Aehr Test Systems ($AEHR) earnings deep dive: the $100.6M backlog, 3x fiscal 2027 revenue guide, FOX-XP and Sonoma burn-in systems, unnamed AI customers, revenue upside, and bull/bear risks.
More Macroplane supply-chain guides · Baskets · Investment theses · Macro trends
Aehr Test Systems ($AEHR) just guided fiscal 2027 revenue to $130–150 million after producing only $50 million in fiscal 2026. The headline is 160%–200% growth. The more important number is $100.6 million of effective backlog: customers have already placed enough orders to cover roughly two-thirds to three-quarters of the guide. The July 14 earnings release also showed that $AEHR is no longer mainly an electric-vehicle silicon-carbide story. AI processors and silicon photonics supplied more than 80% of Q4 revenue, record bookings reached $60.7 million, and the company swung back to quarterly profit.
This is what $AEHR actually makes, what can be established about its unusually secretive customer base, how the revenue math works, and where the bull case can still break.
👉 Start with the live company page: Aehr Test Systems ($AEHR) sits in Macroplane's Semiconductor Test Equipment category. Its newest growth vectors run directly through the Silicon Photonics & Optical Interconnects trend.
| Fiscal Q4 2026 | Result | Why it matters |
|---|---|---|
| Revenue | $18.8M | Up 34% from $14.1M a year ago; about 5% above consensus |
| Non-GAAP EPS | $0.11 | Consensus expected a $0.01 loss |
| GAAP net income | $1.4M | Reversed a $2.9M year-ago loss |
| Bookings | $60.7M | Record quarter; more than 3× reported revenue |
| Backlog at May 29 | $80.6M | Up from $15.2M at the prior fiscal year-end |
| Effective backlog | $100.6M | Includes $20M of orders received after May 29 |
| Fiscal 2027 revenue guide | $130–150M | 2.6×–3.0× fiscal 2026 revenue |
| Fiscal 2027 profit guide | 18%–22% | Non-GAAP pretax net income as a share of revenue |
The stock closed July 14 at about $72 and then traded roughly 29% higher after hours. That reaction was mostly about the guide and backlog, not the modest $0.9 million quarterly revenue beat.
There is an important reality check inside the same report. Full-year revenue still fell 15% to $50 million, full-year GAAP net loss widened to $7.1 million, and non-GAAP gross margin fell to 38.5% from 44%. This is a forward-order story emerging from a weak year, not an uninterrupted growth record.
$AEHR sells machines that deliberately stress semiconductors with heat, voltage, current, and workload before those chips reach customers. This is burn-in: forcing early-life defects to appear in a controlled factory environment instead of inside an AI server, optical transceiver, or vehicle.
Burn-in becomes more valuable as chips become larger, hotter, and more expensive to package. If a bad compute die is discovered only after it has been combined with other compute dies and high-value memory, the manufacturer may have to discard the entire package. $AEHR's pitch is simple: find the weak die earlier, when it is cheaper to lose.
*$AEHR's FOX-XP multi-wafer burn-in system. Image and product specifications: Aehr Test Systems ($AEHR).*
The FOX-XP is the flagship wafer-level platform. A production configuration can burn in as many as nine 300 mm wafers in parallel, using $AEHR's proprietary WaferPak contactors to make electrical contact across the wafer. $AEHR says a FOX-XP can dissipate up to 3,500 watts per wafer and test logic, memory, photonic, and power devices.
This is the growth platform for AI processors, silicon photonics, silicon carbide, gallium nitride, and—if current development work succeeds—future NAND and high-bandwidth memory applications.
*A FOX WaferPak full-wafer contactor, customized to a customer's device. Image and product specifications: Aehr Test Systems ($AEHR).*
The WaferPak is the device-specific interface between the test system and the customer's wafer. New chip designs and capacity additions need new contactors, so the installed machine base can generate repeat consumables revenue. Contactors were $5.8 million, or 31% of Q4 revenue. Management told analysts that roughly 30% is a reasonable consumables assumption for fiscal 2027 when WaferPaks and Sonoma burn-in modules are combined.
That installed-base model matters. A one-time FOX sale can lead to years of contactor orders as customers add devices, designs, and wafer capacity.
*The Sonoma high-power package-level burn-in system. Image and product specifications: Aehr Test Systems ($AEHR).*
The Sonoma system tests packaged AI accelerators, GPUs, CPUs, and other high-performance processors. Its newest configuration supports up to 2,000 watts per device with individual thermal control and liquid- or air-cooled options. $AEHR gained Sonoma through its 2024 acquisition of Incal Technology, adding package-level burn-in to the wafer-level FOX franchise.
That acquisition created the company's two-sided position: $AEHR can sell a customer package-level burn-in today, then pursue wafer-level burn-in when a future multi-die design makes late-stage failures too expensive.
The honest answer is that $AEHR does not disclose their names. It identifies current customers by market and program, and internet guesses should not be presented as facts.
What the Q4 call did disclose is that three customers each contributed more than 10% of quarterly revenue: two serve the AI market and one focuses on data-center optical transceivers. The current customer map looks like this:
| Customer description disclosed by $AEHR | Product/program | What is confirmed |
|---|---|---|
| Lead hyperscale AI customer | Sonoma package-level burn-in | Placed a record $41M follow-on production order in April for custom AI ASIC burn-in; a second, higher-power device is planned to ramp; management estimates about $50M of Sonoma revenue in the fiscal 2027 midpoint |
| Lead AI processor wafer-level customer | FOX-XP + WaferPaks + auto aligners | Placed a $14M order in February, doubled its $AEHR systems, and moved all production screening for the program from system-level to wafer-level burn-in |
| Lead silicon-photonics customer | Automated FOX-XP cells | Ramping high-volume optical I/O and data-center interconnect production; placed another follow-on system order on July 9 |
| New major silicon-photonics customer | Two nine-wafer FOX cells + two FOX-NP systems | Described as a global networking-products leader; forecasting more systems as hyperscale deployments ramp |
| Lead silicon-carbide customer and a large automotive OEM | WaferPaks on installed FOX systems | Part of more than $8M of SiC orders announced July 14; the lead customer is adding EV capacity, particularly for China programs |
These descriptions may overlap with the three greater-than-10% Q4 customers, but $AEHR does not provide enough detail to map each percentage to a named account. That opacity is both a commercial reality and a material investment risk: investors can track purchase orders and programs, but they cannot independently evaluate each customer's budget, product schedule, or credit quality.
The hyperscaler clues can be tested against the public processor roadmaps. Our separate investigation—Aehr Test Systems ($AEHR)'s Mystery Customer: Google ($GOOGL) TPU, AWS Trainium ($AMZN), or Microsoft Maia ($MSFT)?—finds that Alphabet ($GOOGL) is the strongest circumstantial fit and Amazon ($AMZN) is a credible alternative, but neither identity is confirmed.
The official guide is unusually large relative to $AEHR's history, but it is not built entirely on hope.
The $100.6 million effective backlog equals 77% of the $130 million low end and 67% of the $150 million high end. Put differently, $AEHR needs another $29.4–49.4 million of recognized revenue beyond today's backlog to land inside the range.
Backlog is not revenue—shipments can move, configurations can change, and orders can be delayed or cancelled—but this is much stronger visibility than a forecast built only from sales conversations.
Management's rough fiscal 2027 mix is:
At the $140 million midpoint, 70% AI implies roughly $98 million of AI-related revenue. Management separately framed Sonoma package-level revenue at about $50 million, leaving a substantial wafer-level AI component. A 30% consumables mix would imply about $42 million from WaferPaks and Sonoma burn-in modules, although that is a mix assumption rather than formal line-item guidance.
Three opportunities could lift the ceiling:
That is the right way to frame the revenue potential: $130–150 million is the supported one-year case; new AI benchmarks and memory are options, not forecast revenue.
$AEHR has outsourced Sonoma production to a Southeast Asian contract manufacturer with capacity for more than 20 systems per month, while Fremont can support a mix of roughly 20 wafer- or package-level systems per month after further staffing and supply-chain work. Management said the company is not capacity-limited at $150 million.
The call also supplied tempting theoretical math—FOX systems with WaferPaks can approach $5 million and high configurations can exceed that—but factory capacity should never be mistaken for customer demand. The relevant ceiling is orders, qualification, and delivery timing, not what the assembly floor could produce in a perfect month.
$AEHR has crossed from qualification into production. The lead wafer-level AI customer has doubled its installed systems and eliminated system-level screening for the program. The lead hyperscaler placed a $41 million Sonoma order. Multiple silicon-photonics customers are ordering repeat systems. Those are production signals, not lab demos.
The business now covers both sides of the burn-in decision. Sonoma handles finished AI packages; FOX handles wafers before packaging. As advanced packages contain more compute dies and expensive memory, the economic case for finding failures earlier gets stronger. $AEHR can win whether a customer stays at package level or migrates upstream.
Consumables can turn equipment wins into recurring revenue. Device-specific WaferPaks, burn-in modules, and sockets represented roughly 31% of Q4 sales. Every new design and capacity ramp can pull more of these higher-repeat products through the installed base.
Operating leverage could be dramatic. At the official $130–150 million range and 18%–22% non-GAAP pretax profit target, the guide implies roughly $23–33 million of pretax non-GAAP income. Fiscal 2026 produced only $0.9 million of non-GAAP net income. If the mix and shipment schedule cooperate, revenue can grow much faster than fixed costs.
The guide excludes two potentially large markets. The second AI benchmark customer and any memory revenue are mostly outside the range. A production decision from either would add a new customer and reduce concentration while increasing the revenue ceiling.
Customer concentration is extreme. Three unnamed customers each supplied more than 10% of Q4 revenue, about 70% of fiscal 2027 sales are expected to come from AI, and most of the guide comes from current customers. One delayed accelerator or optical program can move an entire quarter.
Execution must jump almost 3× in one year. The second hyperscaler device has already been delayed, the bulk of Sonoma shipments are concentrated in fiscal Q2, and management openly describes the business as lumpy. Building the systems is only one step; installation, acceptance, and revenue recognition have to land on schedule.
The trailing business is still loss-making. Fiscal 2026 revenue fell 15%, GAAP loss widened to $7.1 million, and full-year gross margin declined. The excellent Q4 exit rate does not erase the weak first nine months.
The balance-sheet strength came with dilution. $AEHR raised about $100 million primarily through at-the-market share sales during fiscal 2026. Cash reached $116.5 million and gives the company working capital to fill large orders, but existing shareholders paid for much of that cushion through a larger share count.
Some of the best stories are still evaluations. The new top-tier AI benchmark, high-bandwidth memory, NAND, gallium nitride, and the hyperscaler's third device may all become material. None should be valued like booked production revenue today.
Competition and intellectual property remain live risks. $AEHR is suing SemiE, now operating as NEXUSTEST, over alleged patent infringement in China. $AEHR says two Chinese patents were upheld and that it recently won a Taiwan SiC customer against that competitor, but the litigation adds cost and shows that lower-cost competition is pursuing the same market.
The post-earnings valuation leaves less room for error. A roughly 30% after-hours move immediately capitalizes part of the guide before $AEHR ships the backlog. The higher the stock re-rates on the forecast, the more violently any order delay, margin miss, or customer-program change can be punished.
The next four checkpoints matter more than another press-release headline:
$AEHR makes semiconductor reliability test and burn-in equipment. Its FOX-XP systems stress and test entire wafers before chips are packaged; its proprietary WaferPak contactors connect the tester to a specific wafer design; and its Sonoma systems burn in high-power packaged AI processors. The equipment is used for AI chips, silicon photonics, silicon carbide and gallium nitride power devices, memory, and other high-reliability semiconductors.
$AEHR does not disclose their names. In fiscal Q4 2026, three customers each represented more than 10% of revenue: two targeted AI and one supplied data-center optical transceivers. The largest disclosed program is a $41 million Sonoma order from an unnamed hyperscale data-center customer for custom AI processors.
Management guided to $130–150 million, up 160%–200% from fiscal 2026. Effective backlog was $100.6 million after the quarter, covering 67%–77% of that range. The guide assumes roughly 70% AI revenue, 15%–20% silicon photonics, and no memory revenue.
$AEHR was profitable in fiscal Q4, with $1.4 million of GAAP net income and $3.6 million of non-GAAP net income. It still lost $7.1 million on a GAAP basis for the full fiscal year. For fiscal 2027, management targets non-GAAP pretax income equal to 18%–22% of revenue.
AI processors are large, power-dense, and increasingly assembled from multiple compute and memory dies. A failure found after packaging can destroy the value of every good component in the package. Wafer-level burn-in screens weak dies earlier, while package-level systems stress finished processors before deployment in expensive data-center infrastructure.
This is research and education, not investment advice. Backlog can be delayed or cancelled, customer identities are undisclosed, and forward guidance is not guaranteed. Do your own work.
$AEHR makes semiconductor reliability test and burn-in equipment. Its FOX-XP systems stress and test entire wafers before chips are packaged; its proprietary WaferPak contactors connect the tester to a specific wafer design; and its Sonoma systems burn in high-power packaged AI processors. The equipment is used for AI chips, silicon photonics, silicon carbide and gallium nitride power devices, memory, and other high-reliability semiconductors.
$AEHR does not disclose their names. In fiscal Q4 2026, three customers each represented more than 10% of revenue: two targeted AI and one supplied data-center optical transceivers. The largest disclosed program is a $41 million Sonoma order from an unnamed hyperscale data-center customer for custom AI processors.
Management guided to $130–150 million, up 160%–200% from fiscal 2026. Effective backlog was $100.6 million after the quarter, covering 67%–77% of that range. The guide assumes roughly 70% AI revenue, 15%–20% silicon photonics, and no memory revenue.
$AEHR was profitable in fiscal Q4, with $1.4 million of GAAP net income and $3.6 million of non-GAAP net income. It still lost $7.1 million on a GAAP basis for the full fiscal year. For fiscal 2027, management targets non-GAAP pretax income equal to 18%–22% of revenue.
AI processors are large, power-dense, and increasingly assembled from multiple compute and memory dies. A failure found after packaging can destroy the value of every good component in the package. Wafer-level burn-in screens weak dies earlier, while package-level systems stress finished processors before deployment in expensive data-center infrastructure.