Every AI Chip Has to Survive the Oven First - $TRT
$TRT is a 67-year-old microcap that runs the last test the most expensive AI and automotive chips have to pass, its revenue is accelerating, the market that already found this
Long $TRT. Everything here is built from public filings and public data. Nothing here is investment advice. For educational purposes only. I may hold positions in names I discuss. Do your own research.
Written 17 June 2026. As I write this, Trio-Tech trades near $14.82, up roughly 18 percent on the day, against a 52-week range of $2.44 to $21.38. That range is the first thing to understand. This is a roughly $150 million company on about 10.1 million shares with no sell-side coverage, and it moves violently on small flows in both directions. Read the position-sizing and the bear section before you act on anything here. The stock has already run hard. The point of this letter is not the candle today. It is the gap between what Trio-Tech earns and what the market that already loves this exact business is willing to pay for it.
Let's go.
The layer below the layer everyone already bought
The AI trade has already paid up for everything one rung beneath the models. The compute. The memory. The networking. The foundries. That framing is consensus now, and consensus is priced.
There is a rung below that one, and it is still not priced, because it is unglamorous and almost nobody covers it.
When a chip is built, it does not ship straight to a customer. The expensive ones, the AI accelerators, the automotive controllers, the memory that has to live for years inside a data center, get stress-screened first. The most important screen is burn-in. The chip is mounted on a specialized circuit board, the burn-in board, then run hot, at elevated temperature and voltage, under electrical load, for hours or days. The purpose is to force the weak units to fail in the lab instead of in the field. A part that survives burn-in is far less likely to die early inside a customer's product.
For a cheap consumer part you can argue about whether that screen pays for itself. For an AI GPU that sells for tens of thousands of dollars, an automotive chip where a field failure is a safety event, or memory that runs continuously for years in a data center, the screen is not optional. The economics of the part demand it.
Trio-Tech International sells the burn-in boards. It sells the reliability-test equipment around them. And it runs the testing itself as a service, including final test for advanced chips, from its own floors across Asia. It does not design chips and it does not compete with the chip makers. It sells to them and to the houses that assemble and test for them. When AI and automotive chip volume rises, demand for burn-in and reliability capacity rises with it, and that demand lands on a very small number of specialists.
The whole thesis is one sentence. Every expensive AI and automotive chip has to survive the oven first, and Trio-Tech sells the oven, the boards that go in it, and the oven time, for a fraction of the multiple the market already pays the one burn-in name it has discovered.
Why the screen is becoming mandatory right now
The reason this is a today story is that the AI parts are breaking the old way of testing.
The cleanest read on the category does not come from Trio-Tech. It comes from its larger, pure-play public cousin, Aehr Test Systems, which has spent the past year telling the market two things. First, that rising power densities in next-generation GPUs and accelerators are overwhelming legacy test tools and forcing customers onto new high-power burn-in platforms. Second, the number that frames the entire opportunity, that only about 5 percent of ASICs and roughly 50 percent of AI accelerators currently go through production burn-in at all.
Sit with that. Half of AI accelerators and the overwhelming majority of custom AI silicon ship today without the production burn-in screen. As these parts get more expensive, hotter, and more mission-critical, and as the hyperscalers and the automakers demand field reliability they can underwrite, that attach rate only goes one way. The burn-in attach rate on the most valuable chips on earth is low and climbing. The screen is also spreading into new device classes, silicon carbide power devices and silicon photonics among them. That is the tide. Aehr is the high-multiple way the market already plays it. Trio-Tech is the way the market has not noticed yet.
The inflection is in the filings, and it is accelerating
This is not a story about what might happen. It is reported.
In fiscal Q3 2026, the quarter ended 31 March 2026, revenue was $16.5 million, up 124 percent from $7.4 million a year earlier. The Semiconductor Back-End segment, the part that matters, did $13.1 million of that, up 141 percent. Industrial Electronics added $3.4 million, up 76 percent. Across the first nine months of fiscal 2026 revenue reached $47.7 million, up 85 percent, with the back-end segment at $36.9 million, up 104 percent.
The shape of that growth matters as much as the size. The year-over-year rate went 58 percent, then 82 percent, then 124 percent across the three quarters of fiscal 2026. This is a curve, not a single print.
And the nine-month figure already sits above any full year Trio-Tech has reported since fiscal 2022. This is the part the market is still missing.
Read what that says. Revenue sat on a roughly $42 to $44 million plateau from fiscal 2022 through 2024, then slid to $36.5 million in fiscal 2025. The market, reasonably, priced a tired, cyclical, slowly declining industrial. Then nine months of fiscal 2026 came in at $47.7 million, already past the old plateau, with the rate of growth still rising. The stock is being valued on the company that was declining. The opportunity is the distance to the company the filings now describe.
The segment doing the work is unambiguous.
The customers, by name of theme, and one the company already disclosed
The growth has three sources, and one of them is bigger than the founder letter let on.
The first is final test for a leading AI chip maker. In its fiscal Q1 2026 release, Trio-Tech disclosed its entry into providing final testing services for next-generation high-performance AI devices for what it called a leading AI chip manufacturer, and management guided explicitly to additional revenue from that AI customer for final test services. This is not an inference. The company put it on the record. That is a recurring services relationship with a top-tier AI chip vendor for production test of next-generation parts. It is the highest-quality leg of the story, and it is exactly the kind of relationship that compounds as the customer's volume ramps.
The second is the AI GPU burn-in board order book, which has kept filling after the quarter closed.
The June 4 order matters more than its size. It landed after the quarter ended and after the earnings print, which is the cleanest possible answer to the bear worry that this is one front-loaded burst. The demand is still arriving.
The third is automotive, and it is structured as a service, not a one-time sale. On March 4 Trio-Tech secured an initial roughly $2.5 million production burn-in order from a leading automotive integrated device manufacturer, described as an ongoing program that ramps in phases through calendar 2026, supported by targeted investment in capacity, workforce, and process controls to meet automotive reliability standards. Automotive burn-in is recurring by nature. Every production lot gets screened. That is the steadier counterweight to the lumpier board sales.
The hidden leg: Penang, and the memory customer the market has never tied to this name
Here is the part that turns a good setup into an asymmetric one, and it did not come from a press release. It came from the company's own footprint.
On April 28 Trio-Tech's Malaysian subsidiary signed a lease for about 104,000 square feet of industrial space in Perai, Penang, at a base rent near $115,000 a month, with the term running from June 1, 2026, into 2028 and a one-year extension option. That is a large block of new testing capacity in the heart of the Penang semiconductor cluster, funded by a roughly $10 million raise the company closed days earlier. You do not lease that much floor and fund it fresh unless you intend to fill it.
The more telling signal is in the hiring. Trio-Tech is staffing roughly thirty positions in Batu Kawan, Penang, a park where the company has not historically run a facility, and at least one advert states the role is based at MMP. In the Penang cluster, MMP is the routine internal shorthand for Micron Memory Penang, Micron's memory assembly-and-test operation anchored at Batu Kawan, where Micron has committed billions of dollars and built a center of excellence for memory and storage.
A burn-in and reliability specialist staffing engineers to sit physically inside a memory plant is the signature of a supplier being qualified into that plant's supply chain.
Be precise about confidence here. Micron has not named Trio-Tech. Trio-Tech has not named Micron. This is an inference drawn from public hiring data, the new Perai lease, and the timing of the raise, three signals pointing the same direction at the same time. Treat it as a strong, well-supported lead, not a fact. But the logic is clean, and the direction is memory, the third giant end-market of the AI cycle alongside compute and automotive, and the one Trio-Tech has never been associated with. Penang is also where Coherent has built a datacom and silicon-photonics research presence and where the broader AI back-end is concentrating. Trio-Tech is expanding straight into the densest reliability-test cluster on earth. If the memory read is right, a fourth leg is being added to a story the market is already underpricing. If it is wrong, the AI final-test, AI board, and automotive legs still stand on their own disclosed numbers.
The number that defines the trade: the comp pays 75 times sales
To value Trio-Tech you do not need a discounted cash flow you can argue with. You need the comp.
Aehr Test Systems is the public, pure-play, AI-burn-in name the market has already found. Here is what the market pays for it, next to what it pays for Trio-Tech.
This is the whole letter in one table. The market has decided burn-in levered to AI is worth paying for. It pays Aehr roughly 75 times sales for the privilege, while Aehr loses money. It pays Trio-Tech roughly two times sales, while Trio-Tech makes money and grows faster. The two companies sit on adjacent rungs of the same ladder. Aehr sells the capital systems. Trio-Tech sells the consumable boards that populate those systems, the surrounding reliability equipment, and the test service, including final test for a leading AI chip maker. On burn-in boards specifically, there are only a handful of specialist suppliers in the world. Trio-Tech is one of them.
I am not arguing Trio-Tech should trade at 75 times sales. I am arguing that the distance between 2 and 75, for a profitable, faster-growing company in the same niche with no coverage to close the gap, is the largest mispricing I have put in front of you in this book.
The one thing the bulls are not modeling: the margin
Now the honest center of the analysis, the part that separates a careful read from a press-release read.
The revenue doubled. The gross margin did not come with it. In fiscal 2025 Trio-Tech earned a roughly 25 percent gross margin. Across the first nine months of fiscal 2026 the gross margin ran closer to 16 percent, and the March quarter itself produced a small operating loss of $81 thousand and a net loss to common of $38 thousand, despite record revenue.
That is the swing factor, and it cuts both ways. The bear read is that the explosive revenue is coming through lower-margin burn-in board hardware and new-capacity startup cost, that the surge therefore does not drop to the bottom line, and that a microcap selling capital boards at 15 percent margins is not the high-quality compounder the headline growth implies. The bull read is that this is exactly what an inflection looks like before it converts. The lower-margin board orders are the door opener. The higher-margin work, the final-test services for the AI chip maker, the multi-phase automotive program, and the apparent Micron memory engagement, are recurring service revenue that scales into the new Penang floor and carries a structurally better margin. If the mix shifts back toward services as that capacity fills, the gross margin recovers toward and through the old 25 percent line on a much larger revenue base, and the operating on a near-fixed Asian cost base is violent. The entire profit case rests on that recovery. Watch the gross margin line on every print from here. It is the single most important number in the model, and it is the one the headline-reading bulls are ignoring.
The balance sheet does not have the usual microcap problem
A company inflecting this fast usually carries a stretched balance sheet. This one does not. At the March quarter Trio-Tech held $13.0 million of cash, and on April 24 it priced a registered direct offering of 1,052,632 shares at $9.50, raising about $10 million gross, which closed April 27. That puts pro-forma cash near $23 million against a company with almost no debt, a current ratio near 3.0, and total debt to equity around 0.11. It sits in a net-cash position.
The dilution was real, a bit over ten percent, and the stock sold off on the news, which is the normal reaction to a raise. The constructive read is that management raised from strength rather than distress, after the stock had already run, and now has the cash to build the Penang capacity that the AI, automotive, and apparent memory demand all require. There is also a Real Estate segment, a legacy of the company's long history in Malaysia and China that earns rental income and holds tangible property value. It is not the thesis. It is a small cushion under a stock the market still prices on its sleepy past.
Alignment, control, and the float
Trio-Tech has been led since 1990 by S.W. Yong, chairman since 2023 and inside the company since the 1970s. He owns about 16.9 percent of the shares, roughly 1.47 million, and in January he exercised options to buy 80,111 shares at $2.64 each, deep conviction at a price a fraction of today's. This is an owner-operator who has carried the company across decades and is buying, not selling.
Institutions hold only about 16 percent. That is the other half of the setup. A profitable, fast-growing semiconductor reliability play levered to AI, automotive, and possibly memory, with almost no institutional ownership and no sell-side coverage, is a stock where the marginal buyer has not arrived yet. The float is tiny, roughly ten million shares, which is why the stock can run from $2.44 to $21.38 and give back a third of it inside weeks. That illiquidity is the risk and the fuel in the same sentence.
What a re-rate is worth
I will not hand you a single false-precision number. I will hand you the framework and four scenarios, each built the same way: a fiscal 2027 revenue level, a gross-margin assumption, and a price-to-sales multiple that stays far below the 75 times the market already pays the comp. Run the math yourself and change the inputs. The shape is what matters.
Read the moonshot line carefully, because it is the point. An 11 times sales multiple, the one that implies a roughly $150 stock, is still only about one-seventh of what the market pays Aehr today. To reach the upside in this letter, Trio-Tech does not have to be loved more than the comp. It has to be loved a fraction as much, while being profitable and growing faster. That is what makes the asymmetry real rather than rhetorical. The base case alone, an $80 million revenue year at a 4 times multiple the market would consider cheap for any AI-levered name, roughly doubles the stock. The bull and moonshot cases are what happen when coverage initiates, the Micron memory leg gets disclosed, and the gross margin recovers on a much larger base at the same time.
What would make this wrong
You get the bear case in full, because the bull case is only worth anything if it survives it.
The first risk is liquidity. This is a genuinely tiny, illiquid stock that has already run roughly six times off its low and trades a third below its high. It moves hard in both directions on small flows. Position sizing matters more here than in anything else in this book. Do not chase a green candle in a name that can erase a week in an afternoon.
The second risk is the margin, the one the whole bull case rests on. If the revenue mix stays weighted to lower-margin board hardware and the new Penang capacity carries fixed cost faster than it carries revenue, the gross margin does not recover, the company stays at breakeven on a bigger base, and the re-rate never comes because there are no earnings to re-rate. This is the live debate, and the next two prints settle it.
The third risk is cyclicality. Trio-Tech has had revenue surges before that faded back to the plateau. The bull case requires that the AI cycle is structurally larger and more durable than a normal back-end upcycle. The disclosed final-test relationship with a leading AI chip maker, the multi-phase automotive program, and the Penang buildout are the evidence that it is. They are not yet proof.
The fourth risk is the inference. The Micron memory read is drawn from public hiring signals, not confirmed by either party. If it does not convert into disclosed business, the most exciting leg of the story is softer than it looks today. The thesis does not depend on it. The moonshot does.
The fifth is concentration and control. An owner-operator with a 16.9 percent stake gives you alignment and key-person and minority-shareholder risk in the same breath, in a name with thin governance scrutiny because no one covers it.
What to watch
The fiscal year ends June 30, so the quarter happening now is the close of fiscal 2026, and the full-year report around September is the first hard test of whether the order surge converted into sustained revenue and, more important, whether the gross margin started to recover.
The single highest-value catalyst is the one nobody can schedule: the first analyst who picks up coverage. In a name with none, the day the gap gets a voice is the day it starts to close.
Is $TRT a buy? The bottom line
$TRT is a 67-year-old, no-coverage microcap sitting in the one layer of the semiconductor supply chain that every expensive AI and automotive chip has to pass through. Its revenue is accelerating, it runs final test for a leading AI chip maker, its order book keeps filling with AI GPU and automotive work after the quarter closed, it is building new capacity in the heart of the Penang cluster, and the hiring trail points at a memory customer it has never been tied to. The balance sheet is clean, the insider owns one share in six and is buying, the institutions have not arrived, and the closest comp trades at 75 times sales while losing money. Trio-Tech trades at two times sales while making it.
The honest catch is the gross margin, and the next prints decide whether the inflection converts to profit. Size for the illiquidity. But understand the asymmetry: to reach the upside in this letter, this company does not have to be loved more than the burn-in name the market already found. It has to be loved a fraction as much.
The oven does not care how good the chip looks on paper. It only cares whether it survives. Right now the entire AI build-out is feeding more and more expensive silicon into that oven, and almost nobody has noticed who owns it.
The full valuation model and the catalyst-by-catalyst case go out in Part 2. Subscribe so the next one lands in your inbox before the gap closes.
Disclosure. Long $TRT. This is research synthesis for educational purposes, not investment advice. You should not buy or sell securities based on anything written here. I am not a registered investment advisor and I do not owe you a fiduciary duty. The price scenarios above are illustrative model outputs built on stated assumptions, not forecasts or guarantees, and the inputs can be wrong. The Micron memory connection is an inference drawn from public hiring and corporate records, not confirmed by the companies involved, and may be wrong. Figures are drawn from public filings and announcements and may contain errors. Position sizing in a microcap this illiquid is your responsibility. Do your own due diligence.
More in this series