芯源系統 (MPWR) 2026 年第三季法說會：營收達 12 億美元

重點摘要

• 芯源系統 (Monolithic Power Systems, MPWR) 公佈 2026 財年第三季營收達 12 億美元，首度單季突破 10 億美元大關。營收季增 22%，年增 63%。
• 企業數據 (Enterprise Data) 業務仍為主要成長動能。在客戶多元化、模組採用率提高以及新平台平均單機價值 (content) 提升的支撐下，管理層將該領域的年增率下限從 130% 上調至 150%。
• 第三季可支撐年營收的產能已突破 60 億美元。MPS 預計在邁向 100 億美元目標的過程中，產能將於 2027 年下半年達到 80 億美元。
• 在光模組與交換器產品的共同貢獻下，通訊業務成長加速。管理層預期通訊將成為 MPS 成長第二快的終端市場。
• MPS 接獲車用 48 伏特電源解決方案的首批訂單，並預期 48 伏特與區域 (zonal) 架構將為 2027 年的成長提供支撐。此外，機器人產品也在多個地區開始放量。
• 該公司在本季度實施庫藏股買回，金額超過 4.5 億美元。

關鍵財務數據

業務與營運表現

既有客戶對企業數據的需求保持強勁，管理層預計該動能將持續至 2027 年。MPS 也指出 ASIC 客戶群實現快速成長，這類客戶主要採用平均售價 (ASP) 較高的模組。該公司正運用內部開發的 AI 設計工具來擴大涵蓋的客戶與應用範圍，包含桌上型電腦 AI 設計。

通訊業務受惠於光模組與交換器產品。由於基期較低，以百分比計算，交換器業務是通訊領域中成長最快的細分市場。管理層預期過渡至 1.6T 光模組將同時帶來出貨量與單機價值的成長。

在儲存與運算領域，DDR5 PMIC（電源管理 IC）是最大成長貢獻來源。傳統硬碟 (HDD) 與固態硬碟 (SSD) 產品亦錄得健全成長。MPS 正開發時脈驅動晶片 (clock driver) 與暫存時脈驅動晶片 (RCD)，預計將於 2027 或 2028 年產生營收，並預期於 2028 年迎來更大幅度的放量。

車用業務成長擴及美洲、歐洲與亞洲，其中中國的設計導入 (design-in) 活動特別熱絡。MPS 已收到 48 伏特電源產品的首批訂單，並將 48 伏特與區域架構視為下一波潛在成長浪潮。

工業領域呈現廣泛成長。用於機器人的感測器與馬達控制產品開始放量，不過管理層將 2026 年主要定位為累積設計獲勝 (design win) 的一年，預期 2027 年將會有更顯著的貢獻。

MPS 還贏得了首個大型資料轉換器 (data converter) 設計案，管理層預計該產品將於 2027 年開始放量。此外，該公司已開始提供針對 800 伏特資料中心架構自主研發的氮化鎵 (GaN) 產品送樣，與既有的碳化矽 (SiC) 解決方案並行。

管理層展望

管理層預計可支撐年營收的產能將於 2027 年下半年達到 80 億美元。公司未提供具體的 2027 年營收預測，表示擴充產能旨在超前佈署以因應預期需求。

在毛利率方面，MPS 繼續以 55% 至 60% 為目標模型，保持漸進且持續的擴增。第四季財測中位數顯示將比上一季再增加 10 個基點。

在營收高成長率下，營業費用預計將維持低於公司的模型水準，因為支出擴增速度無法像營收那般迅速。管理層強調注重研發效率，而非僅為了搭配營收成長而增加支出。

MPS 預計 800 伏特架構將於 2027 年底迎來初步放量，而更大的機會將在 2028 年。管理層強調，目前的企業數據展望並不依賴 800 伏特產品的重大營收貢獻。

風險與關注焦點

• 由於下游記憶體供應限制，筆記型電腦短期需求較為疲弱，抵銷了儲存業務的強勁成長。管理層形容整體儲存與運算領域年增率可能持平或僅成長數個百分點。
• MPS 正吸收部分上升的投入成本，並與客戶共同分攤部分成本。價格調漲並非營收成長的主要驅動力，其主要用於保護毛利率。
• 氮化鎵 (GaN) 仍需經過廣泛的可靠性驗證。管理層將失效率目標定在 1 ppm 以下，並允許客戶在 GaN 與成熟的碳化矽 (SiC) 解決方案之間進行選擇。
• 採行 800 伏特架構的時間點與規模仍具不確定性。管理層預期營收貢獻將出現在 2027 年較晚時候及 2028 年，而非近期。
• 通路庫存仍處於公司正常範圍的低檔水準。管理層表示庫存目前並未限制銷售，但產能與晶圓投片量必須持續維持以支撐成長。

分析師問答亮點

分析師聚焦於企業數據的成長能否持續至 2027 年。管理層指出涵蓋更廣泛的客戶群、模組滲透率提升、平台單機價值增加，以及代理型 AI (agentic AI) 工作負載帶動的潛在 GPU 需求，但未提供明年的具體成長率數據。

在資料中心電源架構方面，管理層表示更高 GPU 功率需求有利於垂直供電 (vertical power delivery) 與模組採用，因為水平佈局面臨空間限制。據 MPS 表示，業界大部分業者已朝模組化邁進。

管理層表示，無晶圓廠 (fabless) 的營運模式應能在不顯著增加資本支出強度或改變資本結構的情況下擴充產能。公司強調，80 億美元的產能目標代表對需求的準備就緒，而非特定 2027 年單季營收年化率 (run rate)。

關於利潤率，MPS 重申偏好穩定改善毛利率，而非隨後可能回吐的暫時性提升。管理層亦指出，營業利益率最近每季大約擴增一個百分點。

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 day, and thank you for standing by. Welcome to the Monolithic Power Systems Third Quarter 2026 Earnings Conference Call. [Operator Instructions] Please be advised that today's conference is being recorded. I would now like to hand the conference over to your speaker today, Arthur Lee. Please go ahead.

Arthur Lee

Earlier today, MPS released a written commentary on the results of its operations for the third quarter ended September 30, 2026. This document can be found on our website.

Before we begin, I would like to remind everyone that in the course of today's presentation, we may make forward-looking statements and projections within the meaning of the Private Securities Litigation Reform Act of 1995 that involve risks and uncertainties. The risks, uncertainties and other factors that could cause actual results to differ from these forward-looking statements are identified in the safe harbor statements contained in the Q3 2026 earnings commentary and in our SEC filings, including our Form 10-K and Form 10-Q, which can be found on our website.

Our statements are made as of today, and we assume no obligation to update this information. Now I would like to turn the call over to Tony.

Tony Balow

Thanks, Arthur. Good afternoon, and for us, good morning from Shanghai, and welcome to our Q3 2026 earnings call. In Q3, MPS achieved another significant milestone, exceeding $1 billion in quarterly revenue for the first time. Q3 revenue of $1.2 billion was 22% higher than the second quarter of 2026 and 63% higher than the third quarter of 2025.

During the quarter, our capacity surpassed $6 billion in annual revenue support. We expect to have $8 billion of capacity in place by the second half of 2027 on our way to $10 billion in support of our rapidly expanding business. In our enterprise data end market, we saw strong demand across existing customers, which we expect to continue into 2027. In addition, using our in-house developed AI design tools, we significantly expanded our customer base in traditional areas as well as landing new wins in applications such as desktop AI designs.

Beyond our strong enterprise data growth in automotive, we received first orders for our 48-volt power solutions and expect design wins for 48-volt and zonal architectures to be a driver of growth in 2027. We started ramping robotic solutions with sensor and motor control products across multiple geographies. We continue to increase our SAM with the ramp of our high-speed memory products.

We landed our first large data converter design win, which we expect to ramp next year. And we continue to build out our high-voltage portfolio and began sampling GaN-based solutions for use in 800-volt data center architectures, complementing our existing silicon carbide-based designs. And finally, in Q3, we repurchased over $450 million of our own stock. Overall, we see accelerated growth. Our diversified growth strategy is paying off.

Operator, you may now open the webinar for questions.

Operator

[Operator Instructions] Our first question comes from Tore Svanberg with Stifel.

分析師問答

Tore Svanberg

Michael and Tony, congratulations on that $1 billion milestone. And to you, Tony, congratulations on your CFO appointment. I guess for my first question, you talked about accelerated growth. I was hoping you can maybe elaborate a little bit of that. I mean I'm sure you're not going to give us calendar '27 guidance, but you are on track to growing 50%, 60% this year. And especially in light of your $8 billion capacity and then $10 billion, when could we potentially see $10 billion capacity in place?

Michael R. Hsing

Yes. Thanks, Tore. See as Tony mentioned is only a few items, so we see beyond that. And I see this explosive growth period in the next few years. And even this year, you see the first 9 months, okay? And we don't see anything -- any signs of the slowdown.

Tony Balow

Yes. The other thing I'd add is I think as we've come into the second half, you started to see growth across multiple of our end markets, including like auto, which the ramp started here in Q3, and we've talked about continuing here through the second half.

And then as we start to see applications like robotics begin to ramp, we'd expect that then to continue into the future. So I think it's not just the existing businesses, it's adding on the additional growth areas over time, which is really giving us confidence going forward.

Tore Svanberg

Very good. And as my follow-up, you mentioned the 800-volt GaN product sampling. Now that you are sampling GaN, could you maybe elaborate a little bit more, help us understand a little bit more the positioning there? Is this vertical GaN? Is it horizontal GaN? Is it your own process? Any more color you can help us there? And when do you expect to start to see some 800-volt GaN revenue?

Michael R. Hsing

Yes. We let -- it is based on MPS own development, which is later than we developed silicon carbide device. And we start to sampling that this type of solutions. We let our customers choose in either using GaN, the GaN is a new technology for everyone, okay? And for everyone adopting it in the data centers and the industrial automotive type of applications, okay? This is the first time.

And we are always very cautious about the reliability. We want to achieve below 1 ppm failures and for these modules. And we let our customers choose, okay? GaN or silicon carbide. Silicon carbide is known, we know the data. And so that's a very high reliability product and GaNs we have to ship somewhere 50 million units before we can say this is a good -- this is a new one for GaN, but we let our customers choose.

Tony Balow

Yes. I'll just add on it, I think all of the experience -- I was just going to add a few things there. This is just the experience we've already gained in data center, I think, puts us in a really good position to do what Michael is talking about in terms of making sure we can meet all the standards that customer requires.

You asked a question on the timing of it. I think right now, we're still -- we've always kind of been late in '27 for I think kind of the 800-volt kind of architecture ramps with probably more in '28. I don't think our thinking on that has changed. We're just obviously getting ahead of that design work now.

Michael R. Hsing

Even for our customers, they use both. They don't want to have a failure either. They use both solutions.

Operator

Our next question comes from Rick Schafer with Oppenheimer & Co.

Richard Schafer

Congratulations to you guys and congratulations, Tony, on the appointment. If I could ask a question on storage and compute for my first one, I was hoping maybe you could level set us on some of the moving parts in there. I mean, how does that segment split now if you look at SSD, HDD, DDR5, PMIC versus the PC, the VCore power piece? Because I know we all see PC units tracking down about 20% here in the second half. So I'm just curious, is that kind of what you guys are seeing on the PC side of that segment? And then how much is the storage piece growing? And what's it contributing?

Tony Balow

Yes, Rick, let me start, and then I'll let Michael come in. I think just kind of mechanically, if you look at it, the big driver of growth in that segment has been the DDR5, primarily on the PMIC solution. As we said, we're just starting the high-speed journey on the DDR5. We have seen pretty healthy growth over the past year or so on the hard disk drive and SSD part of that storage business as well. But clearly, DDR5 is the biggest portion of that.

On notebook, sort of tactically, yes, I think we're -- we've been talking about this, and that's what we're seeing as well, right? We're seeing not the same strength on notebook due to what's happening downstream with some of the memory constraints. But I think, if you look forward, I think we're still pretty bullish on notebook in the long term because we have really good technology and good share -- and then I think there's applications around AI that will continue to build out over time at the edge, of which notebook will play a part. So I think what you're seeing in the short term around notebook tracks with what you're seeing, Rick. But I think long term, I think we're still pretty high on what notebook could be for the company.

Michael R. Hsing

Yes. For that segment, we have a new products like a clock drivers and RCD. These all lined up and other than in addition to our power management product. These will be lined up for revenue in 2027 or '28. And we see even '28, these products will really ramping up, okay? So that's part of a TAM is much bigger than power management.

And the notebooks, as Tony mentioned it, we will see this as a very fast growing market segments because after the manufacturing capacity opens up, that the segment will grow along with the desktop.

Tony Balow

If you sum that up a little bit, what we said for kind of the segment overall is with the puts and takes, it could be kind of flattish, up a couple of points year-over-year is kind of how we positioned it, but it's really a tale of 2 cities, as we've talked about with really strong storage and then the TAM just being limited on the notebook side.

Richard Schafer

Got it. And then if I could -- this is your first call with the CFO moniker. And I'm curious how you look at OpEx growth going forward? I mean, any meaningful changes to the prior targets that you guys have laid out? I mean, at these type of top line growth rates that Tore called out, 50%, 60% type top line, how fast can you actually scale R&D, et cetera? It seems like we're -- it seems like you guys are set for some pretty material operating leverage to be flowing through the model. Is that the right way to be thinking about it? Any rules of thumb in there?

Michael R. Hsing

Let me answer that part of the question, Rick. I never believe the big dollars R&D equivalent to a bigger result. It's -- for our business, it's inversely proportional to that. We just need some good ideas, good people that we're going to make everything happen.

Tony Balow

So yes, I think, Rick, you're not going to see us necessarily change our model. What we've seen pragmatically, though, as we start to get to higher levels of growth that we start to fall below our model range in terms of we just can't spend money that fast. But I completely can agree with Michael on that. It's about the efficiency of the R&D and the people that you get versus just spending more. But in general, I'd say at the growth rates that we've seen, for example, this year, you're going to fall below the model from an OpEx perspective.

Operator

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

Quinn Bolton

I'll echo my congratulations and Tony, congratulations on the new role. I guess I want to start with the enterprise data. Last quarter, you gave us a floor of 130%. With September results and December guidance, it kind of looks like you're tracking to maybe 150% growth. But wondering, Tony, if you might be able to level set us what a new floor growth for enterprise data might be? And then I know it's early, but any sort of thoughts on that segment as you look into 2027, what kind of growth rate you might be looking at in the forward case? And then I've got a follow-on in the comms business.

Tony Balow

Sure. Let me start just kind of with the -- to answer your question, straight. I think at this point, given what we've seen in Q3 and Q4, we're comfortable raising the floor from -- we were at 130% year-over-year. But to your point, I think we're comfortable with a floor of 150% year-over-year now going forward.

On 2027, I think it's a little early for us to go put a percentage out at this time, but I'll start and Michael, I'm sure, will follow up in that. Structurally, the growth drivers that we're seeing this year are all intact, right, whether you see the diversification of customers, the growth in modules, the content gains as new platforms roll out. And I think over time, as you start to see even stronger growth in GPU potentially driven by the Agentic AI workloads, that's going to be even a further tailwind out in time. So while a little early to call next year, I feel really good about the growth drivers.

Michael R. Hsing

Well, I see in the next years and beyond next years, and I'm very excited to see that we build our own data centers. And as Tony mentioned earlier, the result is a tremendous improvement. And so this is not in the data centers, but in MPS own servers. And all these server, we believe -- I believe it will change and change to a a GaN type of servers, and again the hardware will change and that will trickle down to desktops and even the notebook I said earlier.

And this type of growth, it will trickle down in the next year and the year after. We see those trends, these design activities is really happening. And so this is not only this year, not only in the data centers, it will trickle down to all other areas.

Tony Balow

Do you have a follow-up on comms?

Quinn Bolton

Yes. It's obviously congrats on the nice growth. I think it was 43% sequentially. It really looks like it was a breakout quarter in September. Wondering if that was primarily driven by optical modules? Or are you seeing a broader set of contributors to that growth? And just any thoughts into the December quarter? Does that sort of pace of growth continue into December?

Tony Balow

It's a great question. And to take us back here, obviously, coming into the year, we obviously had very strong growth in optical modules. And what we added beginning of the year is what we've been calling the switch business. And I know I've talked about this before. But to answer your question, why are you starting to see the accelerated growth in that particular end market is because you are adding both of those growth drivers together, and we saw the effect really in Q3.

And I think for the year, we've been positioning it is it's -- I think comms is going to be our second fastest-growing end market for sure. It won't be quite as fast as enterprise data, but certainly won't be far behind. And just like enterprise data, I think if you go into next year, all of those growth drivers are intact. On optical modules, you see the transition to 1.6. That's not just a volume play, but there's a content play there as well.

And then certainly, as we continue to build out additional sockets in the switch business, that's going to be a driver of growth as well. So I think right now, we really like the comms business, not just this year, but going into next year.

Michael R. Hsing

And on the macro economy and market segment predictions, I only care what is the design activity and what our customers are excited about it, what we're excited about it, all these new technology, new product introductions. And as I said earlier to [ market ], all these segments, we introduced the new product and all the design activities that give us an indication clearly the growth for even beyond next year.

Operator

Our next question comes from Joshua Buchalter with TD Cowen.

Joshua Buchalter

Congrats on the results and echoing the congrats to Tony on the CFO gig. Best of luck. I'll stick with policy and ask one on enterprise data and then one on another segment. Maybe can you comment on the diversity of enterprise data? Like how has that business evolved over the last year or 2 and looking into 2027? I would just be curious to hear how much you see coming from GPUs versus ASICs or even any comments you can help us on how much is coming from modules versus discrete at this point and expectations into '27?

Michael R. Hsing

Yes. Modules versus the chip design, chip power supplies, and then ICs on the chip all of them, okay? I mean even for the server side, and okay, we see some -- many activities and going through using modules because of the high current high-power demand. That's my comment, okay? And it's all segments -- from growth -- the real growth percentage, I'll let Tony to...

Tony Balow

Yes, I think -- and that -- I think it's an important point around this customer diversity that starting in the middle of 2025 until now, what we've seen is a really rapid growth in a lot of ASIC customers, which again are primarily using modules over time. And so that's benefited from a revenue diversity perspective as those carry the higher ASP associated with that.

I think furthermore, what Michael was talking about in terms of how we've been using AI to create scale to allow us to go address even a longer tail of small customers over time, which may be big customers in the future. And that's really what we try to do, go in every socket, not knowing who ultimately in the long term might grow into a sizable customer.

Michael R. Hsing

Yes. All we clearly see is that we have the huge beneficial from these artificial environment, these artificial intelligence based on these large language model. Of course, MPS [indiscernible] is internal. And it benefit us, it will benefit for everybody. And that's where I'm very optimistic about this and our growth trend will continue.

Joshua Buchalter

Maybe in the same theme of starting with small sockets that become big opportunities. Any context you can provide on how much revenue contribution we can expect from the initial high-speed data converter design win? And how big is that TAM as we think about you growing into this new market?

Michael R. Hsing

Yes. This is -- we can't tell because we have -- we just counted, these are over -- conservatively well over 6,000 products, okay? And we continue to develop newer products. So what we do is we do every segment in our best, far better than -- or to meet the new market demand. And we are much faster than any of our competitors.

And as long as we do, the segments do the best and the revenue will come. And we don't -- we just don't know what -- which segment. 7, probably 8 years ago, we look at our product lines, look at the optical market segment, one of the perfect example. We didn't know today was that high demand. And we got it in there. We serve our customers regardless of the large or small. And that's what happened today. I cannot predict next 10 years, next 5 years, we will be bigger, and we serve our small customers well.

Tony Balow

Yes. I think -- and again, to Michael's point, we're not ready to call the long term. We're just beginning that process. So I wouldn't necessarily say it moves the needle here in 2026. And I think as we get into next year, we'll talk more and see how that ramp goes.

Operator

Our next question comes from Will Stein with Truist Securities.

William Stein

First, on data center vertical power. I think there's still one meaningful customer doing lateral and I'm wondering if you can comment as to whether you think that converts to vertical in the coming year and whether that would necessarily require a module from you instead of some other sort of solution.

Michael R. Hsing

Certainly, okay, I don't want to speak for our customers, in particular that one customer. But I said earlier, and you know that what we did in the -- we predicted in the 2016 -- and for the server market, MPS solution is so small, and we have to use the adapter to meet the common footprint. And so when -- for the server market segment, we predicted and if you go above 600, 700 watts and you run out of peripheral and you have to use a smaller footprint.

But the server GPU never get that never get to the 600, 700 watts, but the GPU did it, okay? And as GPU power increase, I don't have a generalized numbers, okay? It will -- the power goes up, you run out of peripheral, you have to do vertical. And we said this many years earlier, okay? I still believe that very logical to me.

Tony Balow

Yes. I mean I'm just going to echo Michael. I think that we won't speak for that. Most of the industry has already moved to modules, right? And that's what we've seen in terms of content gains as well as the volume gains as we've gone through enterprise data. And I think that will be a great question for our customers. Yes, take a last jump [indiscernible].

William Stein

Great. And a different topic. Someone asked about the OpEx leverage. And I guess I understand that for the time being, with the rapid growth, you're going to sort of undershoot on spending, overshoot on margins. What about gross margin? I think the company has talked for a long time about 10 to 20 bps per quarter gross margin expansion. There was a time when you weren't in that range, and it seems like you are today. Any likelihood that because of the growth, you could exceed that expansion rate? Or should we still think about 10 bps to 20 bps is likely? Or any update there would be helpful.

Tony Balow

No. And thanks for calling out. Yes, you saw that we obviously did increase 10 basis points here in Q3. The midpoint of our guide in Q4 is kind of another 10 basis points. So I think we're kind of squarely in the model that we like, right? 55% to 60% is the sweet spot for sure. And within that being kind of slow, methodical about expansion. I think we're going to continue on that path.

Structurally, I feel pretty good about that. We'll see what happens in the future. But remember, what we don't want to do ever is put a bunch of gross margin in one quarter and then take it away the next quarter and move it around a bunch...

Michael R. Hsing

During the capacity constraint, we don't gouge our customers. And so we want to deliver to you guys a very consistent story. And I don't want to answer a question when the margins drops? And I want our MPS to keep it same business, same performance year-over-year -- quarter-over-quarter, year-over-year.

Tony Balow

That's right. So I think the model still holds. I'm very comfortable with that. And I think you called it out, but I'll highlight it again. I think if you look at the flow-through to operating margin, we've been growing operating margin, it's kind of a point a quarter almost. So we've seen a lot of expansion on the operating margin line.

Michael R. Hsing

We are in a making money business. And that's all I would point here.

Operator

Next question comes from Chris Caso with Wolfe Research.

Christopher Caso

Just following from the prior discussion, maybe you could talk about how you're dealing with rising costs. And obviously, costs are going up a lot of different places. There's a lot of talk of foundry wafer prices going higher. So to what extent are you passing that along? And as you go forward, does any of that result in pricing or any kind of revenue tailwind as you have to absorb some of those costs and pass them along to customers?

Michael R. Hsing

We definitely absorb some of the cost increase. And not that we don't increase the price, okay? And we some -- we share the cost with -- the cost increase with our customers.

Tony Balow

Yes, I think that's right, Chris. And I think if what you're asking is from some of the results, how much of that is from pricing, it's really not a significant contributor to the revenue growth that we've had. We are more just making sure that we don't dilute ourselves on the gross margin line.

Michael R. Hsing

Yes. Our customers have a very long memory.

Christopher Caso

Right. Okay. As a follow-up, it's a question on the 800-volt transition. And I know there's a lot of noise out there with regard to the pace of that transition. What -- in terms of MPS, how important is that to you in '27? And I guess maybe the question is, is it really 800 volts a very material driver of content in the EV segment? Or is the majority of that still going to wind up in the existing low power -- I'm sorry, low-voltage designs?

Michael R. Hsing

Yes, we don't want to add more voices, okay? We don't talk about it until the revenue for sure will come, okay? I do see the revenue for sure will come. Not this quarter, not next quarter, probably end of the '27 and 2028. And it's clearly the DC supply for high power demand and will be the trend, will be the data center, all will convert it to that. And the power savings is, I think, beyond like single digit -- single-digit percentage improvement versus the traditional power supply.

I see this is even beyond the data center. And if the data center can improve, let's say, 5%, where do you get the 5% money returns? And all the commercial building, even residential, they all can be like this, 400 volt these are in -- there will be -- in Europe will be 800 volts in U.S. will be 400 volts. And that these will be a significant power savings across the board. That's what I'm excited about this opportunity now.

Tony Balow

Yes. I mean I'll even try to just address the question the other way, which is when we talked about our -- how we are excited about growth in enterprise data, it's not because we're -- we have a big built-in ramp on 800 volt. It's really around the existing customers, the new customers, transition to modules, all the things that we've been talking about, Chris. So if you're kind of trying to figure out how much of that excitement was just at 800 volt, I really think the business we have today is what we're confident in.

Michael R. Hsing

Yes. MPS now we're using our own building do a lot experiment, using these 800 volts and using the technology and see what the power saving is.

Operator

Our next question comes from Sebastien Naji with William Blair.

Sebastien Cyrus Naji

Congratulations, Tony, on your appointment. I wanted to maybe just ask my first question, given some of the news around integrated voltage regulators and the acquisition of Empower by Analog Devices, maybe could you just give us your view on this technology and whether you see it as a competitive risk in the AI data center or the larger data center?

Michael R. Hsing

Yes, we don't comment on other companies' activities. And we have our own solutions. We have our own and we can deliver whatever customer wanted. And this is not BS talks, we do have multiple beyond what you guys beyond the noise levels. And we have those technologies and we can deliver. And the GPU evolves very quickly, the power levels, the architecture and in terms of the power delivery. And we don't talk about those until we have -- we see -- clearly see where the revenue will come.

Sebastien Cyrus Naji

Okay. Okay. Great. That's helpful. We'll look forward to those announcements when those products are ready. Maybe moving to something that's a little bit closer to your heart, Michael, which is the building automation opportunity. I think last quarter, you talked about $40 billion to $50 billion in size, but that the real bottleneck was the software still needed to catch up. Can you maybe just talk a little bit about the progress that's been made here as we get to the end of the year?

Michael R. Hsing

Well, good questions, okay? That's my baby project. And we just have installed in our 1 million square feet building. That's in China. And we -- as I said just a minute ago, we're experimenting our own AC to DC power. These are for 800-volt supply. And for across the building, -- and that project -- these projects just started. But in terms of turning to a $50 billion revenue will take some time. We don't know how to get there yet, but the goal is very clear.

And we will get there. There's a lot of software involved in it, okay? But on the hardware side, we pretty much have every single component. And our single -- our components, our chips, our module technology is far better than -- sorry to say these are the current existing technology. These are 50-year old, and we will do some significant improvement on it.

Operator

Our next question comes from Kelsey Chia with Citi.

Wei Chia

Congrats on the great results again. And Tony, congrats on your well-deserved appointment. So my question is on the switching business. So to what extent is that contributing to the strong communications growth that we saw in third quarter? And how should we think about the ramp as we head into 2027 and if that business is accretive to gross margins?

Tony Balow

Yes, I'll start on that one. So we're obviously not going to start guiding sub end markets a little bit. But certainly, from just a percentage basis, it's been growing very quickly, probably on a percentage basis, the fastest-growing sub end market inside of communications just because it started from a very low amount at the beginning of the year. As we go forward, again, I still think that and optical modules are going to be significant growth drivers for communications into 2027.

On the margin side, I think if you look at all of our end markets, plus or minus, they'll all operate around the corporate average because to Michael's point, we have north of 6,000, 7,000 products. And so if you look at the distribution of that, basically, what we said is as we grow this, we don't think growing one end market over another will necessarily take us out of our model. And so right now, I would still take MPS as a total kind of in that 55% to 60% range with sort of the slow methodical expansion when we can provide it versus worrying about whether growth in one end market would take you out of model.

Wei Chia

And my second question is on the industrial end market. I mean, very strong growth this quarter as well. What drove that strong growth this quarter? And if you could talk about the ramp of the robotic solutions? And how do you see that ramping in '27?

Michael R. Hsing

The robotics, we -- I tried to enter the robotics in 2016. On the Analyst Day, we talked about robotics, that's the area we're focused on. We have a product for these sensors and motor controls -- and I didn't know what I was talking about, the revenue didn't come, okay? And now it shows up.

And I think this is still at the very beginning, once the vision for the recognition and visual recognition and improved from using the AI-based vision improvement technology. These robots, as you know, they will propagate in everyday life. So I see this -- that's the future. And the revenue would be significantly higher than that.

Tony Balow

Yes. I think the only thing I'd add kind of on the overall segment is, I think this year, we still look at it as just piling up design wins in robotics and other places and really look for more of that in 2027. The growth in the quarter was fairly broad-based, but I think you start to see a lot of these other design wins come to turn into revenue in the future. So I think that in kind of consumer and industrial, 2026, I think, really does still remain mostly about winning all the designs that support future growth.

Operator

Our next question comes from Joe Quatrochi with Wells Fargo.

Joseph Quatrochi

My congrats as well, Tony. Maybe just kind of a finer point on the expansion of capacity to $8 billion. I can appreciate that you can't grow OpEx as fast and you don't need to grow as fast. But how do we think about like maybe some of the investments you have to make on the working capital side or go-to-market side to support that?

Tony Balow

Yes. I think on -- one of the nice things about being in a fabless model, right, is we don't have to dig a hole and pour concrete into it.

Michael R. Hsing

We don't make it and it will wake our brain generally.

Tony Balow

That's right. So we -- I don't anticipate at this point seeing a large increase in capital intensity just to go ramp this capacity, right? So it gives us a lot of flexibility. As a matter of fact, I am personally just amazed at the amount of capacity we've been able to put on just even over the last year.

And I think the operations team has done a tremendous job. But I don't see it necessarily reflecting the overall OpEx at this point or any kind of change to our capital structure just to put on that extra capacity.

Joseph Quatrochi

Okay. And then maybe as a follow-up, I think auto is continuing to see some inflection. Can you just talk about the breadth of demand you're seeing there from a geographic perspective and how we think about like the design pipeline looking into '27?

Michael R. Hsing

It's of course, geographic in U.S. side, Europe side and also in Asia, China. We see -- in China, particularly, we see a lot of activities. And so that's probably the world. And we -- as we said -- as I said earlier, we can't predict which market segment. And that's our diversified growth strategy.

And some years, one segment is good, even with a mature market and some years, one market segment is good and the other one is not so good, okay? And in the last few years, all of a sudden, the GPU market happened or the optical market and the revenue happened. And so we didn't anticipate it. We just provide the best highest performance product. Our customers will realize it.

Tony Balow

Yes. The only thing I'd add is I think the reason we've been pretty confident on the ramp in the second half is because it is so diverse amongst customers and regions and not dependent just on one. You've seen that same level of customer diversity grow in auto over the past several years, like we've been talking about in enterprise data. And the pipeline is similar, right? It's very geographically diverse and very customer diverse. We were just very excited to start seeing some of the architectural transitions that we've been talking about in sort of 48-volt and zonal because those sort of form the next wave of growth for us.

Michael R. Hsing

We're not predicting what the market is.

Operator

Our last question comes from Gary Mobley with StoneX.

Gary Mobley

Congrats Tony, again. I wanted to ask about what you might be signaling in your capacity milestones. You're currently at $6 billion in production capacity, which I assume supports your 2027 outlook. It doesn't take much to get there, just extrapolating out Q4 revenue guidance. And you're expected to be at $8 billion in capacity exiting fiscal year '27. So should we think about that in terms of the revenue velocity exiting fiscal year '27?

Michael R. Hsing

It doesn't take much to get there, oh, come on, this is the market that move the numbers. There's a lot of work behind it.

Tony Balow

Yes. I'll add, so right now, I think what we're trying to strongly signal is that just like we've been doing in the past 20 years, we're making sure that capacity stays ahead of the business. And in a world where there's a lot of narrative around there being tightness, we wanted to be very clear to you and our customers how we're growing capacity to support that business.

So I think, again, it's too early to go right into a quarterly revenue ramp and what that means for 2027. But we wanted to make sure it was loud and clear was that we see the demand coming, and we're going to make sure capacity is there to deal with it.

Gary Mobley

And on inventory, I just want to confirm that the distribution inventory is still within that normal 5- to 8-week range. And then it looks like your internal inventory is down to 117 days based on a forward revenue guide basis, and that's been trickling down. Are you inventoried properly to support sales, I guess, is really the question.

Tony Balow

I think right now, we're still on the channel side, certainly in the lean end of our range like we've been for some time now. And internally, I think we grew dollars on a gross basis, right? But obviously, as the revenues ramp, that's come down from a days perspective. So I think that our range certainly hasn't changed, and it's something that as we increase our capacity and our wafer starts, we'll make sure that we're positioned for the business. But right now, it's not limiting us on the business.

Operator

This concludes the question-and-answer session. I would now like to turn it back to Tony Balow for closing remarks.

Tony Balow

All right. Thank you, operator, and thank you all for joining us for this conference call today. I look forward to speaking with you on the next call for the fourth quarter and full year '26 revenue guidance. Thank you again, and have a great day.

Operator

This concludes today's conference call. Thank you for participating. You may now disconnect.