it, is, that's all great. And at this point, they trust us that we can deliver these very complex switching products because we've done it already, which by the way, is no small feat.

There's been a lot of companies that have tried and failed to do these [indiscernible] size very complex Ethernet switches, and it's only getting harder with the SerDes performance and the speeds. But the next level of that, Joe, is that you then need to show a compelling clear road map that you can execute with credibility on the optics side, both NPO and then all the way to full integration on CPO and so our additions while certainly there's opportunities in the next few years that are going to come and we're going to have on the copper side, I think why we really win long term is because we convince our customers we can execute on the full end-to-end, which really is what our customers are looking for at the end of the day, especially in scale up, they need to book end the link, and they need to trust it on both sides of it and everything in between. They've got a partner that can handle that job.

And point solutions at this juncture, we believe are not going to get it done. You really have to have the end-to-end portfolio. And that's what we're hearing from our customers, by the way. So the existing silicon will do fine, and it will do great, but even more excited about the optics attached once that gets going over the next few years because it just effectively almost doubles the [indiscernible] of the switching in terms of the attach you can get Thanks.

Operator

Your next question comes from Ben Reitzes with Melius Research.

Benjamin Reitzes

Matt, so I want to go back to the Google deal, I think people are just wrestling with this. And I realize that you have a great Analyst Day coming. So I -- don't shoot me. But for FY, if you look at the $120 billion over the course of the years, that's about $18 billion a year and at the $120 billion divided by [indiscernible]. And that's like adding an FY '28, Marvell every year. And I realize that we don't know how much is incremental versus what's already in your guide.

But, are we talking like FY '29 and FY '30 big step-ups as a result of this incrementally because these numbers are huge. And I know you're going to tell us at the Analyst Day, but I think we're all just trying to figure out how high to go in those out years? And is it at the comparable margin? And -- just any more color if you're in our shoes, how we should handle it into Analyst Day or just wait, that would be great.

Matthew Murphy

Yes, Ben, thanks. And by the way, fully valid question. I mean, when you look at the scale of this, your math is not wrong. And what you can conclude from what I'm saying is, because most of this is comprehended already in next year, the big impact would be in '29 and beyond. So it's -- if you took the full performance and the full opportunity, then you're right, it's just -- it's just a monster number.

What I would say -- and so we do need the Analyst Day, though, I think you guys understand to contextualize it and probably show some ranges of outcomes. But you should assume in that time frame that on the custom side, these numbers would be a lot larger than -- overall custom than anybody has been modeling so far. I mean, I think there's been doubt for years that we could even do the $8 billion to $10 billion. This should give, I think, investors comfort that we secured a pretty big set of programs, not just here, but across the broad range of our customer set.

On the margins, just in general, this is a custom business, and we've got a financial model for custom. We've got a financial model for our standard and merchant products. It would be in line with that, but it's significant. And I'm not able to quantify it today for you, but you should assume starting in FY '29 and beyond whatever you've modeled previously prior to the warrant for custom numbers definitely goes higher.

How big we'll be happy to show the range of outcomes. But in the context of where Marvell has been, I mean, we were an $8 billion company last year. We just took everybody to $12 billion this year. We haven't even finished that year and $18 next year. And the rough math you do at full performance provides an incredible step up to the scale of the company. If all those programs that come to fruition at their max performance, which -- but I think what this is helpful to show is just the scale of the opportunity in general that Marvell is participating in.

Because we have significant engagements across the customer base. This one is a little unique because of the warrants that given the magnitude and the equity side, we needed to disclose it. But I think you've seen over the last couple of years, we've entered into very strategic partnerships with other big players in the AI market. And some of that we've done publicly and some of that we're doing just on our own.

So I think one takeaway I would have right now is just a significant validation of where Marvell sits in the market today, and we're very honored to be a part of the ecosystem, the TPU ecosystem and we'll see how it plays out. But I hope that's enough for now. I want to save a little bit of firepower from Investor Day.

Benjamin Reitzes

And I appreciate that and congrats on that deal. I'll see you soon.

Operator

Your next question comes from CJ Muse with Cantor Fitzgerald.

Christopher Muse

I think one of the more interesting takeaways from NVIDIA's report last night is Jensen's view of the fungibility of compute, networking and memory to deliver performance in the AI data center in a cost-constrained world. And considering your vast experience and expertise across all these 3 areas, how is this backdrop, particularly in light of where memory pricing is today support your new design wins? And is it really focused on the memory controllers, custom HBM to increase short line? Or is it also driving strength on the XPU side as well? Would love to hear your thoughts there.

Matthew Murphy

Yes. Thanks, CJ. No -- first of all, I see the same thing. I'm very mind-melded with what they talked about. I think at the broadest level to start, it definitely -- this fungibility requirement, given the dynamic nature of the market and some of the constraints we're seeing. Some are supply constraints, some are power constraints, some are architectural constraints, having our ability to very quickly execute custom and semi-custom designs or modify our products or adapt to shifting architectures at Marvell. It's a key capability, the flexibility that the company has because we've built a business around being able to customize quickly, and it's been in our DNA even from 2021 Investor Day.

We talked about how every hop in the network at some point was going to get some level of customization, it wasn't going to just be the accelerator. And here we are 5 years later, and that's playing out. CXL and memory expansion is one example where very quickly, customers are adapting, but there's also other opportunities where to optimize for inference as an example, and this is one of the things that was in the warrant, but it's a trend in the market. AI inference accelerators, that's another trend in the market you see. So I think companies are all looking at how they can maximize their performance, how they can maximize the cost and performance relative to the tokens generated.

And we're in a monetization era. So this stuff really matters. It went from training very, very quickly to inference and to companies now monetizing this. And so we're seeing a big pickup in activity. And it's not just on the custom side. I mean this discussion we had earlier about CPO and NPO and optics and copper, that's all moving very, very fast because, again, companies are architecting at a speed we really haven't seen to make sure that they have the most competitive solutions.

So Marvell fits right in there, CJ, on a number of fronts. It's not just "our custom" revenue, but it's the solutions underneath we provide including our networking and our storage and storage and memory in our compute .

Operator

Your next question comes from Jim Schneider with Goldman Sachs.

James Schneider

was wondering if you could maybe just give us a little bit of update in terms of the growth rate for productivity, you expect -- I think, Matt, you outlined those growth rates for both this fiscal year, next fiscal year last quarter. Can you maybe just give us an update on where those are landing, given the incremental strength. It sounds like most of that's being driven by that subsegment right now?

Matthew Murphy

Well, yes, thanks, Jim. As I said, the way to think about it is the $500 million raise for this year and the [indiscernible] for next year is broad-based connectivity clearly is a driver, and I say connectivity at a broader level because underneath that, we spent time talking about scale-up optics. That's one, which is our NPO and CPO and [indiscernible] products. But it's also just the transceiver market for scale-out and optical DSPs going into that segment, that's upsized versus the prior growth rates we talked about. So that's in the $500 million and the $1.5 billion, that business continues to be absolutely on fire, and we're executing well there.

And then within connectivity as well as a broader segment, you also have our switching which is biasing higher this year, but also from our last quarter update, again next year. So you really have those three, plus, I'd say, custom that are all floating up versus our prior expectations. But Jim, we're really seeing broad-based strength. There's -- I think it's across almost all of our product lines when we look out, but those are a few of the ones I'd highlight. And you're right. The connectivity is a bigger bucket, it is probably the largest driver net-net of the $1.5 billion raise. If I just look at all those pieces I mentioned underneath, scale out, scale up and switching.

Operator

And our last question for today comes from Quinn Bolton with Needham & Company.

Quinn Bolton

I guess, Matt, just wanted to come back to the sort of the near-term guidance. Within data center, I think you guided up greater than 20% sequentially. But with margins coming down 90 basis points quarter-on-quarter, is custom the fastest-growing segment within data center that drives that lower margin? Or is there some other mix going on within data center kind of driving the lower 90 -- or sorry, the 90 basis point lower guidance for the October quarter?

Matthew Murphy

Yes, sure. Thanks. Let me -- I'll tee it up, but I'm going to have Dan comment because you guys are sparing them too much on this call. I mean, he's the new CFO. You're supposed to be asking him questions and you're asking me all the questions. So joking aside, you can clearly see it in the numbers. I mean, stronger Q3 guide implied a much stronger Q4 guide. We did say custom was ramped meaningfully in the second half. I said that, Dan said that. But Dan, maybe why don't you make a few comments on the margins and how you see that playing out and also through next year as well as we grow.

Daniel Durn

Sure. So just jumping in, we clearly see the performance of the company doing well. You see the acceleration in Q2, followed with acceleration in Q3, follow with acceleration in Q4. Part of the acceleration story in Q3, as you point out, mix is the primary driver. We've got a strong ramp in custom. And so you can see that play out in the profiling of the margins. Not a surprise. We've been signaling that custom ramp for quite some time as we window into Q4, that strength, we're signaling a bigger step-up, much larger step-up from a revenue standpoint, if you roll in the $12 billion annual target.

And that step up in Q4 is broad-based. You see it not only in custom, you see it in connectivity. You see it in a rebound in comm and other. So broad-based performance net it all out. We see gross -- Q4 gross margins in the same target range as Q3. If we use the back half of the year as a jumping off point and think about how that looks going into FY '28. That broad-based strength continues. We take the monounsaturated iterations and we push that forward into next year, and it's cross custom's, it's across connectivity all parts of the business are higher. They're up.

Gross margin depends on ultimate mix. My preliminary view is, is gross margins next year are going to be in a similar range, same range as we're exiting this year. So back half of this year, same range for FY '28. So we feel good about the performance of the company. We're going to drive growth at this company. We're going to do it with discipline, and we're going to deliver strong margins.

But when we think about that margin profile, operating margin still has significant leverage embedded in it. You're going to see it up in Q3. You're going to see us entering our target long-term model range, 38% to 40% exiting this year. You'll see us achieve the high end of that range as we progress through FY '28, and we're going to reset that long-term target model here in the coming weeks at the Analyst Day. So we feel good about the performance of the company and the broad-based strike.

Operator

Thank you. And ladies and gentlemen, thank you all for your participation. This does conclude today's teleconference. You may disconnect and have a wonderful day.