การประชุมแถลงผลประกอบการไตรมาส 3 ปี 2026 ของ Cohu (COHU): ไปป์ไลน์ HPC แตะ 1 พันล้านดอลลาร์

การแถลงผลประกอบการไตรมาส 3/2026 ของ Cohu เน้นย้ำถึงความต้องการระบบทดสอบประมวลผลประสิทธิภาพสูง (HPC) ที่เร่งตัวขึ้น รายการโอกาสทางธุรกิจของ Eclipse ที่เพิ่มขึ้น และรายได้ประจำที่แข็งแกร่งขึ้น นอกจากนี้ ผู้บริหารยังได้ปรับเพิ่มแนวโน้มปี 2026 และเปิดเผยแผนขยายกำลังการผลิตสำหรับปี 2027

ประเด็นสำคัญ

• รายได้ไตรมาส 3/2026 เพิ่มขึ้น 37% เมื่อเทียบรายปี สู่ระดับ 173 ล้านดอลลาร์ โดยมีกำไรต่อหุ้น Non-GAAP อยู่ที่ 0.43 ดอลลาร์ ขณะที่อัตรากำไร EBITDA ปรับปรุงแล้วแตะระดับ 16.6%
• ยอดจองซื้อทั้งหมดเพิ่มขึ้น 126% เมื่อเทียบรายปี โดยรวมถึงการเติบโต 318% ในยอดจองซื้อระบบ และการเติบโต 36% ในยอดจองซื้อรายได้ประจำ อัตราส่วนยอดจองซื้อต่อยอดส่งมอบ (Book-to-bill) อยู่สูงกว่า 1
• Cohu ได้ขยายรายการโอกาสทางธุรกิจ HPC ต่อปีขึ้นประมาณ 18% เป็นราว 1 พันล้านดอลลาร์ และปรับเพิ่มแนวโน้มรายได้จาก HPC ในปีงบประมาณ 2026 เป็นประมาณ 120 ล้านดอลลาร์
• ผู้บริหารคาดว่าจะส่งมอบชุดจัดการระบบความร้อน (thermal handler) รุ่น Eclipse ในปี 2027 เพิ่มขึ้นประมาณสามเท่าของปี 2026 โดยคาดการณ์การเติบโตของรายได้เบื้องต้นในปี 2027 ไว้ที่ 30% ถึง 40%
• คาดการณ์รายได้ไตรมาส 4 อยู่ที่ประมาณ 178 ล้านดอลลาร์ บวกหรือลบ 7 ล้านดอลลาร์ ซึ่งสะท้อนถึงการเติบโตเมื่อเทียบรายปีที่คาดไว้ที่ 46%
• อัตราการใช้งานระบบทดสอบเซมิคอนดักเตอร์ปรับตัวดีขึ้นสู่ระดับ 82% ณ สิ้นไตรมาส 3 เพิ่มขึ้น 2 จุดเปอร์เซ็นต์เมื่อเทียบรายไตรมาส และเพิ่มขึ้น 7.5 จุดเมื่อเทียบรายปี

ข้อมูลทางการเงินที่สำคัญ

ลูกค้ารายหนึ่งในกลุ่มการประมวลผล (compute) คิดเป็นสัดส่วนมากกว่า 10% ของยอดขายรวมในไตรมาส 3

ผลการดำเนินงานทางธุรกิจและการปฏิบัติการ

การประมวลผลประสิทธิภาพสูง (HPC)

แพลตฟอร์มการจัดการความร้อน Eclipse ยังคงเป็นกลไกขับเคลื่อนการเติบโตหลักของ Cohu โดยรายการโอกาสทางธุรกิจ HPC ต่อปีแตะระดับประมาณ 1 พันล้านดอลลาร์ ซึ่งรวมถึงโปรแกรมลูกค้าที่ได้รับการจัดอันดับ 13 โปรแกรม และข้อตกลงในระยะเริ่มต้นเพิ่มเติมครอบคลุม CPU, GPU, อุปกรณ์เครือข่าย, ตัวเร่งความเร็วที่กำหนดเอง (custom accelerator), ซิลิคอนโฟโตนิกส์ (silicon photonics) และบริการที่เกี่ยวข้อง

โปรแกรมที่คิดเป็นรายได้ต่อปีที่อาจเกิดขึ้นได้ 230 ล้านดอลลาร์ได้รับการรับรองคุณสมบัติ (qualified) แล้ว ขณะที่อีก 440 ล้านดอลลาร์กำลังอยู่ในขั้นตอนการรับรองคุณสมบัติอย่างแข็งขัน ทั้งนี้ โปรแกรม 7 รายการที่คาดว่าจะบรรลุหรือเริ่มก้าวสู่หลักหมุดการรับรองคุณสมบัติสำคัญระหว่างปลายเดือนพฤศจิกายน 2026 ถึงไตรมาส 1/2027 คิดเป็นรายได้จากระบบที่อาจเกิดขึ้นได้ต่อปีประมาณ 340 ล้านดอลลาร์ตามแผนงานปัจจุบันของลูกค้า

ผู้บริหารระบุว่า ภาพความชัดเจนที่ปรับตัวดีขึ้นเกี่ยวกับความต้องการกำลังการผลิตของลูกค้าในปี 2027 ไม่ใช่การเพิ่มขึ้นของจำนวนลูกค้าอย่างมีนัยสำคัญ คือสาเหตุหลักที่ทำให้รายการโอกาสทางธุรกิจขยายตัว โดย Cohu กำลังขยายการผลิตและประสานงานกับผู้ซัพพลายเออร์เพื่อรองรับปริมาณการจัดส่ง Eclipse ในปี 2027 ที่เพิ่มขึ้นประมาณสามเท่าของปี 2026

การตรวจสอบ HBM และบรรจุภัณฑ์ขั้นสูง

Cohu ยังคงจัดส่งระบบตรวจสอบขั้นสุดท้ายสำหรับ HBM3 และ HBM4 ให้แก่ลูกค้ารายหนึ่งในสหรัฐฯ ซึ่งได้แต่งตั้งให้บริษัทเป็นซัพพลายเออร์เชิงยุทธศาสตร์ ทั้งนี้ ยอดคำสั่งซื้อค้างส่ง (backlog) ของระบบตรวจสอบ Neon ได้ขยายตัวยาวไปจนถึงไตรมาส 1/2027

บริษัทยังมีความคืบหน้าในการรับรองคุณสมบัติการตรวจสอบบรรจุภัณฑ์ขั้นสูงโดยใช้ Neon เพื่อตรวจจับรอยแตกร้าวระดับไมโครที่มองเห็นได้และใต้พื้นผิว โดยผู้บริหารคาดว่าจะส่งมอบระบบสาธิตให้แก่ผู้ผลิตอุปกรณ์แบบรวมศูนย์ (IDM) รายใหญ่ในสหรัฐฯ ในไตรมาส 1/2027 ทั้งนี้ Cohu ประเมินตลาดรวมที่สามารถเข้าถึงได้ (TAM) สำหรับแอปพลิเคชันการตรวจสอบบรรจุภัณฑ์ขั้นสูงและ HBM ไว้ที่ 330 ล้านดอลลาร์

โซลูชันอินเทอร์เฟซและการทดสอบพลังงาน AI

Cohu ได้รับคำสั่งซื้ออินเทอร์เฟซขั้นต้นจากลูกค้ารายใหญ่ในกลุ่มผู้ให้บริการศูนย์ข้อมูลขนาดใหญ่ (hyperscale) สองรายสำหรับการทดสอบโพรเซสเซอร์ และได้ส่งมอบโซลูชันการผลิตอินเทอร์เฟซตัวแรกให้แก่ลูกค้ารายใหญ่ในกลุ่ม HPC แล้ว โดยผู้บริหารประเมินมูลค่าตลาดต่อปีสำหรับโซลูชันอินเทอร์เฟซการทดสอบขั้นสุดท้ายและการทดสอบระดับระบบของการประมวลผลไว้ที่ 200 ล้านดอลลาร์ แม้จะกล่าวในช่วงตอบคำถาม (Q&A) ว่ายังเร็วเกินไปที่จะหารือถึงปริมาณการทดสอบระดับระบบที่มีนัยสำคัญ

Diamondx ยังคงก้าวหน้าอย่างต่อเนื่องในแอปพลิเคชันการจัดการพลังงาน AI โดยลูกค้ารายใหญ่กำลังเร่งขยายการใช้แพลตฟอร์มนี้สำหรับ IC จัดการพลังงานที่ใช้ในเมนบอร์ดของเซิร์ฟเวอร์ เพื่อทดแทนซัพพลายเออร์รายเดิม ทั้งนี้ Cohu มีแผนจะเปิดตัวขีดความสามารถกระแสไฟสูงและแรงดันไฟสูงใหม่ในเดือนธันวาคม 2026 ก่อนที่จะมีการเร่งกำลังการผลิตในครึ่งแรกของปี 2027 โดยผู้บริหารประเมินตลาดเป้าหมายที่สามารถให้บริการได้ (SAM) สำหรับแอปพลิเคชันเหล่านี้ไว้ที่ 290 ล้านดอลลาร์

ซอฟต์แวร์และรายได้ประจำ

รายได้ประจำต่อปีจากซอฟต์แวร์วิเคราะห์ข้อมูลเพิ่มขึ้น 85% เมื่อเทียบรายไตรมาส ส่งผลให้ยอดจองซื้อและรายได้ของธุรกิจทำสถิติสูงสุดใหม่ การเติบโตนี้รวมถึงคำสั่งซื้อซ้ำจากลูกค้ารายใหญ่ในกลุ่ม HPC การขยายการใช้งานระบบตรวจสอบ HBM บนพื้นฐาน AI และไลเซนส์เชิงปริมาณรายแรกของ Cohu สำหรับซอฟต์แวร์การบำรุงรักษาเชิงคาดการณ์บนชุดจัดการระบบรุ่นก่อนหน้า

ผู้บริหารคาดว่าโอกาสรายได้ประจำจะขยายตัวขึ้นเนื่องจากระบบ Eclipse ที่ติดตั้งแล้วจำเป็นต้องใช้ชุดอุปกรณ์ใหม่ หัวความร้อน และโมดูลความร้อนสำหรับอุปกรณ์รุ่นต่อๆ ไป อย่างไรก็ตาม ผู้บริหารข้อสังเกตว่าหลักฐานยังคงมีจำกัดเนื่องจาก Cohu เพิ่งเข้าสู่ตลาดนี้เมื่อประมาณหนึ่งปีที่แล้ว

ตลาดหลัก

ยอดจองซื้อระบบภาคอุตสาหกรรมเพิ่มขึ้น 52% เมื่อเทียบรายไตรมาส ขณะที่ยอดจองซื้อระบบยานยนต์เพิ่มขึ้น 23% โดยผู้บริหารกล่าวว่ายอดจองซื้อภาคอุตสาหกรรมกำลังเข้าใกล้ระดับเท่าเทียมกับยานยนต์และอาจแซงหน้าได้ในไตรมาสต่อๆ ไป ซึ่งส่วนหนึ่งสะท้อนถึงการนำแอปพลิเคชันพลังงานสำหรับศูนย์ข้อมูลจัดเข้าอยู่ในประเภทอุตสาหกรรม

อัตราการใช้งานเซลล์ทดสอบแตะระดับ 84% ในกลุ่มการประมวลผล 83% ในกลุ่มอุตสาหกรรม และประมาณ 80% ในกลุ่มยานยนต์

แนวโน้มจากผู้บริหาร

ผู้บริหารกล่าวว่าระดับสูงสุดของช่วงการเติบโตของรายได้ปี 2027 ขึ้นอยู่กับการรับรองคุณสมบัติของลูกค้าและกรอบเวลาการเร่งกำลังการผลิต โดยสมมติว่าสภาวะตลาดยังคงแข็งแกร่ง ทั้งนี้ Cohu คาดว่าประมาณ 50% ของรายได้ส่วนเพิ่มจะสะท้อนลงสู่กำไรจากการดำเนินงานในระดับการเติบโตที่คาดการณ์ไว้

ความเสี่ยงและประเด็นที่ต้องจับตา

• กรอบเวลาการรับรองคุณสมบัติของลูกค้าและการเร่งกำลังการผลิตยังคงเป็นตัวแปรสำคัญสำหรับระดับสูงสุดของแนวโน้มการเติบโตปี 2027 ของ Cohu
• อุปสงค์ HPC ที่เติบโตอย่างรวดเร็วส่งผลให้ระยะเวลาส่งมอบ (lead time) ยาวนานขึ้นและต้นทุนสูงขึ้นสำหรับเซมิคอนดักเตอร์และชิ้นส่วนเฉพาะทางบางประเภท โดย Cohu กำลังจัดหาชิ้นส่วนสำคัญและประเมินซัพพลายเออร์และโซลูชันทางเลือก
• ซัพพลายเออร์จะต้องขยายขีดความสามารถควบคู่ไปกับกำลังการผลิตของ Cohu เพื่อรองรับการเพิ่มผลผลิต Eclipse เป็นสามเท่าตามแผน
• ผู้บริหารมีความชัดเจนจำกัดเกี่ยวกับการจัดส่ง Eclipse ในไตรมาส 4/2027 แม้ว่าจะคาดว่าปริมาณจะเพิ่มขึ้นตั้งแต่ไตรมาส 1 ไปจนถึงช่วงกลางปีก็ตาม
• ลูกค้ารายเดียวในกลุ่มการประมวลผลคิดเป็นสัดส่วนมากกว่า 10% ของรายได้ในไตรมาส 3

ประเด็นสำคัญจากการตอบคำถามของนักวิเคราะห์ (Q&A)

ผู้บริหารกล่าวว่าการจัดส่ง Eclipse น่าจะเพิ่มขึ้นตลอดปี 2027 โดยคาดว่าจะมีปริมาณที่มีนัยสำคัญในไตรมาส 2 และไตรมาส 3 ทั้งนี้ Cohu มีแผนที่จะปิดปี 2026 ด้วยอัตราการผลิตต่อสัปดาห์เป็นสองเท่าของระดับที่บันทึกได้ ณ สิ้นเดือนมิถุนายน โดยกำลังการผลิตจะยังคงขยายตัวอย่างต่อเนื่องจนถึงเดือนมิถุนายน 2027

กระบวนการรับรองคุณสมบัติทั่วไปนับตั้งแต่การเจรจาขั้นต้นจนถึงการผลิตจริง มักใช้เวลาประมาณ 9 ถึง 12 เดือน และหลังจากติดตั้งหัวความร้อนรุ่นใหม่ การรับรองคุณสมบัติต้องใช้เวลาอีกประมาณ 6 ถึง 8 สัปดาห์ ทั้งนี้ คาดว่าโปรแกรมหลายรายการจะเปลี่ยนผ่านเข้าสู่การผลิตระหว่างปลายเดือนพฤศจิกายน 2026 ถึงสิ้นไตรมาส 1/2027

เมื่อถูกถามถึงสัดส่วนของ HPC ต่อการเติบโตในปี 2027 ผู้บริหารระบุว่ารายได้อาจแตะระดับอย่างน้อยที่ช่วงสูงของกรอบ 200 ล้านถึง 250 ล้านดอลลาร์ที่เคยอ้างถึงไว้ก่อนหน้านี้ โดยได้รับการสนับสนุนจากการเพิ่มกำลังการผลิตและผลผลิตของ Eclipse ตามแผน

Cohu ยังกล่าวอีกว่าประสบการณ์จนถึงปัจจุบันชี้ให้เห็นว่า ลูกค้ามักมอบหมายตระกูลอุปกรณ์ทั้งหมดให้กับแพลตฟอร์มเดียว แทนที่จะแบ่งตระกูลนั้นให้กับซัพพลายเออร์หลายราย อย่างไรก็ตาม ผู้บริหารเตือนว่าประวัติการดำเนินงานของบริษัทในตลาด HPC ยังคงค่อนข้างสั้น

ถอดความการแถลงผลประกอบการฉบับเต็ม

บทถอดเสียงฉบับเต็มของการประชุมทางโทรศัพท์ผลประกอบการ

คำชี้แจงจากฝ่ายบริหาร

Operator

Good day, and thank you for standing by. Welcome to Cohu's Third Quarter 2026 Financial Results Conference Call. [Operator Instructions]. Please be advised that today's conference is being recorded.

I'd now like to hand the conference over to your speaker today, Matt Hutton, Vice President, Strategy and Investor Relations. Please go ahead.

Matthew Hutton

Good afternoon, and welcome to our conference call discussing Cohu's third quarter 2026 financial results and our outlook for the fourth quarter of 2026. I'm joined today by Luis Muller, Cohu's President and CEO; and Jeff Jones, Cohu's Senior Vice President of Finance and CFO. If you need a copy of our earnings release, it can be found on our website at cohu.com or by contacting Cohu Investor Relations. A slide presentation accompanying today's call is also available in the Investor Relations section of the website. Replays of this call will be accessible via the same page after the conclusion of the call.

During this call, we will be making forward-looking statements that reflect management's current expectations concerning Cohu's future business. These statements are based on the information available to us at this time, but they are subject to rapid and sometimes abrupt changes. We encourage everyone to review the forward-looking statements section of our slide presentation and the earnings release as well as Cohu's filings with the SEC, including the most recently filed Form 10-K and Form 10-Q. Our comments are current as of today, October 28, 2026, and Cohu does not assume any obligations to update these statements for events occurring after this call. Additionally, we will discuss certain non-GAAP measures during this call. Please refer to our earnings release and slide presentation for reconciliation to the most comparable GAAP measures.

Now I'd like to turn over the call to Luis Muller, Cohu's President and CEO. Luis?

Luis Müller

Good afternoon, and thank you for joining Cohu's Third Quarter 2026 Earnings Call. We delivered solid third quarter progress as high-performance computing adoption accelerated and recurring revenue growth strengthened. Revenue increased 37% year-over-year to $173 million. Book-to-bill was solidly above 1, and total bookings grew 126% year-over-year, driven by a 318% increase in systems bookings and 36% growth in recurring bookings.

Systems represented 59% of third quarter bookings and recurring business represented 41%. These results reflect broad-based portfolio improvement, growing adoption of our Eclipse thermal solutions for high-performance computing and accelerating demand for services, consumables and software that provide a stable foundation for long-term value creation. Our annual high-performance computing opportunity pipeline expanded to approximately $1 billion and continues to mature as more customer programs advance through qualification and into production.

We are raising our fiscal year 2026 high-performance computing revenue outlook to approximately $120 million. To support growing customer interest and an expanding opportunity funnel, we're scaling manufacturing capacity and working closely with supply partners with the goal of delivering approximately 3x as many Eclipse handlers in 2027 as in 2026. Estimated semiconductor test stabilization improved to 82% at the end of the third quarter, up 2 percentage points sequentially and 7.5 points year-over-year.

Computing customers drove system demand with bookings increasing 81% sequentially, driven by rapid adoption of our Eclipse thermal handler for testing processors used in data centers. We also delivered continued improvement in interface solutions, strong momentum in inspection and a record quarter for software analytics bookings and revenue.

Let me now review the quarter by product line. Starting with high-performance computing. This remains one of our most important growth areas for the company. During the quarter, the annual opportunity associated with our Eclipse platform expanded by about 18% to approximately $1 billion. The funnel now includes 13 ranked customer programs, plus additional early-stage engagements spanning server and data center processors, CPUs, GPUs, networking devices, custom accelerators, silicon photonics, HBM inspection and related services. This breadth diversifies the business opportunity across device architectures, customers and revenue streams. The traction is visible in both the scale and the maturity of the funnel.

We now have programs representing $230 million in qualified annual revenue potential and another $440 million in active qualification. Additional systems and application hardware are being shipped for engineering evaluations, production device qualification and capacity planning, while the funnel continues to expand across more programs and strategic accounts. At a major networking customer, the initial device has qualified on Eclipse in the third quarter, and we are preparing to ship a production configuration thermal head for a higher power device validation during the fourth quarter.

Looking at the next wave, 5 additional customer programs are scheduled to reach major qualification milestones between late November and first quarter of next year, with 2 more targeted to start in the same period. Together, these 7 programs represent approximately $340 million of annual systems revenue based on current customer planning. We remain on track with our internal production expansion, and we're working closely with supply partners to ensure they ramp in tandem with our manufacturing capacity. This coordinated execution is important to support customer schedules and expand Cohu's business in 2027 and beyond.

Customers are evaluating Eclipse because it combines precise thermal control of high-power and power density devices during test, scalable automation and an architecture that can support multiple device generations, including future rollout of optical interfaces. These capabilities are increasingly important as AI processors become larger, hotter and more complex and as customers seek standardized platforms that can improve utilization and reduce time to yield.

Now turning to high-bandwidth memory and inspection. We continue to ship final inspection systems for HBM3 and HBM4 to a U.S.-based customer during the third quarter. The customer has designated Cohu as a strategic supplier and is working with us to plan for potentially steeper ramp scenarios. Neon inspection systems also shipped in volume to integrated device manufacturers in the third quarter. Backlog is growing and were booked deep into the first quarter of 2027. We also progressed towards an advanced package inspection qualification with the Neon platform to prove our vision technology's ability to detect both visible and subsurface microcracks.

We expect to soon complete an agreement with a large U.S.-based integrated device manufacturer, supporting shipment of demonstration systems in the first quarter of 2027. We're poised to grow our advanced packaging and HBM inspection business meaningfully in 2027 by pursuing large customers that can benefit from the Neon platform's differentiated vision technology. These applications support expanding AI and high-performance computing ecosystem and represent a $330 million addressable market opportunity for Cohu.

Moving to interface solutions. We delivered another quarter of sequential growth with increasing momentum in computing test interface applications. During the quarter, we secured first interface orders with 2 leading hyperscale customers for processor test. These wins establish new positions in final test and system-level test applications and create pathways to broader production adoption. In co-packaged optics, we won additional interface applications that combine high-speed electrical contact with precise optical alignment for simultaneous electrical and optical testing.

We also delivered our first interface production solution for a major HPC customer and are progressing to follow-on opportunities for devices with substantially higher pin counts and data rates. Together, these design wins position us to participate in an estimated $200 million annual market for computing final test and system-level test interface solutions. Our focus is converting current hyperscaler processor, networking and co-package optics engagements into production revenue.

Turning next to AI power and semiconductor test. AI infrastructure is driving a major transition in power architecture. As rack power moves from traditional levels of 10 to 30 kilowatts to well above 100 kilowatts, data centers are adopting higher voltage distribution and more efficient power conversion to address thermal reliability and operating cost challenges. These technologies are expected to extend beyond hyperscale data centers into industrial automation, automotive electrification, edge AI and physical AI applications. This broadened demand for power devices that operate at higher voltages and currents, increasing the need for scalable and accurate semiconductor test solutions.

In semiconductor test, we continue to advance Diamondx in AI power management applications. A leading customer is ramping production with our platform for testing power management ICs used in server boards. We also remain on track to release new high current and high-voltage capabilities in December ahead of anticipated production ramps in the first half of 2027. These power management applications represent an approximately $290 million serviceable addressable market opportunity for Cohu.

Switching to software analytics. Annual recurring revenue in software analytics increased 85% sequentially, contributing to record bookings and revenue for the business. Growth was led by expanding production deployments, including a large repeat order for a leading high-performance computing customer and an expanded AI-based HBM inspection software deployment at a leading memory manufacturer. We also booked our first volume license for prescriptive maintenance software on earlier generation Cohu handlers, creating a meaningful attach opportunity across our installed base.

Our next release for high-power thermal handlers is designed to reduce troubleshooting from upwards of 5 days to less than 1 hour while keeping customers' data inside the factory through our on-site AI appliance. Beyond Cohu equipment, paid proofs of concept and evaluations are expanding our platform into specialty gas and chemical delivery systems, front-end wafer fabrication equipment, test operations and wire bonding.

Let me conclude with our priorities and outlook. Looking ahead, our priorities are clear: convert the high-performance computing funnel into production revenue and accelerate recurring growth through services, spares and software. We're encouraged not only by the increase in the total value of the HPC pipeline, but also by its improving quality. More programs are advancing through qualification, production monitoring and approval and the opportunity is increasingly diversified across customers and device types.

For 2027, we expect revenue growth of approximately 30% to 40%, driven primarily by conversion of our high-performance computing opportunity funnel. The high end of the range will depend on the timing of customer qualifications and production ramps provided market conditions remain strong. This outlook is supported more by rising power and test intensity than by processor unit volume growth as customers increasingly require the higher-performance thermal solutions Cohu is well positioned to supply. Thank you to our employees around the world for their continued dedication and execution, and thank you to our customers and shareholders for their continued support.

With that, I'll turn the call over to Jeff to review our financial results and outlook in more detail. Jeff?

Jeffrey Jones

Thank you, Luis. Before reviewing the third quarter results and providing fourth quarter guidance, please note that my comments refer to non-GAAP figures. Details about non-GAAP financial measures, including GAAP to non-GAAP reconciliations and other disclosures are included in the earnings release and investor presentation on our website.

Q3 2026 revenue was $173 million, above the midpoint of guidance. Recurring revenue driven primarily by consumables represented 48% of total revenue, providing a meaningful contribution to the quarter. One compute customer accounted for more than 10% of total sales during the quarter. Gross margin was 45.7%, above guidance, reflecting improved cost leverage in the manufacturing of our interface products. Operating expenses were in line with guidance at $53.5 million. Net interest income was approximately $1.5 million after interest expense and a $500,000 foreign currency loss. The Q3 tax provision was $3.7 million, lower than guidance due to improved profitability in the U.S. Non-GAAP EPS for the third quarter was $0.43 and adjusted EBITDA margin was 16.6%.

Moving to the balance sheet. We ended the quarter with increased liquidity and positive cash generation. Cash and investments increased by approximately $10 million during Q3 to $508 million, and cash from operations was approximately $6 million. Total debt is $310 million and includes $288 million from the Q4 2025 convertible debt offering. Capital expenditures were approximately $3 million, mainly for manufacturing machinery and equipment, facility improvements and IT equipment. We continue to target 2026 total capital expenditures of approximately 2% of revenue. This includes capacity expansion of our Malaysia and Philippines manufacturing facilities to support growing customer demand for our Eclipse thermal handler in the high-performance computing market.

Looking ahead, we're expecting Q4 revenue of approximately $178 million, plus or minus $7 million, representing year-over-year growth of 46%. This outlook reflects the ramp of high-performance computing opportunities and continued recovery in our core business segments. We're raising our full year 2026 revenue growth outlook to approximately 38% over 2025. Q4 gross margin is projected to be approximately 44%, and we expect full year 2026 gross margin of approximately 45%.

In the near term, the rapid increase in high-performance computing demand is creating longer lead times and higher input costs for certain semiconductors and specialty components. We're taking proactive steps to secure critical components and minimize the impact on lead times, profitability and customer pricing. We're also working closely with our supply chain partners to expand manufacturing capacity, improve availability and reduce these incremental costs over time.

Operating expenses are expected to be approximately $57 million, reflecting continued investment in resources and product development materials to capitalize on the growing pipeline of high-performance computing opportunities. We believe this is the time to invest behind our differentiated thermal technology, secure customer wins and scale Cohu's business. These targeted investments are expected to support near-term Eclipse system sales and build a growing base of recurring revenue from the consumables and services associated with these highly complex thermal solutions.

Net interest income in Q4 after interest expense and foreign currency impacts is projected to be approximately $1.4 million at current interest rates. The Q4 tax provision is expected to be about $3.3 million and diluted shares are projected to be approximately 55 million, including 5.8 million shares attributable to the convertible debt. And of that amount, 2.4 million shares will be fully offset by the capped call but are required for U.S. GAAP diluted EPS calculations.

In summary, our 2026 priorities remain focused on supporting the R&D investments and production ramp required to secure multiple design wins in the compute market, including AI data center infrastructure, HBM memory and physical AI applications. As we execute on these opportunities, we remain focused on progressively increasing EBITDA margin and free cash flow.

Turning to 2027. Our preliminary view based on our high-performance computing opportunity funnel and the continued recovery of our core business is for revenue growth of 30% to 40% over 2026. At that revenue range, we expect gross margin of 46% to 48%. We expect quarterly operating expenses to remain in the $54 million to $56 million range, similar to 2026 as we realign costs from our core business toward high-performance computing products and customers. This disciplined cost approach, combined with revenue growth is expected to produce significant operating leverage with approximately 50% of incremental revenue falling to operating income.

That concludes our prepared remarks, and now we'll open the call to questions.

Operator

[Operator Instructions] Our first question comes from David Duley with Steelhead Securities.

ช่วงถาม-ตอบ

David Duley

Luis, you mentioned something about a couple of interface wins with hyperscale customers that I guess could lead to a final test and SLT system sales. Could you just elaborate on that particular topic a little bit more?

Luis Müller

Dave, well, I think it's a little too soon to be talking about SLT, to be honest with you. We talk about having hyperscale customers in the funnel, and this is specific to our final test business. We have them qualifying the Eclipse handler, a couple of them actually in late-stage qualification. So we'll describe more of that business when it starts translating into sort of volume revenue here in the coming quarters.

David Duley

Okay. And as far as the Eclipse revenue next year, you talked about shipping several hundred Eclipse systems. And I'm just wondering, when you think about the output for the calendar year, is it going to be second half weighted? Or what does the shape of Eclipse system shipments look like?

Luis Müller

Yes. Good question. Well, I think it's actually going to be a ramping through the year, Dave. We should see a pretty decent volume in the second and the third quarter. There will be volume in Q1 as well. I don't have -- we don't have a good view on the fourth quarter at this point. But I would say sort of increasing from Q1 into the summer months, potentially at peak through Q2, Q3 or possibly even ramping further Q3 to Q4, but we don't know much about Q4. The bottom line is we're looking at putting enough capacity in place to ship 3x more Eclipse systems next year than we're shipping this year in 2026.

David Duley

And so the overall capacity is -- you mentioned, I guess, shipping 3x more. What is the capacity plan for the balance of this year and into next year as far as -- I mean, I imagine you're tripling or quadrupling or something like that. I think you talked about 2xing it a couple of different times. So maybe just elaborate a little bit more on the capacity plan.

Luis Müller

Yes. We're increasing the capacity this year, exiting the year at twice the output per week rate than we exited the end of the first half of the year than we exited June, and that continues to increase through June of next year. So obviously, it's a ramp in totality, if you kind of look at it in totality is that it is to be able to output 3x more test systems next year, but still ramping between January and June next year.

Operator

The next question comes from Quinn Fredrickson with Baird.

Quinn Fredrickson

Can you discuss in that 30% to 40% growth for '27, how much is HPC versus just continued market recovery in the rest of the business? I think you had sized HPC as maybe $200 million to $250 million previously. I'm just wondering if that's changed at all with the incremental qualified revenue and just the overall expanded pipeline.

Jeffrey Jones

No, Quinn, I would say that, that's probably at the higher end of that range would be our estimate at the moment. As Luis said, we're looking to triple our capacity and output on Eclipse systems. And so that would put us at least at the high end of that range.

Quinn Fredrickson

Got it. Okay. And then could you discuss test cell utilization amongst each of your key markets? And just wondering how we should be thinking about that progression over the next several quarters, whether this couple of hundred bps of improvement you saw this quarter is sustainable or whether any markets get a little choppier outside of compute?

Jeffrey Jones

Well, like we said here, we got about 82% test cell utilization at the end of third quarter. Computing -- the compute business is actually leading the pack. We have -- I think it's 84% utilization in the compute sector. We got 83% utilization in industrial, about 80% utilization in automotive. We don't really forecast utilization for the coming quarters, Quinn. Never done that before. We do know that the -- it's more of the core business tends to have a seasonality. I don't think compute necessarily has the seasonality yet for us, but the core business, auto industrial tends to have a bit of a seasonality profile during the year. But that's sort of the number across market segments.

Operator

Our next question comes from Krish Sankar with TD Cowen.

Robert Mertens

This is Robert Mertens on the line for Krish. I guess the first one, just around the new high-performance compute pipeline estimate. It looks like for the year, you raised revenues, maybe like $120 million from $105 million and brought up the full pipeline. Was that more of just giving more clarity from your customers and their ramp plans? Or was it an additional customer qualification or additional engagement that you didn't have last quarter?

Luis Müller

Largely, we're gaining more clarity on the customer forecast for 2027, Robert. So a couple of our customers have come in with specific need dates for systems throughout next year. And generally speaking, they are growing. Their business is growing. So the funnel -- the total opportunity funnel size is getting larger. The customers that we're qualifying or have completed a qualification and now moving to a production device are also giving us a little bit more clarity now on what they need for 2027.

So in aggregate, these numbers are getting a little bigger than what we had before. So the funnel is getting bigger. It's not like we're necessarily adding a lot more customers in the funnel. It's really getting more clarity from those customers on what they need in '27, either from the ones -- both from the ones that we have qualified and the ones that we are in late-stage qualification at the time -- at this time.

Robert Mertens

Great. That's very helpful. And then just, I guess, a broader one in terms of as you're adding capacity and ramping up with customers, we definitely heard a lot of pain points throughout the supply chain in terms of component tightness. Have you seen that crop up anywhere for your own business or for your customers? Or is it more really just around higher component costs more on the memory side?

Luis Müller

No. We see some tightness on specialty components that we use in our systems. We have been working with the suppliers. We have been looking at alternative suppliers and alternative solutions as well to augment what we currently use. I think at this point, we're doing okay. Don't see it as a -- necessarily as a constraint to business today. But obviously, with the intent of tripling the output, the system output next year, we need to be cognizant that so do our suppliers. They need to ramp up their capacity or alternative suppliers have to come in and augment that.

Operator

Our next question comes from Craig Ellis with B. Riley.

Craig Ellis

I apologize. I'm getting into the queue late. So you may have touched on this already. It looks like from the investor presentation, we're doing a nice job of raising our computing pipeline by around 13% to $1 billion. Can you just talk a little bit about what's driving that strength and how you're feeling about the way you're converting from engagement through qualification and then down to what gets qualified?

Luis Müller

Craig, yes, we're just talking a minute ago about this. We have better visibility now from both customers that are in the qualified bucket and the customers that are in late-stage qualification now that are opening up and describing more of their capacity needs in 2027 and the time line of when they need that capacity. So it's just -- the numbers are growing basically. There are elements in the funnel that we don't have any better visibility than we did a couple of months ago. But several of these customers, we have better forecast now into 2027. And the demand is larger than we had originally anticipated. So in aggregate, the funnel is $1 billion, potentially $1 billion plus as we see today for 2027.

Qualification is going well. We have made a comment here during the prepared remarks that we had a network processing customer that qualified the initial device in Q3. We're moving now to ship a new thermal head to those -- to that customer pretty much now, so we can do a final qualification on the actual production device that we should be ramping in 2027. We got a few other customers that have moved down the funnel, a couple of hyperscaler customers that are now providing some visibility and engaging in discussions on needs and volume needs for '27.

Craig Ellis

Got it. And Luis, as you think about the breadth of customers that are in qualified and I take it the ones you mentioned are contributing to revenue this year. Can you just talk a little bit further about how broad our reach is today in terms of the customers that we're engaged with versus those that are revenue generative? And what are you learning about the cycle time from engagement to revenue generation?

Luis Müller

The breadth is pretty much how we described it earlier, Craig. We have customers here that are CPU. There's been a lot of CPU business in 2026. We're seeing quite a bit of custom ASIC business opportunities to augment into 2027, network processors as well into 2027, some GPU business into 2027. So it cuts across a variety of products and a multitude of customers actually pretty broad from hyperscalers to fabless to, I guess, one IDM in the mix. So feeling very happy, very good about the opportunity. This is tremendous actually. So I don't know, what was the second part of your question again?

Craig Ellis

I think it was just about the cycle time that you're...

Luis Müller

Qualification cycle time.

Craig Ellis

Learning from initial engagement to revenue generation.

Luis Müller

Yes. It's about -- as we described before, we have usually a first device to qualify on that is what is currently in production, generally a lower power device. Once that is used to qualify the system, then it comes a second pass for typically a next-generation device thermal head that needs to be delivered on an existing system.

Once that gets to be installed, we're looking at about 6 to 8 weeks to qualify that thermal head and then production follows that. If you put it all together in a time line, approximately 9 to 12 months has been the cycle that we observed for some of these qualifications all the way to being actively in production. So a lot of these things are going to be converting here between kind of late November into the end of Q1, we should be seeing quite a few of these customers converting into business.

Craig Ellis

Good for you. And then if I could just flip to an area that we haven't been talking about as much lately. The mobile business, I think it's showing at about 10% of system revenue in the third quarter. A large entity has changed the way they're releasing products and it's going to a much more balanced 4Q, late 3Q, 4Q, 1Q release. What implication is there for the seasonality of your business based on what's happening with that large smartphone OEM?

Luis Müller

So the mobile business is, like you can see here in the graph, it's the smallest of our business segments at the moment. Orders have increased quarter-on-quarter in the third quarter. But still, if you look forward here, likely to be the smallest or second smallest of our business in terms of revenue over the next 1 to 2 quarters. I don't expect a lot of change necessarily in the mobile business in the near term.

Seasonality-wise, it depends. As usual, we have one major customer in the space here that tends to place orders late in the year, beginning of the year. It's a product line, thermal subsystems that we sell that tend to have a little bit longer lead times. We have another customer that typically comes in second into early part of third quarter. So that is sort of entering a seasonally slow part of the year right now and then picks up again in the second quarter coming up 2027. I don't expect a dramatic change given the size of the business today, a dramatic change to the total portfolio at Cohu.

Operator

Our next question comes from Kevin Garrigan with Jefferies.

Kevin Garrigan

Congrats on the great results. Once Malaysia is operational, how much additional capacity will you have above the 3x Eclipse shipments if customer forecasts were to increase again? And what would be the bottleneck there if you wanted to increase capacity again?

Luis Müller

So just remember, when we talk about the increase in 3x, that's specific to Eclipse, right? So we still have the capacity in Malaysia to do the core business. Same capacity we had back in 2021, 2022. We obviously have flexed direct labor down during the down cycle of the market. We can flex it back up. Supply chain, same story, flex down, flex back up. So -- but that capacity is generally the same subject to flexing the supply chain and the demand power. With that said, if you put it all together, we'll be in a position to I think, ship over 2,000 systems next year if we were to utilize the entire capacity for handlers, right? Eclipse specifically is the one that we've been commenting that is expanding here to 3x of 2026 because that relates to the high-performance computing market.

Kevin Garrigan

Got it. Got it. Okay. That makes sense. And then as a follow-up, on the power management IC program that's ramping for server boards, can you just describe how that opportunity developed? Was Diamondx replacing some of your competition? Or was this a new test requirement?

Luis Müller

No, it's Diamondx is replacing an incumbent supplier. Obviously, there aren't that many options for names here in the market to look into, right? But it is replacing one of those incumbent names that you're familiar with. It's our focus in the analog market, particularly now current high-voltage instrumentation and intercepting the deployment of gallium nitride technology that is being used in power management devices going into server boards.

Operator

Our next question comes from Denis Pyatchanin with Needham & Company.

Denis Pyatchanin

My first question is about your in-qualification system customers. So we saw a $165 million increase in that number versus the last quarter. Can we assume that this is predominantly for Eclipse? And maybe you can perhaps discuss which products or customer groups are driving this growth?

Luis Müller

It is exclusively Eclipse actually, Denis. This is all Eclipse related. I mean, obviously, I'm not going to disclose customer names, but we have -- let me look here specifically, we have a couple of hyperscalers in that bucket. We have a number of fabless companies in that in qualification bucket as well, including fabless that are partnering with and supporting hyperscalers in the process. So there's both the hyperscaler direct business and a hyperscaler through a fabless business in the mix here. So I'm not going to obviously disclose the names, but it's a pretty rich pool of customers on the in-qualification bucket right now.

Denis Pyatchanin

Maybe as a direct follow-up to that. So some of these fabless companies supporting the hyperscalers are -- this is related to networking and data center?

Luis Müller

They are related to data center indeed. They are related to some custom ASIC-type devices as well.

Denis Pyatchanin

Got it. And then for my second question, I think you spoke about some additional business for the Neon system for advanced packaging. Was this for a new customer or a new application at an old customer?

Luis Müller

For Neon, this would be a new customer. Not a new customer for Cohu. It is a computing customer of Cohu today, but we don't sell any inspection systems to this customer. So it will be a new customer using Neon into their product lines, into what they're doing with advanced packaging technology right now. So it would be die level inspection for advanced packaging. And we don't sell any inspection systems today at the moment.

Operator

Our next question comes from Daniela Talio with Stifel.

Daniela Talio

This is Daniela Talio on for Brian Chin at Stifel. My first question is more on the recurring revenue side. As you're having conversations with customers and you're going through the upgrade process, are you seeing a larger opportunity from the actual dollar content being higher per upgrade or more frequent upgrades just due to device complexity?

Luis Müller

So the upgrade cycle, when we talk about computing, in particular, right, the upgrade cycle is about to start to some of the systems we have shipped over a year ago where new devices are getting launched and new thermal heads are being purchased for Eclipse systems that have been installed. We don't know exactly how this is all going to take shape in years to come. What we know today is many of these devices have about 1.5 years life cycle, 2 years life cycle. And at the end of that, there's a new product that comes out.

We typically sell device application kits to those new products. But now on the high-performance computing side, a lot of times, we also have to replace the thermal head. The die size has increased, the power management requirement has increased. So a new thermal head entirely needs to be replaced or a new thermal module has to be replaced going along with device kits. I would expect -- I would hope that we will see an increase in recurring revenue associated with Eclipse because of those thermal heads.

We've seen this in the past when Cohu had a strong business in the computing market, and I expect that to be the case here going forward. But quite honestly, we are new into this market, right? We started getting into this market just about a year ago. So we're now starting to see a little bit of that business coming through as we put more Eclipse out at various customers, I think that's going to pick up some momentum here basically a year, 1.5 years later. So I'm pretty optimistic about the recurring business, but we haven't seen a lot of evidence yet because we're new in the space.

Daniela Talio

Okay. Yes, that makes sense. And then for my follow-up, when you're having conversations with customers, specifically looking into next year, 2027, do you get the sense that they are looking for more than one supplier? Do they prefer to have 2 vendors? What are the conversations that you're having around this type of topic?

Luis Müller

We have seen a mix of both, to be honest with you. We have seen customers putting an entire product line on our system once we break into the device market. We haven't quite seen a lot of split business by device family. So I think the experience here is still early on for us. But as I said, what we see today is, let's say, we get the entirety of a data center device market, while our competitor has the entirety of a different device family, right?

But again, we're coming into the market. So a little early to say when we're still coming up to $120 million revenue this year. I mean it's a good question to reask about a year from now and see how that really has evolved. But up to this point, yes, it seems like when we take a device family, we take the entirety of that device family. That's the limited experience we have, but the experience we have thus far.

Operator

Our next question comes from Christian Schwab with Craig-Hallum.

Benjamin Taxdahl

This is Ben Taxdahl on for Christian here. Great quarter. I just had one question as visibility improved with the customers in the $1 billion pipeline, customers moving from engagement to qualification. Do you guys still think there's a large opportunity to -- for new customers? Or is this -- you're kind of maxed here? I mean I'm just trying to think of as in engagement, is there an opportunity to take that even higher going forward?

Luis Müller

So with final test, Ben, what we're seeing is about $1 billion of opportunity pipeline as we put in our deck here and about $1.5 billion in sort of addressable market. What we're working towards is actually potentially tapping into another vertical in the test space that would create, let's say, another opportunity pipeline on top of final test. We're not there yet, but we're looking more into sort of broadening what we supply, broadening our service and our products so that we can add another SAM or another TAM in total to our business. That will be something we will talk more into -- deeper into 2027. But for now, looking at final test, like I said, opportunity pipeline of $1 billion, SAM of about $1.5 billion.

Benjamin Taxdahl

Great. Really helpful. A lot of my other questions were asked. Great quarter.

Luis Müller

Thank you. Thank you, Ben.

Operator

Our next question comes from the line of Denis Pyatchanin with Needham & Company.

Denis Pyatchanin

Just one question about automotive and industrial since it doesn't seem like anyone is interested in these. Systems are up slightly this quarter. Utilization continues to pick up. What are your key customers in the space telling you about the next several quarters? Are you seeing bookings pick up at all?

Luis Müller

We're seeing bookings -- you asked automotive and industrial. We're seeing bookings pick up in industrial in particular. I mean we've seen a pickup in automotive as well, but predominantly in industrial. I'll give you a sense. Quarter-over-quarter, industrial system orders have increased 52%. Automotive has increased 23%. So sure, a good improvement in auto. But if you look at both side by side, we're now at a point that the industrial market from a booking perspective is almost the same size as automotive.

Historically, automotive being twice the size of industrial. We're approaching parity now. And I would expect in the coming quarters that industrial will surpass automotive. It's just the trend today is much more industrial driven than it is automotive driven. And I think our primary core customers in that segment also have pivoted a lot of their business towards industrial applications, quite a few of those actually towards data center power, which they classify as industrial and so do we at the moment.

Operator

That concludes today's question-and-answer session. I'd like to turn the call back to Matt Hutton for closing remarks.

Matthew Hutton

You will see that we filed a preliminary proxy statement today for a special meeting of stockholders later this year to ratify certain actions previously approved by the Board and our stockholders. The ratification is intended to address a technical issue relating to the record dates for the prior meeting and provide certainty regarding the validity of those actions. Please see the preliminary proxy statement for further information.

Before we sign off, I'd like to note that we will be hosting our Investor Day on November 10 in New York City, where we will provide a deeper look at our strategy, growth opportunities and new midterm financial model. Please visit ir.cohu.com for additional details, including registration information. I'd also like to note, we will be attending the following investor conferences over the next 3 months. Baird's Global Industrial Conference on November 12 in Chicago, the CEO Summit on December 15 in New York City and the Needham Growth Conference on January 14 virtually. Thank you for joining today's call. We look forward to speaking with you soon.

Operator

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