$AEHR EXECUTIVE CALL SUMMARY: Aehr Test Systems (04/07/26)
The central conclusion from the call is that Q3 FY26 was a demand inflection quarter, not a clean earnings inflection quarter. Reported revenue remained weak at $10.3 million, down 44% y/y, and non-GAAP net loss was $1.5 million, or $0.05 per diluted share. Against that, bookings surged to $37.2 million, quarter-end backlog reached $38.7 million, and effective backlog rose to $50.9 million after another $12.2 million of bookings in the first 5 weeks of Q4. On the official numbers, book-to-bill was roughly 3.6x. Management reiterated 2H FY26 revenue guidance of $25 million to $30 million and 2H non-GAAP EPS guidance of -$0.09 to -$0.05, but qualitatively tightened the outlook by saying FY26 revenue should finish on the high side of the prior $45 million to $50 million range and 2H bookings on the high side of the prior $60 million to $80 million range. The correct read-through is that end-demand and order momentum improved materially, while reported revenue, margin capture, and operating leverage remained under pressure. 
A more precise framing is that the quarter widened the gap between demand formation and revenue realization. Through the first 9 months of FY26, revenue was $31.2 million versus $44.9 million in the prior-year period, and non-GAAP EPS was -$0.09 versus +$0.16. Non-GAAP gross margin was 36.5% in Q3, which was 620 bps below FY25 Q3’s 42.7%, although it did recover 661 bps from the 29.8% level in Q2. By contrast, Q3 bookings increased 54% y/y versus FY25 Q3 and 500% sequentially versus FY26 Q2. Effective backlog of $50.9 million now equals roughly 86% of all FY25 revenue. The debate has therefore shifted away from whether the company has demand and toward 3 narrower questions: how much of backlog converts in the next 1 to 3 quarters, how much of that conversion is system revenue versus consumables, and whether mix can support the promised return to profitability. 
GUIDANCE AND WHAT CHANGED
The most important guidance point is that the formal numbers did not rise. Q2 had already reinstated 2H FY26 revenue guidance of $25 million to $30 million and 2H non-GAAP EPS guidance of -$0.09 to -$0.05, while indicating expected 2H bookings of $60 million to $80 million. Q3 kept those numerical ranges intact. What changed was management’s confidence inside the ranges. The call explicitly moved revenue expectation to the high side of the prior FY26 range and bookings expectation to the high side of the prior 2H bookings range. That is a meaningful tonal improvement, but not a headline raise. Since 9M FY26 revenue is $31.2 million, the official full-year revenue range still requires roughly $14 million to $19 million of Q4 revenue, which implies a significant sequential step-up from Q3. The practical implication is that the call improved FY27 setup more than it de-risked FY26 execution. 
A second, more subtle guidance change sits inside the wafer-level AI benchmark commentary. In the January 8 Q2 release, management had said it expected to complete data collection that quarter on the benchmark with a top-tier AI processor supplier. On this call, management disclosed that the program “has taken longer than we originally expected” because of “a technical misunderstanding on the clock configurations,” and that WaferPaks are now being redesigned while additional data is collected over the next several months. That is not a thesis break, but it is a real timing slip versus the prior quarter’s implied schedule. The positive implication is that the program is still active. The negative implication is that some of the highest-value wafer-level opportunities are still engineering-sensitive and harder to underwrite on a quarter-by-quarter basis. 
ORDER BOOK QUALITY AND REVENUE CONVERSION
The record backlog is important, but its revenue timing is uneven. The February 26 $14 million AI wafer-level order is scheduled to ship within 6 months, which means a portion is near-term and a portion likely falls into the transition period or early FY27. The March 31 silicon photonics order from a major new customer is scheduled to ship in fiscal Q4 ending May 29, 2026 and is therefore the clearest immediate revenue candidate. By contrast, the February 11 next-generation Sonoma AI ASIC order is scheduled for summer 2026 delivery, and the March 3 silicon photonics follow-on order is scheduled for 2H calendar 2026. The analytical consequence is straightforward: the backlog surge is real, but a meaningful share of the announced February-March orders is structurally more supportive of transition-period and FY27 revenue than of FY26 Q4 revenue. 
That timing asymmetry is especially important because Aehr’s own FY25 10-K emphasizes that most customer purchase orders are subject to cancellation or rescheduling with limited penalties and that backlog at any particular date is not necessarily indicative of actual sales in a succeeding period. The same filing also highlights the company’s persistent concentration risk: 77% of FY25 revenue came from the top 5 customers, with 2 customers accounting for 39% and 15%, respectively. In other words, the quarter substantially improved confidence in demand, but did not eliminate the company’s historical sensitivity to a small number of order movements and shipment deferrals. For an investment committee, that distinction is critical. Backlog should be treated as strong directional evidence, not as fully contracted revenue. 
AI PACKAGE-LEVEL BURN-IN
Package-level burn-in on Sonoma remains the fastest and cleanest monetization vector in the AI opportunity set. The company has now announced an initial production order for a lead hyperscaler’s next-generation higher-power AI ASIC, and management stated on the call that it also expects a “significant near-term follow-on order” for the current device family. The qualitative Q&A color was also important. Management said that many AI ASIC suppliers still do not run production burn-in today, but that “the common theme is they’re all moving to burn-in.” That comment matters because it implies the current Sonoma opportunity is not confined to one customer or one device cycle. It suggests a category shift in high-power AI silicon reliability screening as thermal density and current levels rise. 
The other important package-level read-through is that management increasingly framed Sonoma not as a standalone niche tool, but as the first monetization step in a larger account lifecycle. Qualification burn-in often starts at package level because customer test modes and production flows are already understood there. Wafer-level is economically superior in many cases, but more design-for-test dependent and more likely to require engineering iteration. In that context, Sonoma appears to be the lower-friction entry point, with FOX wafer-level potentially following on later generations. That interpretation was reinforced by management’s disclosure that the lead hyperscaler is already discussing DFT for a 3rd device so that wafer-level burn-in can be considered on the roadmap. The implication is that package-level and wafer-level should not be modeled as competing products. The more likely pathway is package-level first, wafer-level later where economics justify pre-package screening.
Manufacturing capacity also appears materially less constrained than in prior periods. The company said it expects to begin shipping Sonoma systems this quarter from a newly upgraded contract manufacturing facility with capacity for more than 20 additional systems per month, in addition to Fremont. If that capacity comes online as described, supply should stop being the binding constraint for Sonoma. The gating factors would then shift to customer qualification timing, module availability, and end-customer ramp cadence. That is constructive because it increases the probability that FY27 revenue is limited by demand conversion rather than internal manufacturing bottlenecks. 
AI WAFER-LEVEL BURN-IN
AI wafer-level burn-in remains the potentially larger long-term economic vector, but it is also the segment with the greatest execution complexity. Management was emphatic, stating that “AI wafer-level burn-in is really hot right now,” and the $14 million follow-on order from the lead AI wafer-level customer clearly demonstrates that the company has moved beyond proof-of-concept with at least 1 customer. The strategic logic is compelling and was articulated clearly on the call: screening high-power AI devices before they are combined with expensive substrates, HBM stacks, photonics, or other die inside advanced packages can save substantial downstream cost and avoid multi-die package scrap. That logic becomes more powerful as package content and packaging cost rise. 
However, the call also provided the clearest evidence to date of what can slow this adoption curve. The top-tier benchmark did not fail, but it did slip. The explanation was operationally specific rather than market-driven: “a technical misunderstanding on the clock configurations” forced redesign of the WaferPak approach. That makes the right analytical conclusion neither fully bullish nor fully bearish. The conclusion is that wafer-level AI remains highly valuable and apparently technically viable, but schedule predictability is lower because each program can involve customer-specific DFT assumptions, custom contactor work, power/current constraints, and a learning curve on both sides. The stock can support a higher multiple on wafer-level AI only after more of these benchmark programs convert into repeatable production orders, not merely because the TAM is large.
The Q&A also sharpened the adoption model. Management effectively described a market still in discovery mode. Some customers start by assuming package-level burn-in is the only option, then pivot toward wafer-level once they see the economic trade-off. Others are already asking what DFT changes should be embedded into next-generation devices so wafer-level screening becomes easier. The practical implication is that wafer-level AI revenue should be expected to compound in a stair-step pattern rather than in a straight line. Large orders are possible, but they are likely to remain lumpy until customer design flows standardize around wafer-level burn-in requirements.
SILICON PHOTONICS
Silicon photonics was arguably the cleanest positive surprise in the call. The March 31 order from a major new customer was not a narrow pilot. It included both engineering qualification and high-volume production tools, and it is scheduled to ship in fiscal Q4. Management also discussed a follow-on order from the lead silicon photonics customer and described integration into fully automated, lights-out operation with autonomous guided robots. This combination of new-customer acquisition, production intent, automation integration, and near-term shipment timing makes silicon photonics look more de-risked than several of the still-evolving AI wafer-level benchmarks. 
The segment matters for 3 reasons. First, it gives the wafer-level business a near-term growth leg that is not entirely dependent on AI processor qualification cycles. Second, it is inherently aligned with the same macro driver powering AI infrastructure buildout, namely the shift from copper to optical interconnect as bandwidth, power, and thermal constraints intensify in hyperscale data centers. Third, it appears to be moving more quickly from evaluation to production than some AI wafer-level programs. Management’s tone on silicon photonics was notably less exploratory and more operational. The company now has both an installed base and new-customer proof points in this market. For valuation purposes, silicon photonics now deserves to be treated as a real growth contributor rather than residual optionality. 
POWER SEMICONDUCTORS AND MEMORY
Power semiconductors and memory should be treated as separate buckets. On power, management was conspicuously disciplined. There was constructive commentary on GaN and a new SiC customer in Taiwan, but the call also explicitly said the company is “not yet counting on significant revenue from this segment to return yet.” That is an important change in narrative emphasis. Historically, Aehr’s equity story was heavily tied to SiC recovery. By FY25 Q3, management had already disclosed that SiC wafer-level burn-in, which represented over 90% of fiscal 2024 business, was tracking to less than 40% of fiscal 2025 business, while AI burn-in had already reached over 35% in its first year. The present call confirmed that the center of gravity has moved decisively away from waiting for a SiC rebound and toward AI, silicon photonics, and broader data-center infrastructure demand. 
On memory, the tone was more strategically interesting than near-term bankable. Management said it achieved the correlation requested by a key memory supplier and is now discussing test system specifications and a development agreement. The call also expanded the memory narrative from flash into HBM-adjacent opportunity. That is conceptually powerful because HBM sits inside the same advanced packaging ecosystem where wafer-level screening becomes more economically attractive. Yet the timeline described on the call still looks long. Management’s own language suggested potential fiscal 2027 orders and fiscal 2028 ramps after a 12 to 18 month development effort. That places memory firmly in the category of meaningful upside optionality, but not something that should be heavily capitalized into the next few quarters of estimates.
MARGIN, CASH FLOW, AND CAPITAL STRUCTURE
The margin profile improved sequentially but remained well below the level needed to validate a durable earnings inflection. Q3 non-GAAP gross margin was 36.5%, versus 29.8% in Q2 and 42.7% in FY25 Q3. Q1 FY26 had already been 37.5%, so Q3 does not yet establish a new upward trend; it mainly confirms that Q2 was an unusually weak trough. Non-GAAP operating expenses were $6.3 million in Q3, essentially flat y/y but above Q2. Management attributed ongoing spend to AI benchmark and memory projects. The economic model still works if revenue scales, but it is not currently showing material operating leverage. 
The consumables thesis remains central to the margin case, but evidence is still incomplete. Management said this year has been light in consumables, particularly WaferPaks, and described that as an outlier tied to customers growing into previously installed system capacity. Management also reiterated the longer-term goal that consumables should exceed 30% of revenue and lift margins as the installed base expands. That framework is plausible, especially if Sonoma and FOX deployments become more production oriented. However, the reported Q3 mix still does not fully demonstrate that operating model. The company is currently asking the market to underwrite future consumables normalization rather than showing it cleanly in current results.
Management also said it expects to return to non-GAAP profitability in Q4. That is directionally reasonable, but the room for error is limited. Officially reported 9M FY26 non-GAAP EPS is already -$0.09. That means the Q4 outcome needs to cluster near breakeven or modest profitability for the full-year framework to hold. The result is that Q4 revenue timing, shipment mix, and consumables attachment matter disproportionately. A modest revenue shortfall or an adverse mix outcome could still prevent the quarter from delivering the desired profitability signal. 
Liquidity risk has improved, but not because the operating model has already self-funded. Cash, cash equivalents, and restricted cash rose to $37.1 million from $31.0 million sequentially. On the call, management said the increase was primarily due to ATM issuance and disclosed that the $40 million ATM has now been fully utilized, with more than 1.13 million shares sold at an average price of $35.38. Relative to the 29.877 million shares outstanding at May 30, 2025, that implies roughly 4% dilution before considering other share movements. This is an acceptable trade if FY27 growth materializes, because it removes near-term balance-sheet pressure ahead of a potentially larger production cycle. It is less attractive if bookings remain lumpy and revenue conversion disappoints. 
The fiscal-year-end change should be viewed as a modeling issue rather than a fundamental issue. The board changed the year-end from the Friday nearest May 31 to the Friday nearest June 30, which creates a 28-day transition period from May 30 to June 26, 2026 before FY27 begins on June 27. The operational business impact should be minimal, but comparability will become temporarily more difficult. Revenue phasing, growth rates, and shipment timing around the transition month will require extra care in model construction and in interpreting quarter-to-quarter changes. 
INVESTMENT IMPLICATIONS
The bullish interpretation of the call is that Aehr now has enough concrete evidence to support a credible FY27 reacceleration thesis. The evidence base is materially stronger than it was 90 days ago: Q3 bookings of $37.2 million, effective backlog of $50.9 million, a $14 million AI wafer-level follow-on order, an announced production order for a lead hyperscaler’s next-generation Sonoma ramp, a new silicon photonics customer with Q4 shipment timing, and incremental Sonoma manufacturing capacity. That is a much stronger setup than a generic “pipeline” story. If these orders convert on schedule, the company can move from subscale FY26 profitability to meaningfully improved FY27 volume absorption, broader market diversification, and better consumables leverage. 
The bearish interpretation is that the company still has not fully solved its oldest public-market problem: translating exciting technology and large customer opportunities into smooth, predictable quarterly financials. Q4 still requires a sharp revenue increase. The highest-value wafer-level programs remain engineering-intensive and timing-sensitive. Backlog carries limited cancellation protection. Customer concentration remains high. The margin thesis still depends on a future consumables mix improvement that is not yet obvious in reported gross margin. In that sense, the call materially de-risked demand, but did not fully de-risk execution. The immediate market reaction reflected that balance: AEHR traded at $50.25 at 6:27 p.m. ET on April 7, down 3.6% from the prior close. 
The most important confirmatory signals from here are straightforward. 1st, Q4 revenue needs to convert into roughly a mid-to-high-teens level for management’s qualitative tightening to be validated. 2nd, the promised near-term follow-on Sonoma order from the lead hyperscaler would further strengthen the package-level ramp case. 3rd, the top-tier AI wafer-level benchmark needs to exit redesign and move toward production order visibility. 4th, silicon photonics needs to prove it is repeatable across multiple customers, not just a single quarter spike. 5th, consumables need to become a visibly larger share of revenue if the longer-term margin algorithm is to command a premium multiple. Absent those signals, this quarter risks being remembered as a peak-bookings, delayed-revenue quarter. If they appear, the call will likely prove to have been the point where Aehr’s story shifted from narrative optionality to a tangible FY27 growth setup. 
COMPANY PARTICIPANTS
•Gayn Erickson, President and Chief Executive Officer
•Chris Siu, Chief Financial Officer
•Jim Byers, Investor Relations, PondelWilkinson
•Unidentified Speaker, title not disclosed in the transcript
RESEARCH ANALYSTS
•Christian Schwab, Craig-Hallum
•Larry Chlebina, Chlebina Capital
•Mark Schuder, William Blair
•Max Michaelis, Lake Street Capital Markets
