泰瑞達 (TER) 2026 年第三季法說會：AI 驅動營收成長 70%

重點摘要

• 泰瑞達 (Teradyne) 2026 年第三季營收突破 13 億美元，年增 70%，non-GAAP 每股盈餘 (EPS) 大增逾 150% 至 2.29 美元。這兩項指標均超出公司財測的高標。
• AI 驅動的營收佔總營收 60% 以上。半導體測試、產品測試和機器人部門皆實現大幅年增。
• 半導體測試營收突破 10 億美元，年增 77%。記憶體測試營收創下 3.01 億美元的新高紀錄，而系統整合測試 (IST) 營收也以 9,500 萬美元刷新紀錄。
• 產品測試營收成長 56% 至 1.38 億美元，訂單出貨比 (B/B ratio) 為 1.4 倍。機器人部門營收成長 28% 至 9,600 萬美元，該部門並實現單季轉虧為盈。
• 展望 2026 年第四季，管理層預期營收介於 12.5 億至 13.5 億美元之間，non-GAAP EPS 介於 1.95 至 2.25 美元之間。
• 在積壓訂單、記憶體需求、運算市佔率提升以及新增超大規模雲端業者 (hyperscaler) 計畫的支持下，管理層預計 2027 年將保持健全成長。公司預計 2027 年各季度營收將逐季成長，儘管出貨時程可能會帶來一些波動。

重要財務數據

業務與營運表現

半導體測試

AI 運算仍是主要成長驅動力。隨著客戶持續實施雙供應商測試策略，運算營收年增近 50%。泰瑞達表示，其首家超大規模雲端業者客戶已進入成熟的雙源採購階段，預計 2027 年年中將迎來另一波大幅量產。第二家超大規模雲端業者已進入快速跟隨者階段，預計將於 2027 年下半年進行產能採購，而第三家則處於初期認證階段。

汽車和工業營收年增近 90%，主因是與 AI 資料中心建設相關的電源管理需求帶動。行動裝置業務自低基期大幅成長 60%，但仍低於歷史水準。

泰瑞達擴充了 UltraFLEXplus 產品組合，推出專為下一代 AI 和資料中心設備設計的儀器，包括 UltraPin5000-EM、UltraVS64-HP 和 UltraPort-PCIe6。公司還推出了用於光學測試的 Iris 100 以及全新 Titan HP 預燒 (burn-in) 測試功能。

記憶體與儲存

受 HBM 和 DRAM 測試需求以及 NAND 復甦的支撐，記憶體業務創下單季 3.01 億美元的紀錄。管理層預計 HBM 將成為 2027 年記憶體成長的最大貢獻者，同時看好 HBM、傳統 DRAM 和快閃記憶體全面成長。

公司表示，嚴峻的記憶體供應限制正在推動部分 AI 加速器轉向較低的 HBM 堆疊高度。管理層預計這一趨勢將增加需要進行效能測試的 HBM 堆疊數量，而泰瑞達在此領域的市佔率高於 DRAM 晶圓測試 (wafer sort)。

由於三大主要供應商對 AI 相關 HDD 儲存需求持續強勁，IST 營收達 9,500 萬美元。泰瑞達表示，2026 年全球 HDD 容量 (exabyte) 年成長率超過 20%，同時新增的第三家主要客戶也支撐了測試需求。

產品測試與機器人

產品測試營收成長 56% 至 1.38 億美元。成長廣泛來自於生產電路板測試、光學測試和高速互連應用。合資企業 MultiLane Test Products 對 TDR 和 BERT 設備強勁的需求，推動整體產品測試部門實現近 30% 的季增長。

機器人部門創下歷史最高的第三季營收並實現盈利。隨著製造商採用機器人進行資料中心托盤組裝與測試，電子和半導體已成為該部門最大的垂直領域。泰瑞達在本季推出了優傲機器人 (Universal Robots) Gen 7 協同機器人平台。

管理層預計 2027 年與研究實驗室合作的實體 AI (physical AI) 活動將更加活躍，但指出更廣泛的終端市場成長更有可能在 2028 年變得顯著。來自機器人部門最大客戶的營收從 2025 年到 2026 年成長了兩倍，預計 2027 年將進一步成長，且營收比重將更偏向下半年。

管理層財務預測

13 億美元的中位數意味著約 20% 的年增長率。若達成目標，2026 全年營收將達到約 52 億美元，增幅超過 60%。

管理層預計 2027 年第一季營收將高於 2026 年第四季，並預計 2027 年將實現逐季成長。然而管理層提醒，季度業績可能會隨運算訂單時程以及客戶組裝完整測試單元 (test cell) 的進度而有所變化。

泰瑞達計劃在 2026 年投資近 10 億美元用於研發和資本支出。公司亦正在擴充 SoC、記憶體與電源測試平台的產能，以支援預期需求。

風險與關注焦點

• 主要超大規模雲端業者的運算計畫預計將集中在 2027 年下半年，這可能帶來季度時程上的波動。
• 零部件交貨期拉長，需要增加庫存並提前採購。管理層將約一半的庫存累積歸因於因應需求準備，另一半歸因於零部件緊缺與正常化。
• 行動裝置業務仍低於歷史營收水準，且邊緣 AI 驅動的復甦時機仍具不確定性。
• 共同封裝光學 (CPO) 的採用速度取決於經濟性、測試強度以及尚未解決的組裝流程挑戰。管理層維持到 2028 年 CPO 測試設備市場規模為 3 億至 7 億美元的預估，但表示產能提升的時機仍不確定。
• 除該部門最大客戶外，機器人業務的可見度仍然較短。更廣泛的實體 AI 部署需要實驗室、OEM 和代工製造商的開發工作。
• 成本通膨、供應限制和額外的產能投資可能會影響毛利率，而產品組合和新品上市預計將使第四季毛利率呈季減趨勢。

分析師問答亮點

記憶體展望：管理層預計第四季記憶體營收將維持強勁，儘管在經歷客戶快速採購期後，第三季預計仍為 2026 年的高峰。在 HBM 帶動下，預計 2027 年記憶體營收將實現顯著的年成長。

2027 年營收節奏：與 2026 年初約 13 週的可見度相比，泰瑞達目前的訂單可見度已延長至數個季度。管理層預計 2026 年第四季至 2027 年第一季將實現成長，且 2027 年全年將逐季成長，其中大型超大規模雲端業者和機器人計畫的比重更偏向下半年。

產能與交貨期：UltraFLEXplus 的產能約為 18 個月前的四倍，目前正計劃進行另一輪擴產。泰瑞達已核准該平台的第二家代工製造商，並正針對記憶體測試系統採取類似策略。

運算市佔率提升：管理層表示，透過雙供應商計畫，泰瑞達在 2026 年運算領域取得實質進展，儘管其整體市佔率仍受到初始基期較低且市場快速擴張的影響。管理層預計 2027 年超大規模雲端業者計畫將有良好表現，而 2028 年將更加強勁。

長期模型：公司計劃在第四季法說會上更新其目標獲利模型。管理層表示，先前的模型基於 120 億至 140 億美元的自動測試設備 (ATE) 市場規模，但該市場已大幅擴張，且公司的營收結構已進一步轉向半導體測試。

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

Ladies and gentlemen, good morning, and welcome to the Teradyne Third Quarter 2026 Earnings Conference Call. [Operator Instructions] As a reminder, today's call is being recorded.

I would now like to turn the call over to Amy McAndrews, Vice President of Corporate Relations for Teradyne. Please go ahead.

Amy McAndrews

Thank you, operator. Good morning, everyone, and welcome to our discussion of Teradyne's most recent financial results. I'm joined this morning by our CEO, Greg Smith; and our CFO, Michelle Turner. Following our opening remarks, we'll provide details of our performance for the third quarter of 2026 and our outlook for the fourth quarter.

The press release containing our third quarter results was issued last evening. We are providing slides as well as a copy of these prepared remarks on the Teradyne Investor website that may be helpful in following the discussion. Replay of this call will be available via the same page after the call ends.

The matters that we discuss today will include forward-looking statements that involve risks that could cause Teradyne's results to differ materially from management's current expectations. We caution listeners not to place undue reliance on any forward-looking statements included in this presentation. We encourage you to review the safe harbor statement contained in the slides accompanying this presentation as well as the risk factors described in our annual report on Form 10-K for the fiscal year ended December 31, 2025, on file with the SEC. Additionally, these forward-looking statements are made only as of today.

During today's call, we will refer to non-GAAP financial measures. We have posted additional information concerning these non-GAAP financial measures, including reconciliation to the most directly comparable GAAP financial measures, where available, on our investor website.

Looking ahead between now and our next earnings call, Teradyne expects to participate in a technology-focused investor conference hosted by UBS. Our quiet period will begin at the close of business on December 18, 2026.

Following Greg and Michelle's comments this morning, we'll open up the call for questions. This call is scheduled for 1 hour.

Greg?

Gregory Smith

Thanks, Amy. Good morning. AI was the driver again this quarter. Total company revenue topped $1.3 billion, up 70% year-over-year and consistent quarter-over-quarter, with non-GAAP EPS of $2.29 up over 150% year-over-year. All 3 of our business groups: Semi Test, Product Test and Robotics grew versus the prior year, highlighting AI demand across all parts of the business. AI-driven revenue represents more than 60% of total revenue, a key proof point that our wafer to data center strategy is delivering results. From wafers to modules, to trays, to racks and interfill data centers, Teradyne equipment is used to build and test the high-performance computing that underpins the growth of agentic AI.

We've previously discussed our long-term perspective on the AI transformation of the semiconductor industry and global economy. The first wave started back in 2024 with the build-out of general-purpose data centers which were used for both model training and inference. In 2026, we are in the early stages of the second wave, where data centers are optimized to serve these models to support agentic AI and inference on a global scale. The third wave is physical and edge AI. While there are already compelling examples of physical and edge AI in automotive, robotics and wearables, the full impact of AI on these segments is still to come. The strength of these ways gives us great confidence in our midterm growth plans.

Let's start with Semiconductor Test. The demand for compute is far greater than supply, and the semiconductor industry is seeing dramatic expansion in capacity to meet this burgeoning demand. The portion of semi CapEx associated with test is increasing as well, from 6.4% in 2025 to 8.1% for the first 8 months of 2026. With forecasts for semiconductor CapEx being revised upward for '27, '28 and beyond, we believe that the conditions are right for prolonged expansion of the ATE TAM.

CPUs and AI accelerators are key drivers of the expansion. Through our accelerated investments, Teradyne is positioned for continued growth in both the overall ATE market and share gains, driven by customers adopting dual vendor strategies. Dual vendor provides greater assurance of getting capacity when needed, minimizes the risk of supply chain disruptions and ensures that the test capability in -- that the test capability arrives in time for new device technologies.

Dual vendor programs progress through several phases over multiple quarters, including initial qualification, fast follower and mature dual sourcing. We have multiple customers at different stages of this process. Our first hyperscaler customer is in the mature phase, and we expect our next major ramp in the middle of 2027. Our merchant GPU customer is in the fast follower phase with multiple devices in development, and we expect to see additional revenue in 2027. Our second hyperscaler customer has passed qualification and is now in fast follower, with an expectation of capacity buys in the second half of 2027. We also have a third hyperscaler in initial qualification.

Networking devices, both copper and optical, are growing proportionately with compute. The majority of this TAM is in switches. However, retimers and DSPs are driving significant business. The majority of the networking market is currently copper and optical pluggables. Over the next few years, we believe the NPL and CPL markets will grow strongly from a small base.

We are working with leaders in this segment on solutions across all test insertions, and it is clear that there are many different approaches and technologies being tried by different customers. As a result, the types of tests in the handling and alignment techniques are in flux. The key to our strategy here is to develop a strong open ecosystem -- to develop strong open ecosystem partnerships for instrumentation, handling and alignment. We have spoken before about our partnerships with Technoprobe and FiconTEC, and we have just announced a partnership with FormFactor to address optical test amongst other opportunities. Our open nonexclusive ecosystem of probing, handling and interface partners allows us to validate complete test cell solutions before they reach our customers' floors, giving them more interoperability, full control over their supply chains and lead times. This provides a faster path to high-volume production.

The pace of innovation in compute and networking has accelerated, and we have been investing in R&D to ensure that our product road map is aligned with those needs. We've discussed before that there is no current ATE solution for the most complex next-generation AI and data center devices that are coming to market in 2027.

The trio of new UltraFLEXplus instruments that we just introduced address that need. The first is the UltraPin5000-EM. The EM stands for extended memory. This instrument has up to 80x deeper memories than our current instrument for intensive device characterization work. The second instrument is the UltraVS64-HP. The HP stands for high power. This instrument enables testing accelerators that consume up to 15,000 amps. And finally, the UltraPort-PCIe6. This is the industry's first ATE solution for PCIe Gen 6. We also introduced the Iris 100, a solution for optical test of MicroLED and VCSEL emitters. This instrument serves both mobile applications and optical networking.

To assure data center reliability and uptime, customers are increasingly turning to system-level test and burn-in. In the third quarter, we introduced new burn-in capabilities for the Titan HP platform, which enables full system test coverage across burning temperature ranges.

As AI moves out in the data center and into cars, robots and wearables as part of the third wave of AI, mobile processors and microcontrollers are adding neuroprocessing units and other AI features. To address that coming market, in the third quarter, we launched the UltraFLEXplus S, a small footprint system that covers both engineering and production insertion for edge AI devices.

Moving on now to memory. Memory manufacturers are bringing on more capacity for HBM, DDR, LP-DDR and flash memories. More capacity means more wafers, more wafers means more bits and more units, which means -- which means more to test in memory.

In the past 2 quarters, it has become clear that the constraints on memory supply are so severe that AI accelerators are turning to lower stack heights of HBM memory to get a larger number of stacks out of the same volume of DRAM wafers. This will tend to increase the TAM for HBM performance test, an area where we have higher share. The HBM, DDR, LP-DDR and GDR markets are moving quickly from generation to generation, with higher data rates and new encoding approaches. To support these new technologies in the third quarter, we introduced the Magnum E2, our next-generation DRAM tester, with the signal integrity and throughput needed to keep pace with our memory customers.

Linear and power segments, which were previously driven mainly by automotive and industrial verticals are adding capacity to deliver data center power distribution and regulation components. Our product revenue in the automotive and industrial space has grown from less than $400 million in 2025 to over $600 million in 2026. This trend will accelerate with new 800 volt standards and new technologies like IVR.

The demand for AI accelerated compute goes beyond the semiconductor market. The need for nearline storage and data centers has accelerated the annual exabyte growth for HDD to over 20% in 2026. Since the test time for HDD is more closely related to the capacity than the unit volume, this as well as adding a third major customer, is driving expansion of our HDD test revenue.

Scale-up networking to support large GPU clusters is predominantly served by copper connections via backplanes and cable spines, running at unprecedented data rates. This has turned backplanes and cables from commodities to highly engineered and tested elements of the HPC system, which is driving significant growth in TDR and BERT test equipment served by our MLTP joint venture. In addition, we have seen strong demand for our Omnyx board test products. Together, these contributed to almost 30% quarter-over-quarter growth for the entire product test group.

Increasingly, exotic packaging and materials are being used in data center tray assemblies. These trades demand superhuman accuracy and stamina to build. ODMs and CMs are increasingly turning to flexible, physical AI-based robots to assist in assembly and test processes. As a result, electronics and semiconductor is now our largest vertical for robotics. In the third quarter, we introduced the Universal Robots Gen 7 cobot platform. The features of this platform and PolyScope X have been tailored precisely to bring physical AI to electronics assembly work cells and provide a platform for UR+ partners to plug into to support the diverse requirements of server and rack build.

To sum up, our wafer to AI data center strategy is delivering. Within the third quarter, outside of compute, the rest of our portfolio delivered close to 80% of our overall sales. The R&D that we have been investing has yielded next-generation products. These products allow us to capture opportunity along the entire value chain, from wafers to modules to trays to racks. Our continued investment reflects our optimism about the near and medium-term opportunities in front of us.

This was another strong quarter where AI was the driver across our business groups. We are riding a rising tide. AI has driven the semiconductor and electronics industry into a new era, and our wafer to data center strategy has positioned us to benefit from these positive changes.

With that, I'll turn the call over to Michelle.

Michelle Turner

Thanks, Greg, and good morning, everyone. Let's get into the quarter beginning with the total company. Both revenue and non-GAAP EPS came in above the high end of our guidance range as our wafer to data center strategy and the breadth of our portfolio sustained positive momentum. For the third consecutive quarter, total company sales remained at the high watermark of $1.3 billion. For the first 9 months of 2026, we delivered $3.9 billion in revenue and $7.31 of non-GAAP EPS, up over 85% and 235% year-over-year, respectively, driven by all things AI.

Building on that, let's take a deeper look at revenue starting with Semi Test. Our Semi Test team once again cleared $1 billion in revenue, up 77% from Q3 2025. The revenue breakdown within Semi Test was SOC at $676 million, memory at $301 million and IST at $95 million. As expected, compute order timing was predominantly offset by strength across the rest of the Semi Test portfolio.

Starting with SOC, compute revenue grew nearly 50% year-over-year on continued AI-related demand. As Greg mentioned, our customers' dual vendor strategies are opening up more opportunities for us in, and we expect healthy growth in 2027. Auto and industrial continued to strengthen, up nearly 90% year-over-year, driven primarily by power management demand increases for AI data center build-outs. Mobile was up 60% year-over-year from a low base, but remains below historical levels and a muted part of the overall SOC portfolio.

Now turning to memory. Our memory business delivered another record quarter at $301 million in revenue, up from the previous record set just last quarter. This represents our fourth consecutive quarter of revenue over $200 million, driven by continued robust HBM and DRAM test solutions demand and the resurgence in NAND.

Finally, turning to IST. Revenue in the quarter reached a new record at $95 million, up approximately 150% from the prior year, with continued strength in AI-related HDD storage demand from all 3 major suppliers in this space.

Now on to the product test group. Revenue was $138 million, up 56% year-over-year. The group experienced broad-based growth tied to AI infrastructure build-outs from production board test to optical to high-speed [ interconnect ]. Book-to-bill for the quarter was 1.4x, with particular strength in the newest addition to the group, MultiLane Test Product JV, as the need for data center interconnect test solutions increases.

Robotics revenue was $96 million, up 28% year-over-year. Our Robotics group delivered its highest third quarter revenue ever and also achieved profitability within the quarter. AI-related revenue remains elevated with robot-assisted test and assembly across the data center value chain and high demand. In the first 9 months of the year, AI-related revenue was roughly 20%. Aligned with the third wave of AI Greg mentioned, the physical and edge AI ecosystem is in early innings and building momentum. Our Robotics group has announced partnerships with industry leaders such as NVIDIA and Analog Devices as well as innovative start-ups, including Scale AI, Generalist, Bright Machines and Sereact. These partnerships reflect growing interest in physical AI.

Now moving down the P&L. Gross margins for the quarter were 59.1%, up 60 basis points year-over-year, driven by strong volume and product mix, while sequentially, gross margins were down 70 basis points, driven primarily by product mix. As expected, OpEx increased, driven by more R&D and go-to-market investment in support of future growth. Finally, non-GAAP operating income was $413 million, with an operating margin of 31.6%.

Now moving to capital allocation. Reinvesting in the business remains our top priority. In 2026, we will invest close to $1 billion across R&D and CapEx, with R&D investments representing a double-digit percentage of our sales. Beyond that, our strategy is to maintain the cash reserves to grow the business, return capital to shareholders through both dividends and share repurchases and keep dry powder available for accretive M&A. Each pillar of our capital allocation strategy is focused on innovating from wafer to data center, positioning us well over the near and midterm.

We ended the quarter with cash and investments of $625 million, up 21% from last quarter. We had free cash flow of $227 million in the quarter. For the first 9 months of 2026, free cash flow was $806 million, up nearly 250% from the prior year period. Additionally, from a working capital perspective, we increased our inventory in support of future revenue growth. We paid $20 million in dividends in the quarter, and our share buybacks were $99 million.

Now looking ahead to our fourth quarter guidance. For the quarter, we expect revenue in the range of $1.25 billion to $1.35 billion and non-GAAP EPS of $1.95 to $2.25. Gross margins are expected to be in the range of 57.5% to 58.5%, reflective of product mix and new product launches. Operating expenses are expected to run at 28.5% to 29.5% of fourth quarter sales, driven by continued investments in R&D and go-to-market. The non-GAAP operating profit rate is expected to be between 28% and 30%.

Our Q4 guidance midpoint of $1.3 billion represents growth of approximately 20% over the last year's fourth quarter, which would bring full year revenue to approximately $5.2 billion, up over 60%. We expect Q4 will be our fourth consecutive quarter of revenue above $1.25 billion, representing a sustained step change in the business from a year ago. As we said last quarter, we expect 2027 to be another year of healthy growth. Our strengthening backlog and the growth in compute give us momentum, and we expect to carry and build throughout the year.

For your modeling purposes, we expect FormFactor investment to close in the fourth quarter, subject to regulatory approval and customary closing conditions.

The rest of our portfolio is in full effect in the second half as memory, auto and industrial, IST, Product Test and Robotics grow, offsetting a second half program timing in compute.

In closing, Q3 was another quarter of strong execution. AI demand continued to pull through every part of our portfolio, from the chips that power it to the systems, interconnects and robots that build it. This is what our wafer to data center strategy is designed to capture. As we look to the fourth quarter and into 2027, the demand signals across our end markets remain strong, and we look forward to sharing an updated target earnings model on our Q4 call.

I'll close by thanking all of our Teradyne team members around the world for the hard work and dedication that drives our results and positions us well for the opportunities ahead.

With that, operator, we're ready for questions.

Operator

[Operator Instructions] We'll take our first question. from C.J. Muse with Cantor Fitzgerald.

分析師問答

Christopher Muse

I guess first question on memory. You obviously put up a record number for September. I'm curious how we should think about the trajectory of that business into December? And then moreover, as you think bigger picture into '27, I know it's early, you're going to save that for February, your outlook. But as you sit here today and as you look at your book-to-bill, how are you thinking about maybe the rank order of growth between HBM, traditional DRAM and NAND?

Gregory Smith

C.J., it's Greg. So looking out to Q4, we think memory is going to continue at a really strong level. And going into '27, given all of the fab expansion that is coming online in '27, we're expecting significant growth for the memory business year-on-year. And you were asking in terms of like what segment is going to drive most of the growth. We would say that like this year, HBM was probably the biggest thing that drove the growth of the memory TAM. Flash came back pretty strong, and DRAM was surprisingly strong based on what we were thinking at the beginning of the year. Looking at next year, I think HBM is the start. That -- there's going to be growth across all of the segments of memory, but I think the biggest impact is going to be in HBM growth.

Christopher Muse

Very helpful. And then maybe on your gross margin guide and the strength in the 1.4 book-to-bill you highlighted for Product Test. How should we think about Product Test sequentially? I'm assuming that's up. And is that what's driving gross margins lower? And as you think about that business next year, what impact does that have to what we should be thinking about as a more normalized gross margin for Teradyne?

Michelle Turner

Yes. So I'll start with the gross margins, and then we can talk a little bit about Product Test. So within the quarter, gross margins came in at 51.1%. This is up 60 basis points from the Q3 last year. Within the Q4 guide, we're also guiding up about 80 basis points in compared to the last year. And then if you extrapolate that for full year, our overall gross margins for the year will be about 59.4%, which is up about 110 basis points year-over-year. So at the low end of the earnings model that we just put out in January, and we're going to expect to update that in the Q4 call. So thinking about the 3 segments, C.J., Product Test is actually accretive to the overall portfolio. So as you think about modeling for Q4 and into next year, we expect that to continue to be true.

As you think about gross margins for 2027, I put it in a couple of buckets in terms of headwinds and tailwinds. From an overall headwinds perspective, I would say, volume, we're going to have another year of healthy growth when you think about the overall portfolio, along with some elements of mix as we think about next year. Tailwind, though, from an overall cost perspective, we are continuing to see cost pressures, inflationary pressures just across the overall ecosystem across supply, along with potential expanded capacity next year as well. And so as we look at second half of next year, we're continuing to see robust demand signal. So to the extent that we're continuing to build on our capacity, that will also be an expense that we would expect next year as well.

I don't know if you want to add anything else?

Gregory Smith

You missed [indiscernible]. You said 51.1%. I think you meant [ 59.1% ]. Just want to make sure that, that doesn't stay out in the atmosphere. But I completely agree with what you said that there are kind of, I'd say, equal up and down forces in terms of gross margin going into 2027.

Michelle Turner

Yes. And the only other thing I would add, C.J., in terms of Q4 and comparative to Q3, and Greg referenced kind of the continued growth in memory. And we certainly expect that. We do expect though, that Q3 was the record quarter within 2026. So we think there's a bit of a digestion period, although it's going to continue to grow from Q4 of last year. We expect that compute was at its lowest within Q3. And so we expect that, that's going to start to reaccelerate from a growth perspective in Q4. And then to your point, we also expect Product Test and Robotics to grow within fourth quarter as well.

Operator

We'll move on now to Timothy Arcuri with UBS.

Timothy Arcuri

You're in kind of the shape of the year. So this year kind of played out, you got this big step-up in Q1, and then it was basically flat throughout the year, Greg. Inventory though, is up like $200 million over the past 6 months. So obviously, you're portending a pretty big year next year. Can you speak to that?

And I guess, particularly, it sounds like both your first and your second hyperscale customers are a little more back-end loaded next year. So does that mean like next year is going to be a back-end loaded year? Should we think of the first half as kind of flat versus the back half of this year and then a big step up in second half of next year? I ask because people are sort of thinking that it's more of a ratable ramp for the year. So if you can help us shape that, that would be great.

Gregory Smith

So I think there are 2 things that are driving the increase in inventory that you're seeing. One is, you're right, we are preparing for robust growth in 2027. The other is that many of the components that we're procuring to build these testers, the lead times for those have gone out as well. So we are needing to essentially put on order and take inventory against a sort of a longer lead time of our inputs. So there's sort of 2 things going on there. I don't think it's a good idea to just sort of extrapolate from that number to a pure growth number.

In terms of the -- how next year breaks down, we'll have a lot more to say about that in our Q4 call. But I will tell you that in terms of the [indiscernible] programs, I think you're right. I think they are going to tend to be more back half loaded. But I don't particularly expect -- I think it's going to be a strong year all around, and we'll talk more about the phasing next quarter.

Timothy Arcuri

Okay. And I guess just, Greg, on the share gain pace, so last night, your competitor was kind of -- I mean, they keep pushing back on losing any share. And in the release last night, they -- there was some allusions to the contract manufacturing partners adding a bunch of supply for them. So how do you see your lead times affecting the pace of the share gains? Like are you being proactive enough about adding? You're having your supply chain add enough for you to keep lead times when you need them so that your customers don't just go to them and place orders with them.

Gregory Smith

Yes. So we've talked a lot about this in prior calls, that we've added significant capacity already for our primary SOC test platform, the UltraFLEXplus. We're actually at like 4x the capacity that we were about 18 months ago, and we are essentially planning our next capacity step-up for that platform. In addition, we've employed a dual member strategy for our own supply chain. So in 2026, we qualified the second CM to build that platform. We're in the process of doing that for our memory test platform as well. And we are aggressively adding capacity for both memory and our [ Eagle ] power test line. So where we are definitely looking to increase our capacity, we are doing our best to try to hold our lead times.

One thing that is a little bit different is we're holding our lead times, but customers are giving us probably better forecast information about when they are going to need capacity in advance of placing those orders. So we have a pretty good viewpoint going out further than we did before, and that's helping us map out exactly what we need to do for capacity planning. We're doing our best to try and hold our lead times. I would say that if demand comes in that we haven't heard anything about from a forecasting perspective, the drop in place for that is probably longer than it was a year ago. But we're doing our best in terms of procuring material and expanding capacity so that we can still be very responsive to spot demand.

Operator

We'll move next to Mehdi Hosseini with SIG.

Mehdi Hosseini

Two follow-ups for me. Greg, I want to go back and see if you can qualitatively give us an update on the demand vectors. I think the targeted model is now [ stalled ]. And yes, we're going to have to wait till Q4 earnings to get an update. But I think it will be very helpful if you can shed some light into different demand vectors. It seems like everything is tracking better than expected, and next year could be second half weighted. But is there anything else you can provide because I just want to make sure that consensus is already well above your target and we're not being too aggressive. And another follow-up.

Gregory Smith

Yes. No. So I think we can definitely -- we're going to give you an updated model in January. And the last model was based around a $13 billion ATE TAM, $12 billion to $14 billion. And the TAM has grown very, very quickly. And when we were building that model, we were trying to give you an idea of what a larger Teradyne would look like without tying down sort of exactly when that would happen because of uncertainty about these growth rates.

Having said that, there are multiple independent parts of Teradyne that are each growing associated with their markets and our success there. And we had to make assumptions in terms of the mix. That in our model, we assume that Semiconductor Test would be 80% of the revenue and Robotics would be 10% and Product Test would be 10%. Semiconductors are growing so fast that, that mix has changed. Also, the Semiconductor TAM growth has been so concentrated in compute, an area where we have low but growing share that essentially, we need more ATE TAM in order to hit the revenue that we had in the model that we published. So as we're looking to put together the model that we'll talk about in January, we're definitely sort of retiming all of the different groups of the company. But if you sort of run through it, compute is growing faster than we anticipated, and we're gaining share there at a modest rate because that's being driven by dual vendor strategies primarily from those customers and growth in networking.

I would say that mobile is underperforming our expectations from the model as we put it together, and we're going to be relooking at that to see whether and when we would see an edge AI-driven recovery in that space. How are the -- the automotive and industrial space, we're seeing good growth there. And I think that's tracking to sort of the model that we had before. And both Product Test and Robotics are tracking the growth that we had expected in our model, but they've been sort of -- the Semiconductor growth has sort of outpaced it. So when we put it all together, I would expect to see us building an evergreen model against a much larger ATE TAM, and we'll talk about that in January. But the growth vectors are all there, and it's all positive in terms of the direction that we're going.

Mehdi Hosseini

And follow-up, your industrial automation for me is basically physical AI is being there. And also build up manufacturing capacity in the U.S. What needs to happen for a higher growth rate? Is it just a matter of timing? Or do you have to make more modification to the model? And anything there that could help you with a higher growth rate?

Gregory Smith

Yes. So it's a great question. So with the growth of physical AI, the capabilities of physical AI, we think that, that is going to unlock significant growth in the Robotics group. The -- there are a couple of things that have to happen. Over this year, and Michelle talked a little bit about it in her script, we have been building partnerships with the labs that are working on physical AI models and physical AI implementations. We are also aligning with the CMs and OEMs that are serving primarily the AI data center build-out. They're doing server builds basically. These things are really tough to build, and the customers are quite sophisticated in terms of the adoption of technology. That is kind of fertile ground to be able to deploy robots.

The thing that I will warn you is that because we are working with the labs and then the labs are working to solve the problems with the end customers, it's going to have a gestation period. So I would expect us to see robust growth in terms of the transactions that we have to the labs in '27 and the beginning of end market growth. But I think it's probably more of a 2028 thing than a 2027 thing beyond our sort of major customer in Robotics.

Operator

We'll move on now to Jim Schneider with Goldman Sachs.

James Schneider

I was wondering if you could maybe comment on your sort of near-term visibility heading in the next several months? And specifically, I think you've talked about visibility kind of extending beyond the current quarter. So I was wondering if you can maybe, in terms of the longer-dated commentary, help us a little bit, at least directionally, with your view on Q1 preliminarily directionally sort of up, flat or down and maybe what would sort of be driving that beyond the memory drivers you talked about earlier?

Michelle Turner

Jim, it's Michelle. So from a visibility perspective, we talked about this over the last couple of quarters. So coming into the year, historically, we've had about 13 weeks of visibility. Fast forward to where we are today, we have a pretty good line of sight out a couple of quarters. So where we're sitting as we exit '26 is really strong backlog. So we have pretty good line of sight to what will play out in Q1. The caution we have around putting any guidance around it right now is we do continue to see movement within any given quarter depending on how test cells are coming together for our customers. So I would say we would expect Q1 from where we sit today to be some level of growth, but we expect the year to progressively grow throughout quarter-on-quarter versus what we saw this year, which was more of a 50-50 split.

James Schneider

That's very helpful color. And then maybe as a follow-up, I was just wondering if you could sort of give us a view on the Robotics business very broadly speaking. Obviously, the momentum there is improving. And I know that you have some level of visibility on some very large customers potentially ramping next year. So maybe if you could give us any update on like sort of how your visibility in Robotics compares to the broader visibility of the business you just discussed?

Gregory Smith

Yes. So visibility in Robotics, aside from our largest customer, is pretty short term. I mean we have, coming into a quarter, a large portion of our revenue is coming from [indiscernible]. So we manage that business against sort of a pipeline of opportunities. I'll tell you that the pipeline of opportunities is expanding and our closure rates are relatively consistent, but it is essentially probabilistic. Where in the Semiconductor Test space, we're kind of counting those since we know how many testers people will need when they'll need them.

In terms of the Robotics business, a big chunk of it is essentially driven by an opportunity pipeline. Our large customer, we triple our revenue for that large customer from 2025 to 2026. We expect to see robust growth next year. The way that -- what's going on with that major customer is that in 2025, they were largely doing engineering work to prove out the solution they were deploying. In 2026, they deployed that solution to their sort of leading edge of distribution centers and have been working out the kinks and getting it ready and also like sort of developing the processes so that they can install these systems quickly so that they can quickly flip over a distribution center from manual to automated in these workflows. And that's the work that we expect to see happening in 2027, sort of a more broad deployment of these technologies, but we'll have to see exactly what pace they're going to be able to do that work.

Michelle Turner

Yes. And just a follow-up on that, Jim. This is an example of where we see more of the back half weighting based on that customer schedule.

Operator

We'll move on now to Joseph Cardoso with JPMorgan.

Joseph Cardoso

Maybe just for the first one, I just wanted to touch on, in your prepared remarks, I think you've made a couple of references around CPO and NPO. And so I was just curious if we can get an update on how you're thinking about the timing? It sounds like maybe anecdotally, there is a little bit of a deemphasis there in terms of how fast these opportunities could ramp relative to maybe what we're hearing across various suppliers in the broader supply chain there? And just curious whether you're thinking about that any differently as well as the content and TAM associated with the opportunity? And then I do have a quick follow-up.

Gregory Smith

Joe, so I think we remain quite enthusiastic about CPO. And when we've talked about it before, we've tried to be pretty clear that we're kind of in the preseason right now that there is an initial customer that is driving towards a significant ramp in 2027, but it is growing from a very small base. We've talked about the TAM in 2026, probably somewhere in the neighborhood of $100 million for the test equipment portion, and that could grow to $300 million to $700 million by 2028. I think that's still the right range.

As to whether we trend towards the high end or the low end of CPO, I think it really depends on how -- like whether the economics of CPO get to the point where it's practical to ramp it hard. Right now, the test times and the test intensity required to do CPO is quite -- it's quite high. And I think that, that is going to be somewhat of a limiter in terms of how broadly it's deployed until those -- it's part -- and test is only part of it. There are other challenges in the whole assembly flow that need to get worked out and refined. So I think as those things get refined, this is going to turn into something big. That the $300 million to $700 million is just sort of a stop along a path of growth. This is likely to continue to be up to $1 billion market or beyond. But the exact timing, how fast we go up that ramp is something that I think is going to be a question mark.

The thing that you have to remember is these network connections have to get made. Whether they get made with CPO or NPO or pluggables or copper, the key thing is that clusters are going to get bigger and connectivity is going to be increasing. And so our strategy is to make sure that we are able to serve all of those markets and be able to serve that from both our product test group for pluggables and copper and from our semi test market for all of the switch silicon and everything else. So we're definitely investing towards an open ecosystem in CPO, but I think we are trying to make sure that our business strategy is resilient to any uncertainty in the timing. That's not like a lack of confidence in the end goal. It's more just knowing that the importance of these connections is so high that they're going to have multiple paths to success.

Joseph Cardoso

Got it now. Appreciate the thoughts there and totally fair.

And maybe on the second question, and I wanted to go back to compute, but maybe touch on it in a different angle. Obviously, you have your kind of disclosed customers that you'll be ramping through 2027. But maybe if we think about more medium term, longer term, I just want to hear your thoughts around potential engagements beyond those 3 customers, particularly just given that we're hearing more around maybe your, like your step-up or inflection around custom basic opportunities going into 2028 with additional hyperscalers and potentially AI labs.

So just curious how you're thinking about your opportunity or the TAM there as you guys are thinking or maybe hearing about these engagements or discussions with customers. And are you seeing maybe a similar trend as we kind of think about 2028 as well? And I know I'm skipping '27 a little bit here, but it does seem like this is a big talking point around maybe an expanding opportunity set here. So just curious how you guys are thinking about that?

Gregory Smith

I think it's awesome that you're thinking about 2028, and it really reflects sort of the new reality in this market that we used to be in a world where we were -- like there was significant uncertainty quarter-to-quarter around how the end market would develop. Now any quarter-to-quarter uncertainty is really just around like shipment timing and how fast people can build buildings. So it's pretty cool that we have the ability to look out to 2028 and have a conversation about long-term targets.

So in our prepared remarks, I talked about hyperscaler 1 is mature. Hyperscaler 2, we're in fast follower. We expect to see stuff in the back half of '27. We're in qualification for a third. There are other x factors here in terms of what will happen with the [ Tera fab ] investment, what will happen with OpenAI, with Anthropic. All of these folks are leaning hard into developing their own their own silicon technology for sort of a complete stack.

I will say that we are approaching all of them in the same way. In some cases, we are the first in supplier and working to maintain and maximize our share. In other cases, we are positioning ourselves as a capable dual vendor option for them that will allow them to secure their supply chain. It's essentially the same engagement process that we've used for the first 3 hyperscalers, it's going to be what we use on subsequent ones. And we think that the same things that are driving those early hyperscalers towards the choices they've made are going to be driving the rest of the market.

So I think it's logical to be thinking about what's going to happen in the future. I will say that our competitor has established a very powerful share position in compute. We are making significant progress in terms of share gains inside of compute in 2026. But the size of the compute market has grown so much that, that has, essentially had a dilutive effect on overall share. So I'm very optimistic about what will happen with hyperscaler programs, but it is definitely something that takes time. So 2027 is going to be a good year for these programs. 2028 is going to be even better.

Operator

We'll move on now to Vivek Arya with Bank of America Securities.

Vivek Arya

I had near term and then more of an intermediate-term question. So on the near term, Michelle, you mentioned that you expect sales to kind of consistently grow over the next several quarters. If I go back to the Q1, the Q4 to Q1 transition in '22, '23, '24, '25, I think Q1 sales were down double digit. But in Q1 '26, they were up because I imagine compute started to become more important than mobility. So I appreciate it's a lumpy business, but I just wanted to make sure we keep everything in mind or anything in mind that could cause a Q1 to be below Q4 levels.

Michelle Turner

So let me clarify. So you're asking about Q4 to Q1. I was referencing sales growing throughout 2027. So we expect the quarters next year to represent growth quarter-on-quarter when you look at next year. And I also expect between Q4 and Q1, that we're also going to have growth. So just to clarify, kind of, and set the baseline in terms of the notion that you quoted.

The other thing I would point out is coming into this year, when we set the 60-40 split, the visibility we had was really at 13 weeks. As the year has progressed, we have gained additional visibility that's helping better inform how we position for Q1 as and as we think about next year.

To your point, though, we do continue to see lumpiness quarter-on-quarter. And so I think that's the caution that you're hearing. We feel really strongly in terms of the demand cycles. And we feel strongly that next year is going to be another year of healthy growth. But there could be variability within quarter. But where we sit today, we expect each quarter next year to grow in comparative to '26.

Gregory Smith

And just like the -- the thing I want to add is, I think we talked about this probably 2 calls ago, maybe 3 calls ago, that the patterns that we saw up through 2024 are over, like that. So I know it's a useful analytic technique to sort of look at historical patterns. There has been a seismic shift in the semiconductor test market that started in 2024 and really took hold in 2025. So prior to that, the mobile -- sort of the mobile consumer cycle was the tick talk of the industry. But after that, it's all around data center buildup schedules and capacity coming online in semiconductor manufacturers that are vastly underserving on market demand. So it is like I just -- I think it's an important thing for us to try to figure out how to give you good directional guidance about what's going on. But part of that is to like be very careful about applying patterns that existed prior to '24 to a market that's twice as big and acting very different.

Michelle Turner

Yes. And I think the other thing I'd call that is it really is driven by compute order timing, right? That's what drove this year. We expect that, that's going to play out next year as well.

Vivek Arya

Very helpful. And on my follow-up. Greg, one for you. You made this very interesting point that memory constraints are pushing some accelerators towards lower HBM stack item. I'm curious to know, is this something that you think might happen? Is it already happening? Will it become prevalent? And I think you also mentioned that your share might be higher when the wafer heights are short. So I would just -- I would appreciate your perspective on what is the sensitivity here? And for the same amount of HBM bid, how much more test insertion or test time and intensity and opportunity does that create for you?

Gregory Smith

So I'm afraid I can't give you hard numbers, but I can give you some directional guidance here. So I think it is pretty locked in that next-generation, like there was a trend towards higher stack heights, like from '24 to '25 into '26, there was this progression that stack heights are going to go 4, 6, 8, up to 12, 16 high. And I feel like that has reversed in 2026. And I would expect to see lower stack heights across a number of different devices.

Now how is that good for us? The way to think about it is that there is -- the bottleneck is the total number of DRAM wafers that can be produced. Those DRAM wafers, if you think about it, you can build twice as many 4 high stacks as you can [ 8 ] high stacks out of the same supply of DRAM wafers. So that essentially helps deconstrain the shipments of AI accelerators that are using those HBM memories.

The other thing is that with a higher stack height, you are essentially trying to go through a -- like you're trying to get more data through a narrower pipe. So you have some data rate advantages from having more DRAM stacks versus higher stacks. The thing that makes it an advantage to us is that our market share in DRAM wafer sort is lower than our market share in HBM performance test. So if the same number of wafers are going into more stacks, that means more stacks need to get tested, which is where we sell more testers. So we'll -- I think it is a positive indicator for what our share will do going forward, but I don't have a hard number for you.

Operator

We'll move on now to Krish Sankar with TD Cowen.

Sreekrishnan Sankarnarayanan

[indiscernible] argument? There are some charters that there might be some lowering of the SPM signal or [indiscernible]. Would that have any impact on test complexity? And along the same path, I'm curious about HBM competition, not on the Japanese competitors, but more from the Korean players like [indiscernible], et cetera. How is that shaping up? And then I have a follow-up.

Gregory Smith

So Krish, I heard part of your question. So if I get it wrong, please feel free to clarify. I think the first part of your question was around HBM [ destacking ]. And honestly, like while we've heard a lot about lower stack heights, I can't say that we have heard a lot from our customers around destacking or how destacking could affect test. So, so far, that is not a factor in the way we're looking at it.

Now in terms of share versus Korean suppliers. Korean suppliers in test for memory are primarily serving the wafer sort part of the business, not the performance test part of the business. And so we tend to have less of an impact. Given the expansion in capacity that's going on right now, I would expect that they're going to see healthy growth. That they're going to get a fair amount of business as the number of DRAM and flash wafers increase.

Sreekrishnan Sankarnarayanan

Got it. Got it. And then a quick follow-up for Michelle. Thanks for looking up on the capacity and your inventory position. I'm just wondering, how much of the inventory build out is actually for true demand versus actually building inventory of components that could be constrained? For example, I think the testers use DDR4 chips and things like that. So I'm just wondering how much of that is inventory build to prevent any constraints down the road which is true demand?

Michelle Turner

Yes. I think it's about 50-50. And the other thing I would conclude to, and I think this is important to go back and look at is when you look at our inventory from a DIO perspective, we had actually reached unusually low levels in Q1. So part of this was we were building really fast, really quickly to meet the demand in the surge that's occurred over the last 6 to 9 months. And so part of this is also a normalization. And I would argue, the 189 that we're at, I still think that we should probably be around 200 when you think at kind of normal levels in comparative to some of our peers. And so I think there's an element of that, that has to be factored in as well and not just extrapolating on the 35% growth within the quarter.

Operator

At this time, we've reached our allotted time for Q&A. I'd be happy to return the call to our host for any closing comments.

Gregory Smith

Thank you, operator. It was another great quarter for us. We're looking forward to Q4 and especially looking forward to updating you all on an updated model in our next earnings call. We'll talk to you then.

Operator

This concludes today's Teradyne Third Quarter 2026 Earnings Call and Webcast. You may disconnect your line at this time, and have a wonderful day.