There is a number that the advanced packaging industry does not talk about enough: 15 to 30 percent. That is the additional test cost chiplet architectures add over a monolithic system on chip, driven almost entirely by what happens when a single die fails after it has already been bonded into a $1,000 or higher multi chip package. Chiplet architectures add 15 to 30 percent to total test cost versus monolithic SOCs due to Known Good Die testing and interposer yield losses. The market has been laser focused on who makes the substrate, who does the stacking, who pours the underfill. Almost nobody is watching the company that decides whether the die going into that package is going to survive production.
Aehr Test Systems is a $2.5 billion company in Fremont, California. It makes machines that stress test semiconductors at the wafer level, before the die is ever cut, packaged, or shipped. The idea is simple: apply heat and voltage to chips while they are still part of the full wafer, force the early life failures to appear in your factory rather than in a data center server rack six months after deployment. The execution is genuinely hard, and Aehr has spent three decades getting it right.
Fiscal 2026 revenue was $50 million, and yes, that is down 15 percent from the prior year. The electric vehicle power semiconductor cycle that carried Aehr through 2023 and 2024 went soft, and the company spent most of the year rebuilding its customer base around AI. That rebuild is now complete. The guidance for fiscal 2027 is $130 million to $150 million. That is 160 to 200 percent year over year growth, and the backlog to support it already exists. The thesis here is not what was, it is what is lined up.
The curve
Semiconductor advanced packaging is in early deployment. The technology works, leading edge customers are buying it, and the supply chain is expanding hard to meet demand. Sigmaintell forecasts the global semiconductor advanced packaging market will reach $58.7 billion in 2026, a 97 percent year on year increase, with supply shortages persisting until 2027. The number is large enough to be alarming, and the supply shortage tells you exactly where we are on the curve. Demand is moving faster than infrastructure.
Semiconductor packaging has moved from a back end manufacturing detail to a front line performance driver for AI and high performance computing. As transistor scaling gets harder and more expensive, more of the system level gain now comes from how logic, memory, and photonics are connected. That is why advanced packaging methods such as 2.5D and 3D stacking, chiplets, CoWoS, hybrid bonding, panel level packaging, and co packaged optics are drawing so much attention across the semiconductor value chain. Jensen Huang said it plainly at GTC 2026. He stated that the performance limit of the Rubin architecture ultimately depends on whether packaging substrate bottlenecks can be solved, arguing that traditional organic substrates used for 20 years can no longer support the next generation chip carrying 336 billion transistors.
The single binding constraint for this sector is yield. Specifically, the yield problem that emerges when you bond multiple high value dies together into one package and then discover a failure. Wafer level testing is becoming essential for multi die packages to avoid the high cost of discarding expensive substrates when a single die fails after packaging. One bad die in a CoWoS package does not just cost you the die. It costs you the interposer, the substrate, the memory stack, and the assembly labor. CoWoS S packaging costs approximately $750 per chip for H100-class designs, while CoWoS L costs $1,000 to $1,100 for NVIDIA’s B200, a 47 percent premium driven by multi die complexity. You cannot afford to find out a die was defective after the package is assembled.
This is what decides when the constraint inflects. Every generation of AI chip that adds more dies to the package raises the cost of a late stage failure. Management at Aehr estimates that only about 5 percent of ASICs and 50 percent of AI accelerators currently undergo production burn in, representing a significant expansion opportunity. The technology that catches early life failures at the wafer level, before packaging, becomes more valuable with every new chiplet generation. That expansion from 5 percent of ASICs to something approaching standard practice is the inflection this whole thesis rests on.
GPU and AI accelerator platforms are pushing unprecedented demand for advanced packaging technologies like TSMC CoWoS, Intel Foveros, and Samsung I Cube. Packaging capacity has become a bottleneck in the AI supply chain, influencing global chip availability and pricing. Co packaged optics adoption is accelerating as hyperscalers push for major power savings in AI networking, and with TSMC integrating COUPE into CoWoS, 2026 is becoming the inflection point for optical engines in high performance systems. Silicon photonics chips bonded directly to AI processors need the same burn in logic as logic dies. The market is not one wave. It is several waves arriving close together.
The market has repriced the substrate makers and the OSATs. It has not repriced the test layer. That lag is the opportunity.
The company
Aehr Test Systems has been making burn in equipment since 1977. For most of that history it was a sleepy supplier to the memory and logic semiconductor industry. The modern version of the story started when silicon carbide power semiconductors for electric vehicles became a real market, and Aehr’s FOX platform became the standard tool for high volume wafer level burn in of SiC MOSFETs. That business peaked, softened, and Aehr spent fiscal 2026 diversifying out of it at exactly the right moment.
The core product is the FOX platform, a family of systems that can test and burn in semiconductor wafers at high temperature, high voltage, and in the FOX XP configuration, at extremely high power. The FOX XP offers up to 3,500 watts per wafer in its high power configuration. That matters because the next generation of AI accelerators and silicon photonics transceivers runs hot, and legacy test equipment cannot handle the power density. Increasing power densities in next generation GPUs are effectively breaking legacy test tools, forcing customers to adopt Aehr’s high power Sonoma and FOX platforms.
The silicon photonics opportunity is the clearest near term growth driver. A silicon photonics customer is developing advanced devices used in next generation optical interconnects and optical I/O architectures for hyperscale AI and cloud data centers. These devices enable dramatically higher bandwidth and lower power communication between AI processors, memory, switches, and networking equipment, addressing one of the industry’s most significant bottlenecks as AI infrastructure continues to scale. Aehr announced its first major silicon photonics customer in March 2026, received a follow on order in June, and then received another follow on order on July 9. Aehr is the market leader in wafer level burn in for silicon photonics transceivers, with a large installed base at leading global semiconductor and photonics companies.
The AI accelerator processor side is equally large. The company secured production orders for its Sonoma package level burn in systems from a major hyperscale customer for their next generation, significantly higher power AI processors. This customer is forecasting a substantial expansion of Sonoma purchases starting in the second half of calendar 2026 and continuing into 2027. That expansion is not yet in the $130 to $150 million FY2027 guidance, which means the official guidance is actually conservative against the pipeline management has described.
The Fremont manufacturing base has been upgraded specifically to support the ramp. The Fremont facility has been upgraded to support significantly higher throughput, potentially handling $100 million in wafer level systems per month if fully staffed. Aehr is also running a hybrid capacity model where its Sonoma systems are outsourced to a contract manufacturer in Asia, freeing Fremont floor space for the more complex FOX wafer level builds that carry higher margins.
What Aehr has that no one else has is the accumulated IP in full wafer contacting. A WaferPak contactor has to make tens of thousands of simultaneous electrical connections to a wafer while controlling temperature precisely across the full die surface. The FOX XP platform enables high parallelism, high temperature, and high power burn in, with the unique ability to precisely maintain devices at the required temperature to enable highly effective burn in, allowing customers to identify early life failures before packaging while significantly reducing cost of test. Replicating that is not a six month project for a competitor. It is decades of materials science and process know how.
The numbers
Full fiscal year 2026 ended May 29, 2026. Revenue was $50 million, down 15 percent year over year, with full year non GAAP gross margin at 38.5 percent compared with 44 percent in the prior year. Full year non GAAP net income was $0.9 million, or $0.03 per diluted share, compared with $4.6 million in fiscal 2025. Those are not inspiring annual numbers. But the quarter level data tells a completely different story. Q4 revenue of $18.8 million beat expectations and rose 34 percent from a year earlier. Bookings hit $60.7 million, more than five times the level of the year earlier quarter. This is what an inflection looks like in the data before the revenue arrives.
Aehr guided fiscal 2027 revenue of $130 million to $150 million, representing roughly 2.6 to 3 times growth over fiscal 2026, supported by a record $100.6 million effective backlog. Guidance assumes non GAAP pretax profitability between 18 and 22 percent. That margin target, if achieved, would represent a structural improvement from a business that was effectively breakeven at $59 million in revenue a year ago. The operating leverage is real: the Fremont facility can handle much higher volumes without proportional cost increases.
The balance sheet is solid post raise. At the end of Q4, cash equivalents and restricted cash totaled $116.5 million. During fiscal 2026, the company raised approximately $100 million primarily through its ATM program, significantly strengthening the balance sheet. There is essentially no debt. Operating cash outflow for the full year was $3.3 million. At the current burn rate, Aehr has more than ten years of runway on its cash position, which means there is no near term equity raise risk tied to survival. There is, however, a history of using the ATM opportunistically when the stock is elevated, which I will address in the risk section.
On the technical setup: the stock hit an all time high of $126.62 on June 15, 2026. It had risen about 320 percent in 2026 through mid July, sitting at a market capitalization of roughly $2.7 billion. After the July 14 earnings print, shares initially surged 29 percent before pulling back. Profit taking after the strong earnings rally drove a 13 percent single session drop on July 24. As of July 27, the stock trades around $74 to $76, which is approximately 40 percent below the June ATH. The 50-day moving average is approximately $105 and the 200-day moving average is approximately $100, placing current price below both. Volume expanded sharply on the earnings pop and has since contracted on the down days, which is the better pattern of the two. RSI is near neutral at 51. This is a stock digesting a run, not one in a structural downtrend. The 200-day is far below from a year ago when the stock was single digits, and the trend since the October 2025 breakout remains intact at the higher timeframe.
Short interest sits at approximately 13 percent of the float, which is elevated. The latest short interest is 4.09 million shares, roughly 13 percent of outstanding shares sold short. Only four Wall Street analysts cover the name. That combination of high short interest and thin coverage is exactly the setup where guidance beats and order announcements move stocks by 20 to 30 percent in a session, as July 14 demonstrated.
Why it wins
The moat here has two layers. The first is technical. The WaferPak contactor technology requires a custom designed interface between the test system and each specific wafer design. Every new device generation, every new customer tape out, requires a new WaferPak. Aehr designs and manufactures those contactors itself. The project backlog in cleaning increased nearly 50 percent in the first six months compared to the prior year’s first six months, suggesting customers are pulling WaferPaks faster than the annual revenue numbers yet reflect. This is a recurring consumable business embedded inside what looks like a capital equipment company.
The second layer is the installed base. A lead hyperscale customer is evaluating wafer level burn in for their third generation device to avoid the massive yield loss associated with failing a single die in a multi chip package. Once a hyperscaler integrates Aehr’s systems into its production line for generation three, switching costs for generation four are enormous. The DFT, the WaferPak design, the process qualification work, the technician training: all of it has to start from zero with a new vendor. That stickiness is what makes the hyperscaler pipeline so valuable beyond just the headline order sizes.
The silicon photonics angle is a genuine second leg, not a story substitute. Silicon photonics is essential for speeding up data transmission within AI clusters. Because these optical chips require rigorous testing at the wafer level to avoid expensive packaging failures, Aehr’s testing platforms have become a critical tool for global hyperscalers, driving recurring equipment orders. Three separate order announcements in a 90-day window from March to July 2026 from silicon photonics customers is not coincidence. This market is pulling hard.
The constraint that matters for this sector, yield loss at packaging, only gets worse as die count per package rises. HBM demand continues to exceed supply as HBM4 and 16-Hi stacks roll out, raising yield and thermal risks. Aehr has separately indicated that HBM4E standards with embedded Built In Self Test capabilities may allow it to enter the DRAM market sooner than originally anticipated. That would be an entirely new addressable market on top of the AI processor and silicon photonics businesses that are already ramping.
What could go wrong
The ATM dilution history is real and worth naming plainly. During fiscal 2026, the company raised approximately $100 million primarily through its ATM program. Aehr has now established a pattern: raise equity when the stock is elevated, build cash, use it to fund operations and capacity. That is not inherently bad management, but it means every strong rally in the stock carries some probability of secondary issuance. In April 2026, Aehr launched a $60 million ATM share facility with William Blair. That facility is still available. If the stock recovers toward the June highs, expect dilution to be a check on the upside.
The customer concentration risk is high. The revenue ramp in FY2027 is overwhelmingly dependent on one lead hyperscale AI customer expanding its Sonoma package level burn in program, plus a handful of silicon photonics customers still in early production ramp. None of those customers are named publicly. If the hyperscaler delays its next generation processor launch, or decides to pull burn in testing in house, the guidance collapses. Management has said those potential orders are “not yet included in guidance,” which is reassuring, but the base case itself is already concentrated.
The EV power semiconductor business is not gone, it is just depressed. If that market recovers while AI demand softens, Aehr has a natural offset. But right now the stock is priced for AI execution. AEHR’s price to sales ratio hit 58.1 times at peak post earnings levels, and operating margins remain at negative 28.3 percent. At $76, the forward multiple on $140 million in FY2027 revenue is roughly 18 times. That is aggressive for a company that was loss making last year. Any guidance cut in October would likely wipe out 30 percent or more of the market cap quickly.
The technical picture is honest: the stock is below both its 50-day and 200-day moving averages right now, having broken down from the June 15 ATH on profit taking. That is a falling knife pattern by the book, and I am noting it as required. What changes the picture is whether volume expansion returns on up days as the October catalyst approaches. It has not yet. Anyone sizing into this name before October 5 is buying ahead of confirmation.
What I am watching
The primary catalyst is the Q1 FY2027 earnings report on October 5, 2026. Projected fiscal 2027 revenue of $130 million to $150 million is supported by a record $100.6 million effective backlog, but the first quarter print will tell us whether the revenue recognition is actually accelerating as promised. Shipments in Q4 already ran 54 percent above year ago levels at $241 million, meaning the deferred revenue pipeline is real. The question is timing. If Q1 revenue comes in at $25 million or above, that represents the acceleration that justifies the forward valuation.
The second thing I am watching is silicon photonics order flow between now and October. The three orders received in March, June, and July 2026 have each been incrementally larger. The lead silicon photonics customer provided Aehr with a forecast for additional systems, with follow on orders possible as early as later this calendar year as it ramps capacity to support next generation hyperscale data center deployments. A fourth order announcement before the October earnings print would be meaningful confirmation that this customer is moving toward volume production. Watch for press releases from Aehr on the wire, which tend to move the stock 8 to 15 percent intraday.
The third item is any announcement related to the potential new DRAM or HBM market entry. New HBM standards with embedded Built In Self Test capabilities may allow Aehr to enter the DRAM market sooner than originally anticipated. A design in win with a DRAM manufacturer would be a genuinely new chapter for the company, one that the current guidance and backlog numbers do not include at all. I would treat any press release naming a memory customer as a significant upside surprise signal. That is the catalyst that could reset the ceiling meaningfully higher than the current $140 million midpoint guidance implies.
The bottom line
Advanced semiconductor packaging is in early deployment, and the binding constraint is yield at the multi die module level. As chiplet counts rise, as silicon photonics bonds directly to AI processors, and as every packaging architecture becomes more expensive per unit, the cost of discovering a bad die after bonding climbs asymptotically. Wafer level burn in before packaging is the logical response, and right now only about 5 percent of ASICs undergo it in production. That 5 percent number will not stay at 5 percent. The question is who captures the volume as it expands.
Aehr Test Systems has the installed base, the WaferPak IP, the hyperscaler relationships, and the $116 million cash position to execute. The FY2027 guidance of $130 to $150 million is the first quarter by quarter test of whether the booking wave translates into revenue. The stock is pulling back from its ATH, sitting below both the 50 and 200-day moving averages, with 13 percent short interest and only four analysts covering it. That setup does not guarantee a smooth ride, and I said so above. But October 5 is the date where the story either gets confirmed or starts to unravel. That is a specific enough timeline to manage against.
Not financial advice.