에어 테스트 시스템즈(AEHR) 2027 회계연도 1분기 실적 발표 컨퍼런스 콜: AI 수요가 1억 5,000만 달러 전망 견인

핵심 요약

• 에어 테스트 시스템즈(Aehr Test Systems)는 2027 회계연도 1분기 매출이 전년 동기 대비 200%, 전분기 대비 75% 증가한 3,300만 달러로 사상 최대치를 기록했다고 발표했습니다.
• 수주액은 4,670만 달러를 기록해 출하액 대비 수주 비율(book-to-bill ratio) 1.4를 나타냈습니다. 11월 1일 기준 실질 수주 잔고(Effective backlog)는 분기 말 이후 수주된 840만 달러를 포함해 1억 930만 달러였습니다.
• 비GAAP 순이익은 940만 달러(희석 주당순이익 0.28달러)로 증가했으며, 비GAAP 총이익률은 11%포인트 개선된 48.5%를 기록했습니다.
• 경영진은 2027 회계연도 매출 가이던스를 기존 1억 3,000만~1억 5,000만 달러에서 1억 5,000만 달러로 상향 조정했습니다. 비GAAP 순이익은 매출의 최소 20%에 달할 것으로 예상됩니다.
• 이러한 성장은 FOX 웨이퍼 레벨 번인(burn-in) 수요, 소노마(Sonoma) 패키지 레벨 번인 출하량, 그리고 웨이퍼팩(WaferPak) 컨택터 및 관련 소모품의 지속적인 매출이 견인하고 있습니다.
• 한 주요 AI 반도체 고객사는 2027년 초 대만의 OSAT에 완벽한 구성을 갖춘 FOX-XP 양산 평가 셀을 설치할 계획이며, 검증 후 수개월 이내에 대량 양산 여부를 결정할 것으로 예상됩니다.

핵심 재무 실적

사업 및 운영 성과

AI 웨이퍼 레벨 번인

에어의 FOX 웨이퍼 레벨 번인 플랫폼에 대한 수요가 지속적으로 확대되고 있습니다. 회사는 주력 AI 프로세서 생산 고객사로부터 수주받아 지난 8월 발표한 2,200만 달러 규모의 자동화 FOX-XP 시스템, 웨이퍼팩 자동 정렬기(auto aligner), 풀 웨이퍼 컨택터 후속 주문을 이행하고 있습니다.

AI 가속기, CPU 및 네트워크 프로세서를 공급하는 또 다른 주요 업체가 벤치마크 평가를 완료하고 완벽한 구성을 갖춘 FOX-XP 테스트 셀을 구매하기로 결정했습니다. 해당 시스템은 2027년 초 대만 OSAT에 설치될 예정입니다. 경영진은 해당 고객사가 대량 도입을 결정하기 전에 처리량, 가동률 및 기타 생산 요구사항을 평가할 것으로 예상하고 있습니다.

에어는 웨이퍼 레벨 번인을 통해 고가의 멀티 다이 패키지에 통합되기 전에 개별 다이를 격리하고 전압, 전류 및 온도를 제어할 수 있다고 밝혔습니다. 이번 양산 평가에는 테스트 대상이 되는 각 다이의 온도 조건을 독립적으로 관리하는 에어의 특허받은 다이 온도 제어 기술이 포함됩니다.

소노마 패키지 레벨 번인

에어는 주력 하이퍼스케일 고객사의 첫 번째 AI 프로세서용으로 매달 10여 대 이상의 소노마 시스템을 출하 및 설치하고 있습니다. 동일 고객사의 두 번째 AI 프로세서도 소노마 검증을 마쳤으며, 에어는 시스템, 번인 모듈 및 소프트웨어 기능 개선에 대한 후속 주문을 수주했습니다.

경영진은 고객사가 두 번째 프로세서의 생산량이 첫 번째보다 더 많을 것으로 예상하고 있다고 전했습니다. 또한 단일 패키지 내 독립적으로 제어되는 여러 열 구역을 포함하여 더 큰 전력과 더 복잡한 열 관리가 필요하다고 덧붙였습니다.

소모품 및 지속적인 생산 수요

웨이퍼팩 컨택터, 소노마 번인 모듈 및 관련 제품은 분기 매출의 41%에 해당하는 1,350만 달러의 매출을 올렸습니다. 경영진은 2026 회계연도의 29%와 비교해 2027 회계연도에는 소모품 비중이 전체 매출의 약 35%를 차지할 것으로 예상하고 있습니다.

이들 제품은 설치된 시스템, 생산량 및 신규 디바이스 도입과 연동됩니다. 경영진은 디바이스 변경에 따라 고객사가 1년에서 3년마다 새로운 웨이퍼팩을 구매할 수 있어, 추가적인 시스템 구매 없이도 경상적 매출이 발생할 수 있다고 설명했습니다.

기타 최종 시장

에어는 고속 광 상호연결 및 데이터 센터 애플리케이션 분야에 진출한 다수의 실리콘 포토닉스 고객사로부터 추가 FOX-XP 양산 주문을 수주했습니다.

실리콘 카바이드(SiC) 수요는 추가 웨이퍼팩 주문과 기존 설치된 FOX 시스템의 가동률 상승을 통해 개선 징후를 보였습니다. 경영진은 2027 회계연도에 실리콘 카바이드 매출이 성장할 것으로 예상하고 있습니다. 또한 에어는 지난 6개월 동안 2곳의 질화갈륨(GaN) 양산 고객사를 추가했습니다.

메모리 분야에서는 생산 능력 증설의 한계와 고대역폭 플래시 물량 출하 시점으로 인해 NAND 플래시 사업의 진행 속도가 예상보다 더딘 상태입니다. 아울러 에어는 부상하고 있는 고대역폭 메모리(HBM) 아키텍처에 사용되는 로직 베이스 다이에 대한 웨이퍼 레벨 번인 적용 가능성을 평가하고 있습니다.

생산 능력

에어는 동남아시아의 위탁 생산 업체를 통해 소노마 시스템 출하를 시작했습니다. 이 시설은 월 20대 이상의 소노마 시스템을 추가로 생산할 수 있는 능력을 갖추고 있으며, 에어의 주력 하이퍼스케일 고객사를 지원하는 대만 OSAT로 시스템 출하를 이미 완료했습니다.

회사는 또한 대만 신주 지역의 지원 리소스를 확장하고 있습니다. 경영진은 에어가 현재 2027 회계연도 목표치인 1억 5,000만 달러 이상의 매출을 지원할 수 있는 능력을 보유하고 있으며, 향후 수요에 대비해 생산 능력을 더욱 확장하기 위해 노력하고 있다고 밝혔습니다.

경영진 가이던스

경영진은 2027 회계연도 매출 가이던스를 기존 1억 3,000만~1억 5,000만 달러에서 1억 5,000만 달러로 상향 조정했습니다. 이 목표는 2026 회계연도 대비 200% 성장을 의미합니다.

에어는 연간 비GAAP 순이익이 매출의 최소 20%에 달할 것이라는 기존 전망을 유지했습니다. 경영진은 소노마 시스템 출하 비중이 높아져 2분기 이익률은 다소 낮아질 것으로 예상되지만, 연간 비GAAP 총이익률은 1분기 수준에서 약 0.5%포인트 이내 오차 범위로 유지될 것이라고 덧붙였습니다.

분기별 매출 흐름과 관련해 경영진은 2회계분기 매출이 1분기보다 높고 3분기에는 더욱 증가할 것으로 보았습니다. 4회계분기는 고객사의 일정에 따라 출하 시점이 변동될 수 있으나 연중 가장 높은 매출을 기록하는 분기가 될 수 있다고 설명했습니다.

경영진은 1억 5,000만 달러의 매출 목표가 현재 생산 능력 대비 보수적인 수준이라고 설명했습니다. 다만 이 전망치는 수주되지 않은 기회가 아닌 기존 수주 잔고, 예상 생산 수요 및 현재 인도 일정을 바탕으로 수립된 것입니다.

리스크 및 주요 관전 포인트

• 대규모 AI 웨이퍼 레벨 사업 기회는 대만 OSAT에서의 성공적인 설치, 양산 검증 및 고객사의 최종 승인이 여전히 필요합니다.
• 개별 고객사의 주문 및 양산 증대는 분기별로 차이가 클 수 있어 매출 인식 시점과 제품 구성(믹스)에 영향을 줄 수 있습니다.
• 소노마 시스템 출하 비중이 높아짐에 따라 2분기 총이익률은 전분기 대비 하락할 것으로 예상됩니다.
• NAND 플래시 프로그램은 생산 능력 제약과 제품 출하 시점 등의 이유로 당초 예상보다 더디게 진행되고 있습니다.
• 에어는 예상되는 AI 관련 생산을 지원하기 위해 납기가 긴 부품을 구매하고 운전자본을 늘리면서 310만 달러의 영업 현금을 사용했습니다.
• 경영진에 따르면 회사는 현재 공급 제약을 받고 있지 않으나, 지속적인 성장을 위해서는 공급업체의 추가적인 확약, 생산 확장 및 고객 지원 투자가 필요합니다.

애널리스트 Q&A 주요 내용

경영진은 신규 AI 벤치마크 고객사가 에어의 가장 큰 사업 기회 중 하나가 될 가능성이 있다고 설명했습니다. 생산 능력 추정 모델의 예시로서, 회사는 연간 100만 대의 디바이스를 출하할 때마다 약 5~10대의 FOX 시스템이 필요할 것으로 추산했습니다. 관련 소모품을 포함할 경우 연간 100만 대당 약 3,500만~7,000만 달러 규모의 생산 능력 투자를 의미할 수 있습니다. 이러한 수치는 고객 주문이나 매출 가이던스가 아닌 단순 예시로 제시된 것입니다.

대만에서의 양산 평가 데이터는 2027년 1월에서 3월 사이에 초기 결과가 나올 것으로 예상됩니다. 경영진은 고객사가 이 데이터를 바탕으로 생산 능력 계획을 결정하고 도입할 첫 번째 디바이스를 선정할 수 있을 것이라고 말했습니다.

공급망 준비 상태와 관련하여 에어는 외주 생산을 확대하고 납기가 긴 부품의 구매를 늘렸으며 더 많은 부분 조립(subassembly) 작업을 위탁 생산 업체로 이전했다고 밝혔습니다. 회사는 이러한 조치가 웨이퍼 레벨 시스템의 내부 완공 작업을 줄이고 잠재적 생산 능력을 끌어올릴 것으로 기대하고 있습니다.

경영진은 신뢰성 가속 선별을 위해 더 높은 온도와 전압에 정밀하게 노출되어야 하므로 시스템 레벨 테스트 플랫폼이 전용 번인 테스트를 직접적으로 대체할 수는 없다고 주장했습니다. 에어는 복잡한 멀티 다이 AI 프로세서를 위해 소노마를 패키지 레벨 솔루션으로, FOX를 차별화된 웨이퍼 레벨 플랫폼으로 위치시키고 있습니다.

실적 발표 컨퍼런스 콜 전문

전체 실적 발표 컨퍼런스 콜 녹취록

경영진 발표

Operator

Good day, everyone, and welcome to the Aehr Test Systems Fiscal 2027 First Quarter Financial Results Conference Call. [Operator Instructions] It is now my pleasure to hand the floor over to your host, Todd Kehrli of PondelWilkinson. Sir, the floor is yours.

Todd Kehrli

Thank you, operator. Good afternoon, everyone, and welcome to Aehr Test Systems First Quarter Fiscal 2027 Financial Results Conference Call. Joining me on today's call are Gayn Erickson, Aehr Test's President and Chief Executive Officer; and Chris Siu, Chief Financial Officer. This afternoon, following the market close, Aehr Test issued a press release announcing its first quarter fiscal 2027 results. The release is available on the company's website at aehr.com. Today's call is being webcast live, and a replay will be available on the Investor Relations page of the company's website.

Before we begin, I'd like to remind everyone that management will make forward-looking statements today that are based on current information and estimates. These statements are subject to risks and uncertainties that could cause actual results to differ materially from those expressed or implied in the forward-looking statements. Factors that could cause such differences are discussed in the company's most recent periodic and current reports filed with the SEC. Forward-looking statements, including guidance provided during today's call, are valid only as of today's date, and Aehr undertakes no obligation to update them. Now I'd like to turn the conference call over to Gayn Erickson, President and Chief Executive Officer. Gayn?

Gayn Erickson

Thanks, Todd. Good afternoon, everyone, and thank you for joining us. I'll begin with a brief overview of our first quarter highlights and provide an update on the key customer programs and markets driving our business with a focus on the strong demand we're seeing in the AI and data center infrastructure markets. Chris will then review our financial results and outlook before we open the call for questions.

We delivered an outstanding start to fiscal 2027 with record quarterly revenue, strong bookings and significant progress across several major AI semiconductor customer programs. First quarter revenue was a record $33.0 million, increasing 200% year-over-year and 75% sequentially. Non-GAAP net income was $9.4 million, up 163% sequentially, reflecting the operating leverage in our business as revenue scales. These results demonstrate expanding AI production demand and our ability to meet customer capacity requirements. Bookings totaled $46.7 million, representing a book-to-bill of 1.4. Our effective backlog, including bookings received after quarter end, reached $109.3 million as of November 1.

Based on our strong first quarter performance, substantial backlog and visibility in the customer production requirements, we're raising our fiscal 2027 revenue outlook to $150 million compared with our previous guidance range of $130 million to $150 million. This represents 200% revenue growth over fiscal 2026 and reflects our growing confidence in the production ramps and customer programs supporting our business. We expect non-GAAP net income of at least 20% of revenue for the full fiscal year. Chris will provide additional details on our financial outlook.

Demand continues to accelerate across both wafer level and package level burn-in, driven by increasing semiconductor complexity, power requirements and deployment in mission-critical applications. As devices become more advanced, the need for comprehensive test and burn-in is becoming essential to ensure reliability and performance, which is driving growing adoption of our solutions across multiple markets. For Aehr, this is translating into expanding production demand from existing customers and new opportunities to qualify our solutions on additional devices with both those customers and new customers.

Starting with our wafer level burn-in business, demand for our FOX wafer-level burn-in solutions is accelerating. Our lead AI processor production customer continues to expand capacity. In August, we announced a $22 million follow-on order for multiple systems consisting of fully automated FOX-XP systems with integrated WaferPak auto liners and full sets of proprietary WaferPak full wafer contactors. We are executing against this customer's capacity expansion plans and are encouraged by its forecast for additional capacity in both the short and long term. We've also made significant progress with a major supplier of AI accelerators, CPUs and network processors that successfully completed a benchmark evaluation last quarter.

This customer has decided to purchase a fully configured XP wafer-level burn-in test cell for installation at one of its key outsourced semiconductor assembly and test or OSAT partners in Taiwan, where the system will be evaluated in a high-volume production environment using the same device that was successfully tested in last quarter's benchmark. Installation is planned for early calendar 2027 with a decision regarding high-volume production anticipated within several months following installation and validation. We believe this customer represents one of the largest potential burn-in system opportunities Aehr has ever pursued with significant potential across multiple product lines and across multiple manufacturing locations.

This engagement represents an important step toward broader adoption of wafer-level burn-in for advanced AI processors as well as deployment of our wafer-level burn-in solutions across multiple products and product lines for a single customer. As these advanced processes are integrated into expensive multi-die packages that include other processors, high-bandwidth memory and other components such as electrical and optical I/O chips, identifying early life failures before final assembly can provide significant cost and reliability benefits. There's also a growing trend that some devices that customers are working on may require production burn-in screening at the wafer level as infant mortality failures of certain devices may simply not be able to be screened effectively at the package level.

In silicon photonics, we received additional production orders from multiple customers for fully automated FOX-XP systems used to burn in photonic integrated circuits for high-speed optical interconnects and data center applications. We believe broader adoption of optical interconnects, including emerging co-packaged optics architectures will create additional demand for wafer-level reliability screening. In silicon carbide, additional WaferPak orders and increased utilization of installed FOX systems point to improving demand. We have direct evidence from EV automotive suppliers that companies using our wafer-level burn-in systems are being seen as preferred vendors for their quality and reliability screening. Increasingly, we see that companies performing reliability screening through wafer-level burn-in see it as a real differentiator in their markets.

We also continue to pursue gallium nitride and other power semiconductor applications where reliability screening has proven to be critical to their end markets as well. GaN has much broader application space than silicon carbide does today and has proven to be much harder to screen for early life failures. We feel we are the market leader in this area and have added 2 additional production customers in the last 6 months. In memory, we remain engaged with major manufacturers on next-generation wafer level burn-in. NAND flash opportunities are progressing more slowly than expected because of limited incremental manufacturing capacity and also the timing of new high-bandwidth flash architecture volumes. Separately, emerging high-bandwidth memory architecture for DRAM with more complex logic in the base die may create wafer-level burn-in opportunities before integration into advanced memory packages. These engagements continue and Aer continues to feel these are long-term opportunities for us.

Now let me turn to package level. Our Sonoma high-power package level burn-in business is also expanding. We are shipping and installing more than a dozen systems per month to support our lead hyperscale customers' production ramp for its first AI processor using Sonoma. Our expanded manufacturing capacities are supporting this delivery pace. The same customer's second AI processor has successfully completed qualification on Sonoma. As announced in October, we received follow-on orders for additional systems, burn-in modules and software enhancements to support initial production sampling and an anticipated production ramp in calendar 2027. Importantly, the customer forecasts that this second processor will ultimately reach higher production volumes than its first processor currently in production on Sonoma.

This processor requires greater power and more sophisticated thermal management, including multiple independently controlled thermal zones within a single package. We're also advancing evaluations of our Sonoma high-power systems with additional AI processor suppliers and hyperscale customers for both engineering qualification and potential high-volume production applications. These engagements include bringing up processors here at Aehr as well as at customers' facilities here and in the U.S. and in Asia. These processors include data center-focused processors for training and inference as well as very high-volume edge-based processors for mission-critical real-time inference applications. We continue to see package level applications driving our business in addition to the growth in wafer level.

On our consumable business, complementing our system sales, our proprietary WaferPak contactors and Sonoma burn-in modules support customers' ongoing production requirements and new device introductions. We currently expect these products to represent about 35% of our fiscal '27 revenue. As our installed base expands across AI processors, silicon photonics, silicon carbide, gallium nitride applications, we see further opportunities for these engineering qualification and production-driven purchases. I also want to highlight an important change in the profile of our business that may not be fully apparent from the individual customer programs we've discussed.

Historically, Aehr had only a handful of customers with the potential to generate significant production orders. Today, we see that potential across more than a dozen customers, spanning AI processors, silicon photonics, silicon carbide, gallium nitride and other applications. This represents meaningful diversification, not only across customers and end markets, but also across both our FOX wafer-level burn-in and Sonoma package level burn-in platforms. Individual customer orders and production ramps can still vary significantly from quarter-to-quarter, but we now have a much broader range of customer and production opportunities contributing to our growth outlook. We believe this breadth reduces our dependence on the timing of any single customer program and positions Aehr for more diversified, sustainable long-term growth.

Importantly, this expanding customer base, combined with the increased number of devices being qualified production and growing installed base of our systems is creating multiple opportunities for additional revenue growth beyond our fiscal year. To support expanding demand, we've increased our manufacturing capabilities, including in Southeast Asia. These investments are supporting our current customer production ramps and providing capacity for additional growth. We continue to work closely with suppliers to meet customer delivery requirements. We're also planning further manufacturing expansion and targeted investments in engineering, customer support and sales to support growth beyond fiscal 2027. Our priorities remain delivering on existing production commitments and advancing additional customer programs through qualification and validation.

Before turning the call over to Chris, I'd like to welcome Vijay Rao to our Board of Directors. Vijay brings 30 years of technology industry experience and deep expertise in hyperscale data center infrastructure, including AI accelerators, high-performance memory and optical networking. His understanding of semiconductor reliability requirements will be valuable as we expand our presence in AI and data center applications.

To close, our record revenue, strong quarterly bookings, record effective backlog and expanding customer opportunities have strengthened our confidence in the year ahead. With existing customers increasing production, additional AI processor programs advancing toward commercialization and a significantly broader pipeline of opportunities across both our wafer level and package level burn-in platforms, we believe Aehr is exceptionally well positioned for continued growth in fiscal 2027 and beyond. Our focus remains on executing against existing production commitments, expanding our installed base and converting additional qualification programs into long-term production business. With that, I'll turn it over to Chris.

Chris Siu

Thank you, Gayn, and good afternoon, everyone. Our first quarter results reflect strong execution against expanding customer demand with record first quarter revenue, bookings that exceeded revenue and significant operating leverage as we continue to scale our business. Importantly, we increased quarter end backlog to approximately $101 million despite substantial shipments during the quarter, and our effective backlog reached $109.3 million, providing substantial visibility into our fiscal 2027 outlook.

I'll begin with a brief update on our manufacturing capacity followed by a review of bookings and backlog, first quarter financial results, cash flow and balance sheet and our updated fiscal 2027 outlook. During the first quarter, we continued expanding our manufacturing and customer support capabilities to meet growing demand. In addition to expanding our Fremont operations, we began shipping Sonoma systems from an existing contract manufacturer in Southeast Asia. This operation adds capacity for more than 20 additional Sonoma systems per month, giving us greater flexibility to support growing customer production requirements. During the quarter, we shipped multiple Sonoma systems from this facility to the Taiwan-based OSAT supporting our lead hyperscale customer and the ramp proceeded smoothly. This asset-light manufacturing approach allowed us to increase production capacity and responsiveness without a comparable increase in infrastructure or capital spending.

We are also expanding our local support capabilities in Taiwan. Following the signing of a new lease in Hsinchu, we added customer service resources and planned selective hiring in sales and customer support as wafer level and package level burn-in programs move toward higher production volumes. Turning to bookings and backlog. First quarter bookings totaled $46.7 million, an increase of more than 300% from $11.4 million in the prior year quarter. The increase was driven primarily by orders for FOX wafer-level burn-in systems, WaferPak contactors and auto aligners supporting AI processor and silicon photonics applications. Quarter end backlog was $100.9 million, up from $80.6 million at the end of fiscal 2026.

Importantly, we increased backlog to above $100 million despite recognizing $33 million in revenue during the quarter, reflecting strong new order activity. Including $8.4 million in additional bookings received after quarter end, effective backlog reached $109.3 million as of November 1. This substantial backlog, combined with our expanded manufacturing capacity and visibility into existing production programs, provides a strong foundation for our updated fiscal 2027 revenue outlook.

First quarter revenue was a record $33 million, increasing 200% from $11 million in the prior year quarter and 75% sequentially. The result exceeded consensus expectations and reflected strong execution across both package level and wafer level burning programs. Revenue included Sonoma system shipments to our lead hyperscale customer in Taiwan, multiple FOX-P system shipments, including a FOX-XP system and WaferPak auto aligners to our lead AI processor customer and FOX-NP system shipments to customers in the power semiconductor market. Revenue from contactors and related products, including WaferPaks for wafer level burn-in and burn-in modules and burn-in boards for package level burn-in totaled $13.5 million or 41% of total revenue.

This compares with $2.6 million or 24% of revenue in the prior year quarter, representing more than fivefold growth. This is an important component of our business model as our installed base expands and customers increase production volumes or introduce new devices, we expect additional demand for our proprietary contactors, WaferPaks, beams and bits. These repeat purchases complement system sales and provide opportunities for ongoing production-driven revenue, although the percentage of total revenue will vary quarter-to-quarter based on product mix and shipment timing.

Non-GAAP gross margin was 48.5% compared with 37.5% in the prior year quarter, an improvement of 11 percentage points. The improvement primarily reflected higher sales volumes and a more favorable product mix. The prior year quarter included a higher proportion of lower-margin third-party probers and an automated aligner upgrade sold as part of our complete solutions. Non-GAAP operating expenses were $7.8 million compared with $5.7 million in the prior year quarter. The increase primarily reflected additional research and development investments, including engineering headcounts and materials supporting AI-related programs. These investments are focused on specific opportunities, including wafer level burn-in validation for a major AI processor customer, next-generation package level burn-in programs and memory-related development. While non-GAAP operating expenses increased approximately 36% year-over-year, revenue increased 200%, demonstrating significant operating leverage as we scale.

We recorded an income tax benefit of $0.3 million, resulting in an effective tax rate of negative 3.7%. The benefit primarily reflected discrete tax items, including disqualifying dispositions from stock options and excess tax benefits from divesting of restricted stock units. Non-GAAP net income was $9.4 million or $0.28 per diluted share compared with $0.2 million or $0.01 per diluted share in the prior year quarter. Sequentially, non-GAAP net income increased 163%. Non-GAAP net income excludes stock-based compensation and acquisition-related adjustments.

Turning to cash flow and our balance sheet. Cash used in operating activities was $3.1 million during the first quarter. We ended the quarter with $105.1 million in cash, cash equivalents and restricted cash compared with $116.5 million at the end of fiscal 2026. Operating cash flow reflected working capital investments, including purchases of long lead item -- long lead time components to support anticipated AI-related production demand and customer delivery schedules. These investments are intended to ensure we have the materials needed as production volumes increase. Importantly, we remain debt-free and our strong liquidity position provides flexibility to fund these investments while supporting continued growth.

Turning to our fiscal 2027 outlook. Based on our strong first quarter performance, substantial backlog, expanded manufacturing capacity and increased visibility into customer production requirements, we are raising our full year revenue outlook to $150 million compared with our previous guidance range of $130 million to $150 million. This outlook represents 200% year-over-year revenue growth or triple fiscal 2026 revenue. We continue to expect non-GAAP net income of at least 20% of revenue for the full fiscal year. Our increased outlook reflects current backlog, anticipated demand from existing production ramps and delivery schedules. With expanded manufacturing capacity, a strong balance sheet and a substantial backlog, we believe we're well positioned to execute against our full year outlook while investing in growth opportunities beyond fiscal 2027.

Before we open the call for questions, I would like to mention that Aehr plans to meet with investors at the NYC CEO Summit on December 15. We look forward to seeing some of you there. That concludes our prepared remarks. Operator, we're now ready to take questions.

Operator

[Operator Instructions] Your first question is coming from Christian Schwab from Craig-Hallum Capital Group.

질의응답

Christian Schwab

Congrats guys on a great quarter and outlook. Gayn, I want to come back to your commentary regarding the new wafer level benchmark and your commentary saying that you'll have an idea whether they put in high-volume orders. It sounds like sometime in the first half of calendar '27. But in particular, your commentary about the largest opportunity ever pursued by the company. I'm just wondering if you can give some relative color about the range of revenue potential that you're referring to there.

Gayn Erickson

Okay. So beyond just the sheer numbers and size, it's the tone that's coming in from them and other customers particularly related to their new devices that are coming out. And I mean like coming up in '27 and '28, thinking like kind of near term, even '27, where the new processors, there's sort of a whole new generation of different types of flavors of processors that are going out there. And I'll talk generically, my intention is trying not to be too specific about anyone. So anything I say, I could point to 2 or more customers that are saying the same thing. But there are things where there are processors that are coming out that are described as a single processor, or like a single chip, if you will. But embedded in that chip might be 4 AI accelerator chips.

There could be 12 stacks of high-bandwidth memory, another 4 or 8 optical or electrical I/O interconnect chips. That's in the near term. There's devices that are out there that have been being talked about publicly that have 12, 16 compute chips inside. The other thing is there's a shift going on with respect to just the sheer number of CPUs versus the accelerators and combinational CPU accelerator infrastructure devices that are very interesting. And so the complexity is going up, and it turns out as you mix even the same devices in the package, the burning challenges are very high because all of the devices run at different thermal levels and power levels based upon the physics of the process, and so once you attack put them together into the same substrate with the same heat sink, it becomes difficult or actually impossible to get the device to the optimum screening temperature because the one next to it is hotter or colder.

And so we're able to isolate the die at the wafer level and maintain every single device at the exact same burn-in condition in terms of temperature and voltage and current that is not even feasible to do on package level burn-in systems, not ours, not anybody's. So that whole tone of like, hey, this is really important to us. And then instead of referring to one device under test that might be on Sonoma, it might be 4 devices or 8 devices that are -- so the actual quantity is higher. So we put some numbers together just to help people understand. If you look at one of our systems, our system is a 9 wafer for the AI configuration, let's say. For every 1 million units shipped in a calendar year, that system would need to test -- you would need somewhere between 5 and 10 of our systems per million units per year. And that assumes pure linearity, meaning it ships the same amount the first day of the year as last year.

So -- and keep in mind, that's times 9. So it's 45 to 90 wafers of capacity. And if you multiply that times our ASPs that can range, including the consumables, $6 million, $7-plus million, it's $35 million to $70 million for each million units. If you look at the accelerator markets that we have been seeing, accelerator markets in 2027 alone are somewhere around 25 million units or so total. This is between the GPUs of NVIDIA and AMD and the TPUs and other ASICs from everybody. And the CPUs are double that. So the numbers are staggering and sobering for us. So why we keep pounding the drum and talking about all the infrastructure that we're putting in place to be able to ship tens of these systems per month, for example, is because of these numbers.

Now are we going to win them all? Someday maybe, but that's not what we're aiming for. But you pick up 1, 2, 3 high runners, and these are big numbers, you get a customer who's saying, I would like to be -- consider this for all my devices or most of my devices or every device you could do it. And all of a sudden, it becomes the biggest deal you've ever looked at before. So that's where those numbers come from.

Christian Schwab

And just a quick follow-up on the silicon carbide with the improving demand for vendor customer stats, et cetera. Can you remind us what your revenue to silicon carbide was in fiscal year '26 and a level of expectations that is reasonable to assume in fiscal year '27?

Chris Siu

We always just overly simplified as saying of the $50 million we did in FY '26, it was less than 10% or thereabouts. It's in silicon carbide. If I blurred GaN and silicon carbide, sometimes I do that with power semis, but it was -- the silicon carbide was that. That by contrast to silicon carbide was 90% of our revenue a couple of years earlier. So it was only about $10 million. And I think we already announced a $7 million order. I think we've already got orders maybe bigger than that already. So silicon carbide is going to grow this year. We've had some recent meetings with customers that are hopeful and optimistic. As I do say, I'm sort of in the -- we put the investments in place. We have the infrastructure. We know how to build these things in high volume, both the systems and the WaferPaks.

So I'll just announce them when they come in. But it does appear that everybody's utilization is about full, and there's people talking about continuing up the ramp of silicon carbide, mostly again for the traction inverters for EVs, some of the charging infrastructure of EVs and solar. And on the horizon more is all the data center stuff. It doesn't feel like yet that's really high volumes for -- in the immediate near term, but that's coming as well.

Operator

Your next question is coming from Jed Dorsheimer from William Blair.

Jonathan Dorsheimer

Congratulations. Great quarter. Nice to see this breakout. Gayn, maybe just I think -- and I could be wrong here, but I think previously, you had mentioned the linearity of revenues or maybe it was Chris, in terms of weighted to front half. I'm just wondering whether or not that holds true in the context of raising the full year guide to the $150-plus million. How do you expect that linearity through the year?

Gayn Erickson

Well, we -- kind of a couple of months in, a month in or so into the quarter, customers started coming back and saying, "Hey, I know I scheduled them into your second quarter. Can you pull them in?" In fact, that's kind of everybody right now ship as quickly as you can. So we were able to smooth some things out a little bit by bringing in Q1 -- but I mean, if you just do the math, if you say you're $150 million and we did 33, you've got 120-ish, something like that left the math. It feels like it's kind of -- each quarter of that 3 will grow a little bit and see it sort of smooth them out. But it feels -- this quarter will be a little bit more than last, and we think the following quarter higher than that. And then the fourth quarter, maybe the highest of the year, we'll see.

But a lot of this, too, is interesting, and I don't -- we just raised our guidance. I don't want to get ahead of ourselves, but there are real potentials both in terms of potential demand and our ability to ship as we head into the second half. So I mean, we're feeling really good about the year and candidly, heading into the following year as well.

Chris Siu

I just want to add for the second quarter will still be a very strong quarter for us, especially we have commitment to deliver our systems to our hyperscale customers. It's all in backlog. Q2, we ship it all. So if not all of them, most of it. So we still have a very strong quarter in the second quarter.

Jonathan Dorsheimer

And Chris, you bring up a good point. I mean a book-to-bill of 1.4 is great, but you typically -- particularly because you've been more of a turns business. As you kind of operate at these new levels and maybe gain your perspective on this, are you seeing any areas of your supply chain that are giving you any pause or in terms of operating at, say, at this level or you're talking about numbers that are much, much larger around this benchmark customer if they go. How are you thinking about shoring up the resilience of your supply chain? And then I have one more.

Gayn Erickson

So we've already been doing that. I mean we were talking up a storm a year ago about adding capacity, and it seemed a little odd that we were adding all this capacity if we're only doing $50 million for the year. And the answer was we could anticipate what they were saying and we were putting the infrastructure in place. So doing $33 million in the first quarter, 3x what it was the year before same quarter. We're already doing really well against it. Now candidly, having said that, we think we put infrastructure in place to be able to do above $150 million, and I'm pressing on the team to be able to put a lot more than that in place for the following year or maybe even pull in. So there are ways that we are working with our subcons. We do final assembly here of our wafer level systems. We can do final assembly of the package level systems at one of our subcons entirely now. Those are all shipping directly to our hyperscaler.

We'll continue to do the final assembly of wafer level here for the foreseeable future, but the subcons are doing more and more of the subassemblies. So today, we have x hours to do a completion of a whole wafer-level burn-in system here. By the end of the year, I think we're down to 1/3 or so of that, which would allow us to do 3x the capacity, if you will, kind of number. So we're continuing to move on all of those fronts. And we're also spending money. We're putting money down with suppliers and making sure we're buying long lead items and additional aligners and things like that to have some buffer so that we can also ship very quickly.

Jonathan Dorsheimer

Last question for me. That's helpful. Just around the benchmark customer and the wafer level burn-in, it does seem like escalation to purchase one of your systems and move it over to Asia. When do you -- do you have any expectation in terms of what that qualification process would look like that moves towards a PO? Is that something -- how should we be thinking about that?

Gayn Erickson

We do. We have, I'd say, almost all of the final details on what we call a statement of work between the 2 companies that make it very clear what is required and needed. And we were, I think, pushing on them too, to push on us to ensure that what's in there makes it qualified as a real production system, too. Units per hour, uptime, all of those normal things you'd see with production. I think for them, it was like, okay, you were able to do it here in Fremont, that's great. But will you be able to do it under the scrutiny of our OSATs and volume production? We already know what that looks like. I mean, whether -- maybe we haven't shared all of the details along the way, but we actually have FOX systems installed in Taiwan.

They happen to be very captive. I can't really go into it, but customers that have bought it and really locked it up for their specific requirements. So we know what high volume looks like on these. But this will be the first for them to really see it. And a lot of people are still just sort of getting their arms around the concept that you can actually test 9 wafers with several thousand watts a piece at a time and do it in high-volume production hands-free environment. And when they see that, the lights sort of go on like, well, how many things I could do that with. So we have a good idea what it looks like. We have a gnat chart on the schedule. There's a lot of details that are working through that need to be done to make sure that they get the facility in place. And all that's part of the reason it will -- we will be able to ship it until the beginning of the year. It's actually their stuff.

And we have some work to do. There are some things that we've shown them during the benchmark that were prototype, if you will, and validated, but we need to just show that in volume production configuration. It's -- there are some features that we can upgrade our installed base to, and they'll be seeing that for the first time in high-volume production. So I want to -- there's always a lot of work. I don't want anyone to think that this stuff is easy to do. But around January, February, March, they'll have seen the data. Should be able to be making some decisions around their capacity planning. We'll be working with them on their forecast and which devices they would want to go first on. And we're optimistic and hopeful, and we're going to try really hard. We met directly with them face-to-face. I have in the last 1 week. And we heard directly from their management that this is really important. They really want this to work. So that's usually a good sign.

Operator

Your next question is coming from Max Michaelis from Lake Street.

Maxwell Michaelis

Just first question, you guys snuck it into the prepared remarks talking about consumables revenue mix. I think you said it was 35% for fiscal year '27 is the expectation. What was that in fiscal year '26? And I can't remember if you guys have shared sort of a long-term target. But if you have thought about that, I guess, as of recently, can you share what you expect that to be in the future?

Chris Siu

Yes. '26, if I recall correctly, that's in the 20s. But for the long-term model, we always think about around the 30%. That's kind of what we discussed. But obviously, this year, Q1 was exceptional 41%, but for FY '27, we're looking at something like around 35%.

Gayn Erickson

There are scenarios where it could be 50% and our business is strong. Why I say that is part of the, I guess, pros and cons of being in the consumable business proprietary to our things is that as you have an installed base of, say, 1,000 blades or 1,000 wafers of capacity, every 1 to 3 years, as new devices come out, they'll buy all new sets of WaferPaks. So for maybe 1,000 wafers of capacity out there, which would be 100-ish of our systems, let's say, 9. So you would get maybe 300 WaferPaks even though they didn't buy new systems that year. So over time, that number adds to the -- currently, about 1/3 of every sale is WaferPaks, 2/3 is systems. So when you think 30%, you're like, oh, that's just a good growth year, not a lot of rebuy. And that is somewhat true.

So as we start to see that infrastructure -- in fact, that's the first indicator. The silicon carbide guys in the last 1.5 years have only been buying WaferPaks. And right now, they're leading with more forecast for more WaferPaks or more consumables. So maybe it could be 50%. And again, it could be 50% because you're having a bad year for systems, and it's all that. But I'm talking even in a good year, it could be 50%.

Chris Siu

Yes, I can double check for you Max.

Maxwell Michaelis

And then last question here. I mean, awesome guide, and you blew numbers out of the water this quarter took guidance up to the high end of the range from last quarter. So at $150 million. Looking out for the last 3 quarters, it's about $117 million to get there. You got about $109 million, $110 million in the backlog right now, really only calling for $7 million of orders. Help me out if I'm thinking about this wrong, but it feels like it's almost conservative that you guys are only going to get $7 million or $8 million of orders by the end of, I mean, whatever the lead time is. So help me out there. Is there anything capacity, supply chain that you guys are conservative about or help me out there?

Gayn Erickson

I think we're trying to be conservative. Yes, that's fair. I mean that is not -- that's currently not us supply limited, meaning we can ship more than that. That is current customers orders and when they say they would like it shipped, I'd say, for the most part, across the board. And so they do schedule things, they anticipated. And so we could get additional orders and still ship it this year. And we just thought $150 is a pretty nice clean number for us to be conservative at this point.

Chris Siu

Back to answer your question, WaferPak or contactless revenue was 29% last year. Consumables.

Operator

Your next question is coming from Kevin Garrigan from Jefferies.

Kevin Garrigan

Let me echo my congrats on the great results. Once this benchmark customer is kind of fully up and running with wafer level burn-in, I mean, do you expect that to accelerate adoption with other customers? Or are there any potential customers that are just waiting until they see additional customers adopt wafer level burn-in before they go online?

Gayn Erickson

I believe that this customer has very, very high visibility. And let me be clear, that imposes upside and risk, right? We're really excited about what we've shown them being able to do the proof of concept and everything here and validate it before we go. They have given us new devices. So far, we've been able to look at and say we can test all of them. I'm sure they maybe they'll give us one we can't. But for the most part, I think they're trying to help us along too, let's say. But they -- I think it's not just them. What we're seeing is there's more and more buildup and chatter in the industry and on the RAs and the trade shows talking about wafer level burn-in. And I think it's funny sometimes people talk about wafer level burn-in as a test insertion and yet we're the only ones that can do it. So they talk about like as a segment, but we're really the only ones doing it right now, and people are giving us that.

So I think that this absolutely has the potential of -- as this goes well, other people will be even less concerned. Now to be fair, the very first wafer level burn-in customer that we did is also an AI accelerator supplier. They're using TSMC wafers. TSMC was involved in the qualification and the buyout process of what that looks like. And they, we believe, have been very visible to other customers who are aware of it, either they walk by it at the OSATs or they're just aware of it. And there have been even some rumors about them going out and publicly talking about it that I'm not ready to verify yet right now. But they're very happy with us, and we're a partner to them. So that itself has helped.

In fact, there's some potential that they are the ones that caused the new benchmark customer to take interest once they understood that they were able to do it. So they all feed on each other, not dissimilar to some of the stuff that was going on silicon carbide before obviously, the EV sort of just went quiet, right? We were starting to win everything.

Kevin Garrigan

Yes, that makes a ton of sense. And then Chris, just on gross margins, they stepped up to 40.5% this quarter. How should we think about that going forward? I mean, is that kind of sustainable? And what are kind of the puts and takes there?

Chris Siu

Yes. I would say for the entire year FY '27, that is -- that's the right range. Obviously, we're going to have variations between quarters and especially Q2, we're going to ship a lot of Sonoma systems to our hyperscale customers. So Q2 would probably be lower than what we're presenting -- what we presented in Q1. But for the year, it would be plus or 0.5 percentage, plus or minus.

Operator

Your next question is coming from Edward Yang from Oppenheimer.

Edward Yang

Chris, congrats on the nice quarter. My first one, just on this again, this benchmark AI processor opportunity seems to be progressing very nicely. But can you help us understand the pacing with the Taiwan installation not planned until early next year? Was this timetable consistent with your expectations?

Gayn Erickson

Yes, pretty close. I mean I think it's within maybe 30 days or so of what we were originally thinking. So I think that's pretty consistent. They've been -- we've been meeting with them regularly working through -- looking at WaferPak designs on multiple different devices. Part of it was we weren't even sure which device they wanted to do the production benchmark or production evaluation on. They chose the one that we did the original benchmark on, but we weren't even sure if it was going to be that one or one of the others. And so I think we're working through that. And things take time. People are very busy right now, as you can imagine. And in the background, we're fully committed to work through these things.

Some of the stuff I alluded to, and I'm just not going to go into the detail here related to some of the new features that we have in the system that we've actually patented and haven't gone without NDAs, we're just not talking about some of those features. We're just -- we're implementing those in the background and doing it. So their schedule hasn't slowed us down on that at all. But again, I would think somewhere in that first couple of few months of the quarter, they'll have seen enough to trigger kind of next steps. Maybe it will go sooner, we'll see. But if we can be successful at this and they continue to engage with us on the different devices they're talking about, that's fantastic.

Edward Yang

And for my follow-up, just on competition, one of the larger system-level test suppliers recently introduced what they call AI burn-in capabilities. It doesn't seem really comparable to what Aehr offers. But maybe talk us through how their approach differs from FOX's accelerated live screening.

Gayn Erickson

Okay. Well, they're within -- so first of all, as we understand what they're -- when people talk about screening and burn-in, all semiconductors are tested. They're tested good from bad, and they may be tested for how good they are, speed grade or whatever. And there's platforms out there, the 2 dominant platforms for all the AI high-performance computing platforms are the 93,000 from Advantest. I always remind people that, that came from my old company from HP, Agilent and Verigy. One of the greatest acquisition ever was when they bought Verigy because it's done so very well for them. And the Ultraflex from Teradyne, who's been gaining market share. But those are the 2 players. If you told me it was 99%, I believe you, okay?

All testing is done at wafer and at a package level. When the device is good at the very end, right, a customer will take that, put it into a computer, put that into a rack and then ship it to the end customer, a data center. But what they're finding is those devices would fail early, days, weeks later. So what first was happening is the data center, the suppliers of the racks, computer Micro and Super Micro and these guys, they would actually do a burn-in of the rack, run it for a week to screen out infant mortality, they would call it, of the processors that were failing. And there were public comments by Jensen and others to say, listen, in a big white paper that was done by Meta that was talked about the failures that were happening in the data centers and now the people needed to implement burn-in and screening.

These are old -- these are techniques that have been used for years and years with the likes of Intel and AMD for screening their processors, but all of the new processor companies were fabless going through TSMC and didn't have any of the infrastructure in place. And so people are scrambling what do I do because I don't want to do it at the rack level. So one of the first things they did is they actually looked at what are called SLT machines. So in addition to the final test systems I described for Advantest and Teradyne, there's another insertion called an SLT insertion, which for all intents and purposes is to boot up the device in its native application. Think of it on its own motherboard. And there's companies like Han, Precision, Chroma, AEM, Teradyne has just introduced one, which is what I think you're alluding to with their Titan, I think Titan or Tritan, I forget which one, platform, which is an SLT machine, which is at speed, high speed, intended to look and feel like the actual application.

And that has -- they have great thermals on it, and they have ATE performance, but done for a specific insertion. Now you could do a burn-in on that thing. They typically are very expensive platforms that have high-speed digital signals, et cetera. And -- but one of the challenges is the burn-in condition for accelerating these infant mortalities is not the same as the user mode. We actually bring the devices up to much higher temperatures than they run in the normal user environment to basically strain them and age them. So within 3 hours, we might make the device look like it's several months old or more. And we can correlate to screen out those infant mortality rates by bringing the devices up to very high voltages and very high temperatures extremely accurately and basically precisely age them 3 months, not 3 years, okay?

So we don't want to wear them out. And so that's what we do with our burn-in insertion. But there have been several companies out there that have talked about trying to use SLT derivatives, system-level test derivatives to try and do the burn-in. When we're engaged in package level burn-in deals right now, which there's a number of them going on, people are not talking to us about the SLT alternatives. They're looking for what they would call a real burn-in system like Sonoma that's aimed at addressing those high voltages and high temperatures and massive parallelism at low cost. So we think we have the best package level burn-in system on the market with Sonoma, okay? But to be fair, there are other customers out there, other companies that make them. And in some cases, one of the challenges is they might make one for their own use, and they might have an advantage because they have some cost because they build it themselves. They may not be able to hit our performance, but maybe they're willing to make some of the other technical trade-offs.

So in package level burn-in, we actually see that business growing and growing, and we think that the real competition isn't really an SLT machine, okay? But more interestingly, the real competition to package level is the FOX system from us. The wafer level system provides capabilities both technically. We can do things at wafer level that nobody, including us, can do a package level. And that -- we're proving that to customers through these benchmarks, which would get a customer to want to shift from package level to wafer level. And in that case, the only alternative, unless you want to violate our IP, and we ferociously will defend it is to try and use a standard prober and an ATE system coming down and trying to test as many devices you can at a time to try and get the cost effectiveness down.

So what we're able to do is we're often testing 1/4 of a wafer at a time of all the processors, which people think is nuts, considering that the 93K and the UltraFlex are typically testing one device at a time. So we're testing 1/4 of all the die in parallel on 9 die and 9 wafers at a time. Sometimes, by the way, we may test 1/8 or something, but we're able to do that very cost effectively. And by doing it at wafer level, the cost to test at wafer level is a fraction of the yield improvement by moving it to wafer level. So we think that ultimately, that's where people will shift. And it's exciting and scary as you throw a party and what happens if every single person comes, but that's kind of where we're at. And it's -- we have a great front row seat using the Sonoma platform to have a conversation about people's next-generation devices that, in some cases, they say, you know what, we need to go to FOX by the time that part comes out. So we have a different -- we're playing with a different set of deck of cards, if you will, than our competitors are.

Operator

Our next question is coming from Larry Chlebina from Chlebina Capital.

Larry Chlebina

You mentioned that you're getting an opportunity on the HPC memory. And I think I heard you say via the base of the HPC stacks. Did I hear you correctly?

Gayn Erickson

Yes. On the high-performance memory, HBM, HBM DRAM, there's a -- one of the features of the HBM standard to the JetX standard, and there's new devices coming out that are even outside of the normal JetX standard, where the DRAM devices are stacked on top of a logic interface chip that is done on a high-speed logic process. So like a TSMC would build it and then a Micron or Hynix would build the DRAM and put it on top of that, for example, or Samsung. And that a high-speed interface and other logic, including some built-in self-test features. But that device itself is subject to needing to be screened and bur in and have some sort of infant mortality because if it fails later, you might have 12 DRAM die on top of it, $500 or $1,000 worth of die.

Larry Chlebina

And is that base die opportunity, is that coming through the paid evaluator? Is that one of the opportunities down the road?

Gayn Erickson

We've heard about this opportunity from several different flavors of suppliers. How is that?

Larry Chlebina

Okay. You also at the CEO Summit at SEMICON West, you were telling us about this breakthrough that you had in the -- being able to control the temperature of each device on the wafer within a 0.5 degree centigrade, which is phenomenal. And that breakthrough involves a super-duper new WaferPak that you developed that you recently patented. Will this -- or this next step in the pay-to-value wafer, will that involve that full complement of 9 super-duper WaferPaks? What are you calling them?

Gayn Erickson

Super-duper WaferPaks. No. We've been calling it the die TC for die temperature control. And it has the ability to actually independently manipulate the thermals of each of the die under test at the wafer level, which is nuts. People have never even contemplated that before. And it's integrated into our WaferPak and custom design for each of the wafers under test. So it's a pretty cool feature.

Larry Chlebina

And that will be in that full production system that you're going to start the first of the year in the evaluation?

Gayn Erickson

Yes, it will be.

Larry Chlebina

And well, that seems like a heck of a way to differentiate, actually be able to show that you can get a full burn-in system, actual burn-in and not burn out any devices that have a thermal runaway potential because of leakage and so on, as you described it.

Gayn Erickson

So it's actually even -- so you got it right, but it may be the opposite of -- if I describe it differently, it may be even more impactful. So if you go into thermal runaway, what happens when these devices heat up, they actually drop their resistance, which draws more current and then they heat up at the square of the current. You can see the problem. When they heat up, they drop more resistance, drop more current. In our world, that's called thermal runaway. The device immediately within milliseconds or seconds will go into thermal runaway and burn itself up and actually burn up. That's bad, okay?

So you're saying we can avoid doing that. That is true. But what's actually more critically important, if you look at some of these packages that are coming out, they may have multiple die inside, but they're all tied to the same substrate. The die don't all burn in and have the same leakages. So they're all going to generate different power and different heat, self-heating. The way people will do that is they actually self -- they look at the hottest die and control the temperatures on that one to avoid it from going in the thermal runaway. But when they do that, the coldest die isn't at the burn-in condition. And their customers are like, well, what are we going to do? So that device is actually not burned in optimally. So those -- that infant mortality or that failure rate will sneak through all the way to the data center. The physics. That's all we can do.

Larry Chlebina

The customer is actually paying for a burn-in and actually isn't getting it on the full package. So still suffering from mortality failures and use, you pay for it.

Gayn Erickson

And if you take that to an extreme, which isn't that far out in time, you just simply won't be able to have an effective burn-in. And you're just going to have to say, well, I'm going to have failures in the data center. And some people say, you know what, those data centers are going to be a long ways away, and you can't have failures, right? And so we start using the term internally here. Terrestrial data centers are a lot easier than some of the new ones that are coming up. And you start looking at these real-time edge inference devices where the failure is literally life and death. And so we're getting to these scenarios where you have to do the burn-in and people are like, but how are we going to do it, right? And so it's a great time to be in the burn-in business.

Larry Chlebina

It sounds like your wafer level burn-in system with your super-duper or die TC WaferPak ought be a slam dunk.

Gayn Erickson

I agree with you.

Larry Chlebina

How do you sell it? Sell it. All right. Good job. That's all I had.

Operator

That concludes our Q&A session. I will now hand the conference back to management for closing remarks. Please go ahead.

Gayn Erickson

All right. Thank you, operator. And thank you, everybody. We really appreciate you guys joining in for this. This is an interesting one. We were avoiding election day tomorrow, so didn't anticipate being opposite of Apple. So we appreciate those that are listening in or catching up afterwards. And as always, if you have any questions, please get in touch with IR, Chris or I, and we'll be happy to respond to you as fast as we can. Thank you very much. Bye-bye.

Operator

Thank you. Everyone, this concludes today's event. You may disconnect at this time, and have a wonderful day. Thank you for your participation.