邁凌科技 (MXL) 2026 年第三季財報電話會議：營收成長 72%，上調光學業務展望

重點摘要

• 受惠於光學 PAM4 產品加速採用及基礎設施業務成長，2026 年第三季營收年增 72%、季增 29% 至 2.181 億美元。
• 基礎設施營收達到約 1.25 億美元，年增 210%，且超過 MaxLinear 其他所有營收類別的總和。
• Non-GAAP 毛利率為 60%，Non-GAAP 營業利益率擴大至 29.5%。Non-GAAP 稀釋後 EPS 為 0.62 美元。
• MaxLinear 上調 2026 年光學資料中心營收展望至 2.5 億至 2.7 億美元，理由是需求更加強勁且客戶參與度提高。
• 管理層預測 2026 年第四季營收為 2.4 億至 2.6 億美元，受光學資料中心互連需求推動，基礎設施業務預期將強勁成長。
• 公司預期 1.6-terabit Rushmore 產品將於 2027 年上半年開始貢獻首波營收，下半年將出現更強勁的量產成長。在手訂單已延伸至 2027 全年及 2028 年。

關鍵財務數據

業務與營運表現

AI 資料中心連接性仍是 MaxLinear 的主要成長引擎。公司的 Keystone 系列為 5 奈米 CMOS 單通道 100-gigabit PAM4 DSP 產品群，持續在多家超大型雲端業者進行大規模量產，並支援 800-gigabit 光學部署。

Rushmore 是 MaxLinear 用於 1.6-terabit 架構的單通道 200-gigabit PAM4 平台，目前正進行客戶認證。管理層預期 2027 年上半年將貢獻首波營收，下半年量產力道更強，並預計在 2028 年和 2029 年帶來更大的成長貢獻。公司表示，在過渡期間，800-gigabit Keystone 產品應會持續成長。

客戶群廣泛涵蓋主要光收發模組廠商，合作對象遍及亞洲與美國。管理層表示，美國資料中心在採用 1.6-terabit 技術方面進展更快，應會成為日益重要的需求驅動力。

MaxLinear 也正在跨足光學 DSP 以外的領域。Washington 為其單通道 200-gigabit TIA 平台，支援基於 DSP 的光學與線性光學架構。管理層預期 Washington 將於 2027 年貢獻首波營收，並於 2028 年實現更強勁的量產。Big Sky 則瞄準未來的單通道 400-gigabit 及 3.2-terabit 光學與電學互連。

AI Rack Commander 產品組合將 MaxLinear 的角色延伸至機架級 AI 基礎設施的遙測、連接性、電源管理與保護。用於超大型資料中心的 XGS-PON 控制平面設計獲獎項目已完成認證，預計於 2027 年進行首次部署，並有已獲得的在手訂單提供支撐。

管理層預期公司的儲存加速器平台 Panther 在 2026 年的營收將倍增，並有潛力在 2027 年實現另一次顯著成長。

受惠於北美與歐洲一線服務供應商部署光纖 PON 和 WiFi 7 閘道器，寬頻與連接性業務在第三季均實現季成長。在電信支出及 PON 市占率提升的支撐下，管理層預期寬頻業務將於 2027 年實現成長。

管理層財務展望

按終端市場劃分，管理層預期受資料中心光學互連推動，基礎設施業務將實現強勁成長。寬頻與連接性業務預計將呈季節性下滑，而工業多市場營收則預計增加。

公司亦將 2026 全年光學資料中心營收展望上調至 2.5 億至 2.7 億美元。管理層預計 2027 年營收將再次顯著成長，但未提供正式的全年展望。

風險與觀察重點

• 管理層指出晶圓代工產能是光學 DSP 供應的主要限制因素，並強調新晶圓廠產能需要多年的投資。
• 晶圓預付款導致第三季營業現金流為負數。隨著預估營收與產能需求增加，公司預計預付款將進一步上升。
• MaxLinear 將其循環信貸額度增加 7,000 萬美元至 2 億美元，為供應承諾與成長機會提供額外的流動性。
• 2027 年上半年 Rushmore 出貨的確切規模仍存在不確定性，但管理層表示，預計初始設計獲獎產品將在該期間開始產生營收。
• 同時支援多個 DSP、TIA 及電學互連認證計畫造成了內部資源限制，這需要 MaxLinear 對合作項目進行排程與評估優先順序。
• 寬頻與連接性業務在 2026 年第四季面臨季節性疲軟。

分析師問答重點

光學業務成長與能見度：管理層未提供正式的 2027 年營收展望，但預期 MaxLinear 的光學 DSP 業務成長率將超越市場整體表現。光學產品交期據述約為 26 至 28 週，而客戶正提前更久下單以確保供應。在手訂單已延伸至 2027 全年及 2028 年。

客戶多元化：MaxLinear 表示，光學營收分布於各大模組廠商，且參與 1.6-terabit 計畫的業者數量持續成長。預計美國超大型雲端業者將比中國同業更早採用 1.6-terabit 技術。

市占率：管理層表示，MaxLinear 在光學 DSP 的市占率持續擴大，並設定目標在未來兩年內達到十幾個百分比的市占率。管理層指出，由於市場需求與出貨預測持續變化，目前市占率估算較難精確衡量。

毛利率：隨著基礎設施占營收比重提升，公司預期毛利率將呈上升趨勢。管理層表示，已將部分生態系統成本增加轉嫁給客戶，同時公司的成本結構仍具支撐力。

供應紀律：執行長 Kishore Seendripu 表示，未發現重複下單的跡象。他將更長的訂單能見度歸因於製造產能限制、嚴格的供應鏈管理，以及客戶為確保取得認證零組件所做的努力。

未來光學架構：管理層預期可插拔光學 (pluggable optics)、近封裝光學 (near-packaged optics) 與共同封裝光學 (co-packaged optics) 將並存。MaxLinear 認為 DSP 技術在更高速率下仍具重要性，並表示其產品組合涵蓋用於多種光學架構的 DSP、驅動晶片與 TIA。

完整法說會逐字稿

完整財報電話會議逐字稿

管理層陳述

Operator

Greetings, and welcome to the MaxLinear Q3 2026 Earnings Conference Call. [Operator Instructions]. As a reminder, this conference is being recorded. It is now my pleasure to introduce your host, Leslie Green of Investor Relations. Thank you. You may begin.

Leslie Green

Thank you, Shamali. Good afternoon, everyone, and thank you for joining us on today's conference call to discuss MaxLinear's third quarter 2026 financial results. Today's call is being hosted by Dr. Kishore Seendripu, CEO; and Steve Litchfield, Chief Financial Officer and Chief Corporate Strategy Officer.

After our prepared comments, we will take your questions. Our comments today include forward-looking statements within the meaning of applicable securities laws including statements regarding our guidance for the fourth quarter of 2026, including revenue, GAAP and non-GAAP gross margin, GAAP and non-GAAP operating expenses, GAAP and non-GAAP interest and other expense GAAP and non-GAAP income taxes and GAAP and non-GAAP diluted share count.

In addition, we will make forward-looking statements relating to trends, or execution of our business plan and potential growth and uncertainties in various product and geographic markets, including, without limitation, statements concerning the future financial and operating results, opportunities for revenue and market share across that market share gains across our target markets, new products, including the timing of production and launches of such products, demand for the adoption of certain technologies and our total addressable market.

These forward-looking statements involve substantial risks and uncertainties, including risks outlined in the Risk Factors section of our recent SEC filings, including our most recent annual report on Form 10-K and our Form 10-Q for the quarter ended September 30, 2026, which we filed today. Any forward-looking statements are made as of today, and MaxLinear has no obligation to update or revise any forward-looking statements. The third quarter 2026 earnings release is available in the Investor Relations website at maxlinear.com. In addition, we report certain historical financial metrics, including, but not limited to, gross margin, income or loss from operations, operating expenses interest and other expense and income tax on both the GAAP and non-GAAP basis.

We encourage investors to review the detailed reconciliation of our GAAP and non-GAAP presentations and the press release available on our website. We do not provide a reconciliation of non-GAAP guidance for future periods because of the inherent uncertainty associated with our ability to project certain future changes, including stock-based compensation and its related tax effects as well as potential impairments. Non-GAAP financial measures discussed today are not meant to be considered in isolation or as a substitute for comparable GAAP financial measures. We are providing this information because management believes it is useful to investors as it reflects how management measures our business.

Lastly, this call is also being webcast, and the replay will be available on our website for 2 weeks. And now let me turn the call over to Kishore Seendripu, CEO of MaxLinear. Kishore?

Kishore Seendripu

Thank you, Leslie, and good afternoon, everyone. As the ongoing AI infrastructure build-out momentum continues, and as AI data centric clusters continue to scale to unprecedented size and bandwidth requirements, we believe MaxLinear is emerging as a key enabler of next-generation connectivity technologies. Demand for our optical data center connectivity and solutions continues to strengthen, and we believe MaxLinear is in the early innings of a multiyear AI-driven growth cycle. Against this backdrop, our data center business continues to outperform our expectations.

In Q3, overall revenue grew 72% year-on-year driven by accelerating adoption of our optical PAM4 products, expanding customer engagements and strong execution across our infrastructure portfolio. Infrastructure as a category grew 210% year-on-year and which is now larger than the rest of our revenue categories combined. Reflecting this momentum, we are raising our 2026 optical data server center revenue outlook to $250 million to $270 million. With strong visibility and bookings extending beyond 2027, we expect another significant step-up in revenue next year. Keystone, our 5-nanometer CMOS 100 gigabit per lane PAM4 CDs family continues to ramp rapidly in high-volume production across multiple hyperscalers. Its industry-leading power efficiency and performance position MaxLinear at the center of today's 800-gig optical infrastructure deployments.

Looking ahead, we are seeing accelerating customer activity around Rush mall-based 200 gigabit per lay in PAM4, 1.6 terabit architectures where qualification programs are advancing towards expected production deployments in 2027. Rushmore is building on Keystone's foundational success and is also expected to become a major multiyear growth driver. Big Sky, our family of next-generation 400-gigabit per lane, 3.2 terabit optical and electrical PAM for interconnect technologies will further capitalize growth as network as data center network speeds continue to accelerate beyond 200 gigabit per lane.

Beyond PAM4 3D's optics, we continue to broaden our AI infrastructure footprint across optical and electrical interconnects, [indiscernible], control plane connectivity, telemetry and power management. This strong expansion of content inside the AI data center stack positions MaxLinear to participate across multiple layers of future AI data center architecture. Already unapproved our 1.6 terabit electrical interconnect product family addresses 1.6-terabit active electrical cable and onward retimer applications for AI scale-up. Washington, our 200 gigabit per day in TIA platform not only supports PAM4 DSP-based optics, but [indiscernible] linear pluggable optics consisting of NPO and CPO solutions. We expect initial Washington and unapproved revenues in 2027, followed by a stronger volume ramp in 2028.

We are also excited about our AI RAC Commander portfolio consisting of USB, PCIe UR connectivity bridges used for telemetry power management and protection solutions for rack scale AI infrastructure. Rack Commander expands our role beyond data movement and into the control plane of the AI data center. Customers require advanced telemetry monitoring, connectivity and power control capabilities to maximize performance and efficiency, which is critical for infrastructure management. We believe this represents a significant long-term growth opportunity while expanding our strategic footprint.

Additionally, our XGS PON hyperscale data center design win, which serves as the control activity network across multiple data centers has completed qualification. Supported by secured backlog, we expect first deployments of this interdata center control plane solution to start in 2027. Separately, Panther, our storage accelerator platform continues to gain momentum as customers seek more efficient AI infrastructure by significantly increasing effective storage utilization and unlocking KV cash capacity for agent AI applications, Panther and able to support larger AI inference workloads without increasing storage infrastructure spending. We expect Panther revenue to roughly double this year with potential for another meaningful increase in 2027.

While AI data center remains our primary growth driver, we expect edge AI-driven upgrades to also drive growth in '27 and '28 for our wireless infrastructure and broadband access and connectivity solutions. These include Sierra single-chip 5G RAN remote radio unit solution, our millimeter and microwave backhaul RF and modem SoCs and XGS PON fiber broadband home gateways. Both broadband and connectivity categories delivered sequential revenue growth in Q3, driven by large-scale deployments of fiber pawn and WiFi 7 gateways at major Tier 1 service providers in North America and Europe.

In closing, accelerating hyperscale deployments, increasing optical connectivity content, new product ramps and broader participation across the AI infrastructure stack give us confidence that MaxLinear is still in the early stages of a significant growth cycle. Our strategy is focused on strengthening our leadership in high-speed data center connectivity, expanding into adjacent infrastructure markets and increasing your content, share and strategic relevance with hyperscale customers with strong customer momentum, great backlog visibility and a differentiated technology portfolio, we believe MaxLinear is well positioned to deliver sustained growth and long-term shareholder value.

With that, let me turn the call over to Steve Litchfield, our Chief Financial Officer and Chief Corporate Strategy Officer. Steve?

Steven Litchfield

Thanks, Kishore. Total revenue for the third quarter was $218.1 million, up 29% and from $168.8 million in the previous quarter and up 72% from $126.5 million in the third quarter of 2025. The Infrastructure revenue for the third quarter of 2016 grew to approximately $125 million. Broadband revenue grew to approximately $53 million. Connectivity revenue was approximately $25 million. Our industrial multi-market revenue was approximately $15 million. GAAP and non-GAAP gross margin for the third quarter were 58.7% and 60% of revenue. The delta between GAAP and non-GAAP gross margin in the third quarter was primarily driven by $2.5 million of acquisition-related intangible asset amortization.

Third quarter GAAP operating expenses were $97.8 million, and non-GAAP operating expenses were $66.5 million. The delta between GAAP and non-GAAP operating expenses were primarily due to stock-based compensation and performance-based equity accruals of $28.8 million combined and acquisition-related costs and other costs of $2.3 million. With the strong growth in our business, coupled with continued spending discipline, we have delivered significant improvement in our GAAP and non-GAAP operating margin over the past 12 months. GAAP operating margin for Q3 2026 was 13.9% and non-GAAP operating margin in Q3 was 29.5% of net revenue.

GAAP and non-GAAP interest and other expense during the quarter was $2.1 million and $2 million. GAAP diluted EPS for Q3 2026 was $0.30 per share. Non-GAAP diluted EPS was $0.62 per share. In Q3, net cash flow was used in operating activities was approximately $13.6 million. We exited Q3 of 2026 with approximately $109.9 million in cash, cash equivalents and restricted cash. This prepayment for wafer supporting rising demand for our data center products for which we have committed order backlog throughout 2027. Our days sales in Q3 was approximately 31 days versus 28 days in the previous quarter. Our days of inventory was down in the quarter from 123 days to 111 days.

Finally, today, we announced that we have amended our existing credit agreement to increase the amounts available under the revolving credit facility by $70 million to $200 million. With this facility, our total available liquidity has much improved, giving us additional flexibility to address further customer demand and other growth opportunities. This concludes the discussion for our Q3 financial results. With that, let's turn to our guidance for Q4 of 2026. We currently expect revenue in the fourth quarter of 2026 to be between $240 million and $260 million. Looking at Q4 by end market, we expect strong growth in infrastructure driven by data center optical interconnects, we expect broadband and connectivity end markets to be down seasonally while industrial multi-market end market is expected to be up.

We expect fourth quarter GAAP gross margin to be approximately 58% to 61% and non-GAAP gross margin to be in the range of 59% and 62% of revenue. We expect Q4 2026 GAAP operating expenses to be in the range of $96 million to $101 million. We expect Q4 2026 non-GAAP operating expenses to be in the range of $65 million to $70 million. We expect both our Q4 GAAP and non-GAAP interest and other expense to be in the range of 5 million to $4.3 million. 1.3 -- I'm sorry, $1.5 million tax benefit on a GAAP basis and non-GAAP tax provision of approximately $1.5 million. We expect our Q4 GAAP and non-GAAP diluted share count to be approximately $99.7 million each.

In summary, our quarterly performance highlights the continued strength of our optical product portfolio and the accelerating momentum across several growth drivers within our infrastructure business. The progress and innovation we have achieved in this area have been transformative, and we believe we remain in the early innings of a multi market by expanding revenue and increasing operating leverage. We're excited about the opportunities ahead and remain confident in our ability to deliver sustainable long-term value for shareholders. With that, I'd like to open up the call for questions. Operator?

Operator

[Operator Instructions]. And our first question comes from the line of Tore Svanberg with Stifel.

分析師問答

Tore Svanberg

Kishore, Steve, congrats on the strong results. I guess my first question is on the PAM4 DSP side. You upped the number there for [indiscernible]. I mean based on my rough math, I think that's now -- it'll be running at about $100 million a quarter run rate. So as we think about '27, is EUR 400 million plus a good number? And how should we think about the mix next year between 800 gig and 1.6?

Steven Litchfield

Yes. Tore, this is Steve. So I don't think we're planning to give guidance in the out year today. But certainly, I do believe you've seen us with some very strong growth coming from our optical DSP business. I think we're actually doing a really great job on the market share standpoint. And so I think as you look into 2027, you're starting to see the momentum pick up, and I would expect to see us outgrow the market in 2027.

Tore Svanberg

Very good. And then as far as customer diversification, I noticed you didn't have any 10% customers this quarter. And I had a question specifically on your optical DSP business. How should we think about the composition there by amount of customers, module makers, geography, any color you could share with us on how diversified that business is becoming real time.

Kishore Seendripu

Thank you, Tore. I think when you go to the first order of diversificate for us is really exciting that now we are transiting from 800 gig fully established as a growth vehicle and market share gains. And we are in strong qual cycles on 1.6 terabit or Rushmore products, along with -- not just only in optical products, but we have a strong qual cycle on electrical products as well as the 1.6 terabit speeds. So that's the first layer of diversification. We will -- so we expect in the near-term Rushmore to start picking up revenue in 2027. And if you look to '28 and '29 is going to be a massive growth driver for us. In addition to Keystone at 800 gig still growing for us as well. So that's the first order of diverter.

The second part is that our revenues are quite broad-based. It's across all the major module makers, and we have a growing set of module makers are also engaged in 1.6 terabit beyond our 800-gig product lines. And the same applies on the data centers as well. Our penetration of geographically in Asia and the U.S. is increasingly now the momentum on the designing cycle is starting to shift towards 1.6 terabit I think the U.S. is in a more advanced phase of 1.6 terabit deployments. We naturally expect the U.S. to become an increasingly strong driver of value proposition in the optical module space as well.

Operator

Our next question comes from the line of Ananda Baruah with Loop Capital Markets.

Ananda Baruah

Congrats on the continued good progress. It's cool to see. I guess, two, if I could, Kishore, just sort of sticking there on the hyperscaler diversification and maybe just part of a partly clarification as well. How much -- you said in the past, 1.60 production ships is really a mid-27% dynamic, how much of the business between now and then and 800G, if any, is from U.S.-based hyperscalers? And it just even added then I have a quick follow-up.

Kishore Seendripu

So Ananda, we have never broken down that revenue in that sort of we never do that. However, it's pretty broad-based. Suffice 2C as 1.6 terabit revenue start ramping with initial revenues actually in early 27 and a stronger ramp in the second half. The U.S. data centers will start picking very important in both -- on a static snapshot basis and also on a trajectory basis, so because the U.S. is the one that's going to add up 1.6 terabit, a much earlier cycle than the Chinese con parts, right? So I think that what we take extreme satisfaction and join is that 1.6 traction is quite across the board substantially.

And the U.S. data since they're just beginning to start ramping 1.6 terabit outside of an NDA and a Google either order, right? So -- and our supply diversity across the various speed nodes across 800 gig and 1.6 terabit on the supply lines, including the fabrication process, really puts us in a unique position where within our company, we are or competing a gross product line to the same technology node. So in a sense, we should be able to win this race in terms of making a major provider of 1.6.

Ananda Baruah

That's good context. I appreciate it. That's helpful. And then -- the quick follow-up is you mentioned MPO, CPO. So is there a useful way for us to think about your SerDes participation, if even just kind of from like a paradigm approach, how involved do you expect to be in what ways you expect to be involved? The question has been coming up and it will be kind of relevant 12 months from now.

Kishore Seendripu

Okay. If you look forward, you see a future where all the Alphabet POs will be in coexistence with the PAM4 SerDes base, optical pluggables, right? And we have seen that the short line density and power consumption scenarios will drive a certain volume of the speed connections into to the optical-based what I call incarnations, right, be it or an LRO, right?

So if you look at the LRO, our DSP naturally lends itself into a full-blown participation. And we believe that in the U.S., LRO is a real option at the 200 gig per lane, and I have a major option in the 400 gig per lane as well. That's my particular view. And then if you look at CPO has gotten delayed because the challenge of the optics in a hybridized co-pack is configuration is a very huge problem, and therefore, I think NPO supplies what we were thinking about the advantages of maintenance, modularity, diversity of suppliers for the product as against being locked to the AC players on the switch side or the GPU side.

So I think it's a natural place where the ecosystem goes. So I think the NPA will be big. How big I cannot speak. But I would say on a snapshot basis in 2030, you should expect 60% to 70% pluggable, 20% NPL and 10% all various forms of it. That's my personal view, not a market view, but I think we'll be a major player in NPLs which will be the natural ideally suited for low-power applications versus an LPO in the future as the speeds increase.

Operator

Our next question comes from the line of Cody Acree with StoneX Financial.

Cody Grant Acree

Congrats on the progress. It's been a great performance. Kishore, maybe can you talk about your visibility into '27 if you're not going to talk yet about the scale of your optical ramp. Can you maybe just talk about your bookings or your lead time visibility for the next few quarters?

Kishore Seendripu

Steve, do you want to take a shot at that?

Steven Litchfield

Sure. Yes, Cody. Look, we have great visibility. As I'm sure you've heard from [indiscernible], we're booking products throughout '27 into 2028 [indiscernible] for optical product or 26, 28 weeks. Customers are certainly coming in and placing orders out much further than that, just kind of given they want to secure supply. You've seen a lot of our upsides that we've delivered, I think you'll continue to see more of that growth in '27.

Cody Grant Acree

And Steve, maybe just continuing with you. Can you just bridge your next gross margin thoughts between mix, pricing, manufacturing costs and product transition over the next couple of quarters?

Steven Litchfield

Yes, sure. So gross margin, as you know, we've been making some nice progress kind of moved up from 58%. We delivered on 60% this quarter, our guidance at 6.5%. So I think we're we'rereally seeing some nice progress kind of despite some of the cost increases that we're seeing across our entire ecosystem here, I guess. Some of that we've been able to pass along to the customer. And I think our cost structure is actually just very good. So as the mix continues toward more infrastructure products, I think you will see our gross margin continue to trend up. Thanks, Cody.

Operator

Our next question comes from the line of Christopher Rolland with Susquehanna.

Christopher Rolland

Regarding DSP, last cycle, maybe the 400, 800 cycle was much more widely available transceivers had shorter lead times and people could wait for land power shell to be built before making what was kind of a last-minute order on optical. Do you guys view that as different now with longer lead times? And does this kind of perhaps create an effect where people are ordering ahead more significantly, perhaps even over ordering to -- they can't get their AI data center up without these transceivers. So does this become perhaps the golden screw that people are willing to purchase more than they might need just for the initial deployment?

Kishore Seendripu

So Chris I don't think so at all. Not at all, actually. And you have to keep this in mind that even in the supply chain, nobody is signing real long-term agreements. So you have strict discipline on lead time. There's a lot of manufacturing shortage. So it is in -- it is really naturally organizing the entire supply chain into very disciplined on -- you'll only get what you need now sort of behavior. So I don't think there's any double ordering. And then you get back to our DSP world. We are the only pure-play DSP provider in the optical pluggable. [indiscernible] capacities are allocated by companies by the major foundry, and if you are in 5 nanometer, for example, you're welcome to choose how you distribute your silicon between switch, pluggables, GPUs, TPUs and so on and so forth, right? So we every company gets to prioritize what is in their best

value interest. So I would say that, that is not the case at all, number one. Number 2 is that even to place over ordering the data since guys are very, very disciplined about creating pure supply chains. That means what I mean by that is the optics, the DSP for a particular qualification in a data center is very unique so that they don't cross corrupt the supply chain. So I think it's golden, it's not a screw. But it's a bulk. So there is a bold there that is really, really fastening the cantilever bridge quite well here. So that's for those in the [indiscernible] area. But yes, it's looking pretty good.

Christopher Rolland

Excellent. Maybe a follow-up there and a second question. I guess, first of all, obviously, you guys have some unique 4-nanometer DSP capacity that others don't. Can you talk about how you see constraints in the optical market playing out? Is it DSP? Is it other parts, lasers, et cetera, if you could talk about your place there? And then secondly, DSP in the world of -- some believe it's less meaningful, would like your take, Kishore, as a technologist.

Kishore Seendripu

Okay. Chris, a lot of questions out there that you posed here to me. I will take the first one, which is basically really to supply. I think the optic supply is easing we are really constrained by foundries on the DSP supply. And the data center build-out is really accelerating. So if you sit here and talk, if the demand -- people talked about TAM units for optical pluggable, no matter what flavors in the order of 120 million units. Now they talk about '27, '28, maybe 200 million units. So I'm not keeping count anymore, right? So however, the fab capacity is not coming online. These are huge multiyear efforts. And so there's really, really a fab capacity shortage, I would say. And so the DSP supply is a pretty strong supply constraint. And we are really being -- and the opportunity on the CPO stuff really is a scale-up side, which is a greenfield opportunity for us. But I do believe the NPU is going to be the more bigger success driver than a CPO.

And in an NPU scenario to as well, as you go to higher longer term, they'd be DSP required just like an LRO configuration on the what you call a half DSP because we can all agree now that the LPOs are really, really a very small fraction, point technology and they don't have legs for the future. right? So we have a product portfolio that spans we have drivers and also, we have announced products along those various categories. So I feel we are very well positioned in whichever way the industry goes, then we'll have a full portfolio.

Operator

Our next question comes from the line of Karl Ackerman with BNP Paribas.

Samuel Feldman

This is Sam Feldman on for Karl Ackerman. So first, based on your extended visibility into C8, can you provide any color on how we should think about the growth rate the relative growth rate between 800 and 1.60. And should we expect 400 bps to grow through calendar '27?

Steven Litchfield

So I'll take that. I mean, with regard -- so this year, our optical growth is almost all in the first half, as we've stated and Kishore mentioned earlier, 1.6 starts in '27, but it will have a much bigger growth year in 2028. -- what is the mix in' '28? I mean I think it's substantially 1.6. Maybe you'll see 800 kind of trailing off. But potentially, it can sustain as well up until probably 2029. So I mean, we see significant growth in 2017 as well as 28% on both of those.

Operator

Our next question comes from the line of Quinn Bolton with Needham & Company.

Quinn Bolton

Congratulations on the results and strong GSP outlook. I guess, Steve, for you, it looks like the prepaid assets on the balance sheet increased by about $50 million in the September quarter, I assume that, that's probably the large majority of that or your prepayments that you talked about. Given that it sounds like you're fairly well booked for calendar 2027 now, in the DSP business. Do you feel that the prepayments and the wafers you secured support that 2027 backlog? Do you still have more work to do in the fourth quarter in terms of prepayments and securing wafers to meet the demand you're seeing for calendar '27?

Steven Litchfield

Yes. So I mean we talked about this a little bit before. Similar to others, I mean, the prepayments, mostly on the wafer side. I mean there's lots of other things going on out there. I mean, on the balance sheet isn't just TSMC. There's other things on prepayments as well that we don't break out. As far as -- I mean, I talked earlier about visibility in '27, yes. I mean backlog, as you would imagine, is very, very good in '27 and into 2028 prepaid. I mean, look, I expect prepays to continue to go up from here as our revenue projections continue to go up as well. So I think that's part of the reason why we increased the revolver itself, just to make sure that we can capture all of that upside capacity that's coming from increased market share that we've gained.

Quinn Bolton

Got it. And then a follow-up. Steve, I think in some of the investor conferences over the past few months, you estimated your DSP share currently maybe in the 5% to 7% range. Do you think you're at the high end of that or potentially above that by the end of 2026, just based on the strong backlog you're seeing and the increase in the DSP forecast this year?

Kishore Seendripu

This is Kishore. I think it's very hard as to be [indiscernible] yes, thank you. I think it's very hard to us to make share given the accelerating demand on units as well. So if I'm looking backward, then I think we are well above the number. If I'm looking forward, maybe we made more progress beyond that 1%, but I don't know at what level we are. So can I say what we will look like 2 years from now? Sure. It should be in the teens for sure, right? And so I think that's a very backward-looking number. on a share percentage basis on a backward-looking TAM basis, it should be larger. But on a forward-looking basis, very hard to estimate because the numbers are moving all around. But suffice to say, we are gaining share every day. I think in a 12 to 24 window, we could be neck and neck with 1 of our 2 competitors, right? That's the hope, that's the goal, and that's the momentum.

Operator

Our next question comes from the line of Tim Savageaux with Northland Capital Markets.

Timothy Savageaux

I'll add my congrats on the results and the outlook. Now I wanted to go back to -- to welcome kind of market growth, market share type discussion. And Steve, you made a comment earlier in the call about outgrowing the market. And I think we may have a baseline there. Marvell talked about kind of the AI interconnect unit for them growing, call it, 65% on a CAGR through 2030. It's coincidentally what MaxLinear is growing overall and in '26. But as we look at the optical DSP business, is it fair to use something like that as a baseline. And when you say you want to outgrow the market, I mean just based on your run rate, it seems like we should be thinking closer to a double here than anything else for '27. But I just kind of wanted to bounce that reasoning of view.

Steven Litchfield

Tim, I feel very confident we're outgrowing the market, which I think is supportive of what you're asking.

Kishore Seendripu

And we can't speak for Marvell's numbers.

Timothy Savageaux

Okay. Well, I'll go on and ask this one anyway, then, which is, yes, I think your market share is like 6% here. If you take those forecasts out to their conclusion, I mean, they're talking about, I don't know, a $15 billion optical DSP market I mean if you get into double-digit before that, you should have a $1 billion business on your hand pretty quickly. I want to bounce that one too, and that will be it for me.

Kishore Seendripu

Tim, we can't comment on that, right?

Steven Litchfield

Yes. But the numbers you're talking about, I mean, look, there's a lot of growth there. I mean Kishore commented earlier about the transceiver growth. I mean people are talking 200 million units next year. I mean -- so these numbers make a lot of sense, as Kishore stated, being up in the teens over the next couple of years, very achievable. I think we are doing a great job. I think you've seen that from us in 2026, but I would say it's just getting going -- you're going to continue to see through that in '27, '28.

Operator

Our next question comes from the line of Joe Quatrochi with Wells Fargo.

Unknown Analyst

This is [indiscernible] on for Joe. So Steve, I just wanted to switch gears to broadband. You spoke of seeing some seasonal softness in 4Q 2026. Just curious how we should think about this moving forward into the first half of 2027 in the second half of '27. And just the overall broadband recovery in general?

Steven Litchfield

Yes. Thanks, Tavis. Yes, I mean, we actually feel pretty good. I mean, we do see some seasonality in Q4, certainly. But as far as '27 goes, I mean you just as we had talked about previously with the telco spending, but then also our market share gains on the PON side, we do expect to see growth in 2027. So we feel very good about that.

Operator

Our Next question is from the line of Tore Svanberg with Stifel.

Tore Svanberg

Now the basic questions are out of the way. I just had a few more specific follow-ups. So Kishore, I think you said you expect 1.6% revenue already first half of next year. I think I think in the last quarter, I think you talked about maybe middle of the year. So what changed there? Is it the market that's accelerating earlier? Is it based on some new design wins? Any clarification there? And by the way, does that include DSP and TIA or is it just?

Kishore Seendripu

So, Tore, obviously, it's we do not know where it would be, but I think there's an initial design means that will start shipping in the first half. The exact amount of quantity is not clear, but it's seal in the sense that it's the second half very nicely. And also second part also is correct what you asked. They are accelerating design call momentum going on. in somewhat we had bottle by our own ability to support multiple calls have commitments, both for DSP and TIA in these design wins. So it's not just our DSP alone. Obviously be in a world where we want to attach our own TAA to our own DSP for selfish reasons beyond just a great performance advantage and cost savings for our own customers. in terms of power efficiency and the total BOM efficiency, right? So I think all the three are a strong affirmative response.

Tore Svanberg

Very good. And last one is on pricing or content. So we're hearing as you go from 800 gig to 1.60 that maybe prices could be 2x you're obviously going to be shipping more than the DSP with the TIA. So in your case, I guess, you could potentially have more than 2x as the market goes from 800 gig to '26?

Kishore Seendripu

He has what I call trajectory guide to our NPL strategy and our CPO strategy and our LPO strategy and LRO strategy. It's pretty complicated. So we're constraining how many we can activate at the same time. So you, of course, sequence and prioritize them. That's the constraint I was referring to.

Got you. And then just lastly, with Washington, can you talk about your expected attach rates to rush more with the power envelope that you need to meet. That is the true despite what anybody can tell. So we hope to be the other guy outside of the 1 guy and we have shown that. So we expect the tax rate to be 100% at least and more in 1.5 years from now, right? That's the #1 answer. The next question you had was related to -- could you repeat that for me, please? I just forgot -- and [indiscernible] yes, it's active electrical cable, right? There's one large leader, which everybody knows is Credo. They're 93% or maybe 99% of the entire AC cable market, but I think the market is diversifying now because the problem is getting a lot more complicated and there are going to be multiple AEC vendors.

And our solution will be a pure silicon play to selling it to a cable maker or somebody who makes the, what I call the paddle card for the -- so our revenues for the AEC will not be as large as selling oval cable, but it will correspond to what an AEC market would dictate in terms of silicon price inside it. So I would think that it'd be a similar pricing as our PAM4 DSP for optical. And we expect not to get a big share, let's assume, but we are planning around 1 or 2 big data centers and then a share of those guys through intermediaries. And so I think we should be able to get to 20% to 30% in the 1.6 terabit cycle. And there in the next couple of years. So -- but we won't expand into multiple data centers, right, we only 1 or 2 data centers in the U.S.

Operator

And we have reached the end of the question-and-answer session. And therefore, I would like to turn the call back over to Leslie Green for closing remarks.

Leslie Green

Thank you, Shamali, and thank you all for joining. This quarter, we will be presenting at a number of financial conferences and virtual events. Details will be posted on our Investor Relations page. Thank you all for joining today, and we look forward to speaking with you again soon.

Operator

Thank you. And this concludes today's conference, and you may disconnect your lines at this time. We thank you for your participation.