Much is happening under the hood at Canatu, and today we once again received more clues to what is going on.
Four releases in a little over 24 hours. Yesterday a second reactor order from FST, paired with a preliminary outlook cut and the H1 revenue number. This morning a new strategy with updated long term targets, change negotiations for up to 17 people, and the half year report itself. That is not a coincidence. The new CEO wanted the bad number, the plan and the repeat order to land as one package, so nobody gets to read the revenue line without the FST order sitting next to it.
I will not dwell on the numbers. Revenue fell, the loss widened, opex is up because the headcount added under the previous CEO is now fully in the cost base, and cash is still ample. That was all expected. What matters is what was said. Here is the recap.
Not financial advice. This post reflects my personal views and analysis only and is provided for informational purposes. I may hold positions in the securities discussed and may change them without notice. Do your own research before making any investment decisions.
The first 100 days
Maximilian Slawinski used his first shareholder letter to say the obvious quietly: he has stopped everything that was not on a path to scalable, profitable growth, and he is accepting the top line hit from doing so. The engineering services business that actually grew in H1 is being discontinued. The executive team has been rebuilt around five people, with a new CTO and CMO brought in from outside. Business development has been folded into the business units rather than living as a separate exploration function. The whole thing has a name, Polaris, which is what you call a programme when you want the board and the organisation to treat it as a break with the past.
The layoffs fit the same logic. They cover Operations, Robotics, Mobility and Defence, and Medical Diagnostics. They do not touch Semiconductor. Priorities do not get much clearer than that.
The business model is now settled
For years there was ambiguity about whether Canatu would make pellicle membranes itself or sell the equipment to others. That is over. Canatu is an equipment and IP vendor for EUV pellicles. It sells reactors, consumables and licences, and collects royalties when the customer produces. Inspection membranes stay as a product business made in Finland. The CEO letter says it in plain words: “technology leadership in the CNT pellicle market as an equipment and IP vendor.”
This is the capital light choice and the right one given the balance sheet. It also means Canatu’s revenue depends on its customers getting their pellicles qualified at the chipmakers, and Canatu does not control that. The risk section still says the productivity claim for CNT pellicles has not been verified because no reactor is in production use yet. Read that sentence twice. FST has had a production licence since October 2025 and there are still no royalties.
FST doubles down anyway
That makes the FST order the most important item in the package. FST is ordering a second reactor before the first one has generated recurring revenue for Canatu. That is a customer paying for capacity ahead of qualification, which is what you do when you believe qualification is coming. The March long lead item order that some of us read as a possible new customer turned out to be the front end of this FST order. Delivery is 2027, so this is next year’s revenue, not this year’s.
The second reactor, shipped in December 2024 to a still unnamed customer, is still in customer acceptance testing. Twenty months. The report describes it as ongoing and nothing more.
The meat of the report
If you only read one sentence in the half year report, read this one, from the Semiconductor business overview on page 6:
“The first CNT100 SEMI reactor, delivered to FST (Fine Semitech), operates under a commercial production license. The remaining steps toward mass production are the end customers’ product approvals in pilot and risk production.”
Most readers skip that page. It sits between the key figures table and the financial overview and reads like filler. It is not. Three things in the sentence do a lot of work.
First, “end customers’.” Plural, apostrophe after the s. FST has more than one chipmaker qualifying its CNT pellicle. Canatu will never name them, but the set of chipmakers who could be qualifying a pellicle from a Korean supplier is not large.
Second, “pilot and risk production.” That is fab vocabulary, not supplier vocabulary. A pellicle does not have a pilot line or a risk production phase. A fab ramping a new node does. The sentence is telling you FST’s pellicle is inside a chipmaker’s ramp sequence right now, not sitting in a qualification lab waiting for a slot.
Third, “the remaining steps.” Not next steps, not future steps. Canatu is saying there is nothing left on its side or on FST’s side. What remains is the chipmaker’s approval clock, and that clock is not Canatu’s to read.
Now put that sentence next to what the Korean side has been saying, because the Finnish coverage has not done it.
Samsung invested in FST in 2021 as part of a programme to localise EUV pellicle supply, after investing in S&Stech the year before. In September 2025 Korean trade press reported that FST was in price negotiations with Samsung on EUV pellicles it would supply, that the prototype had met Samsung’s requirements on transmittance, uniformity, thermal resistance and hydrogen plasma durability with some particle issues left to solve, and that Canatu had supplied the membrane synthesis equipment. Price talks happen at the end of a qualification, not the beginning.
Then in January 2026 the piece that matters. Samsung placed an order with FST worth about 25 billion won for EUV pellicle mounting, demounting and inspection equipment for its Taylor fab in Texas. The reporting was explicit that this effectively confirmed Samsung would use EUV pellicles at Taylor for the first time. The equipment handles both CNT pellicles and conventional metal silicide ones, so it is not proof of CNT alone. But the sequence after it is.
January: Samsung orders pellicle handling tools from FST for Taylor. March: FST places a long lead item order with Canatu for a second reactor. August: FST converts it into a full reactor order worth over five million euros, before the first reactor has paid a single euro of royalty. FST is a small KOSDAQ company. It does not spend that kind of money on capacity for a product with no recurring revenue yet unless its customer and shareholder has told it to. The second reactor is Samsung’s pellicle localisation programme buying capacity through its supplier. Canatu’s press release describes it, accurately and quietly, as “expanding the customer’s manufacturing capacity of CNT pellicles.”
Two bullets above the FST sentence, in the market commentary on the same page, Canatu notes that Samsung reported mass production of its first generation 2nm technology. Taylor is the fab ramping that node, in exactly the pilot and risk production phases the FST bullet describes. Canatu put the two facts on the same page and left the join to the reader.
This reframes three things.
The end customer for CNT pellicles at scale is now identifiable. It is Samsung, and the first insertion point is Taylor, which is also where Samsung’s large US foundry contracts from last year are slated to run. When Canatu writes “end customers” in the plural, the natural reading is Samsung foundry plus Samsung memory, with SK Hynix as the other candidate through the same supplier.
The memory milestone in the new strategy is not a new market discovery. Samsung is the largest DRAM maker in the world and already runs EUV on its leading DRAM nodes. Once Samsung foundry qualifies FST’s pellicle, Samsung memory is the same supplier, the same pellicle, and the same handling equipment. Memory was promoted to a strategic milestone this week because the customer that is furthest along makes memory.
And for those of us who have been building a case around the second, unnamed reactor and the TSMC ecosystem, the FST chain is the mirror image. TSMC has been introducing Mitsui Chemicals’ metal silicide pellicles since 2019. Samsung chose to localise through FST and Canatu instead. Whoever the reactor 2 customer is, the FST precedent is now the case study they will be watching, and the report gives us nothing new on them beyond “ongoing.”
Why has nobody read the sentence this way? Because Canatu’s disclosures are written to be minimal and the company will not say Samsung. Because the Korean trade press is not in the feed of a Nordic small cap analyst. Because the March long lead order landed the same day as the withdrawal of the old targets and was read as a footnote to a bad news day, not as the first visible consequence of the January Samsung order. And because most people go straight from the key figures table to the outlook and treat the page in between as filler.
Where the pellicles go, and what a reactor is for
Everything in this section is speculation built on public arithmetic. Canatu has never disclosed reactor throughput, royalty rates or consumable pricing, and Samsung has never said the word Canatu. I am laying out the assumptions so you can swap in your own.
What Taylor consumes
Pellicle demand is wafer starts times EUV layers that carry a pellicle, divided by how many wafers a pellicle survives, plus an allowance for new masks and swaps.
Taylor phase one is a 2nm GAA fab. A plausible ramp is 10 thousand wafers a month in 2027, 25 thousand in 2028, 30 to 40 thousand at steady state. A 2nm flow runs roughly 25 EUV exposures per wafer. Samsung has said it will use pellicles in certain core EUV processes first, so assume 40% of layers pelliclised initially, rising toward full coverage. Canatu’s own 2030 assumption is 15,000 wafers per pellicle; a new fab on a new material will replace more conservatively, so 10,000 is the prudent case.
Run those numbers. At 25 thousand wafers a month, 10 layers, 10,000 wafer lifetime, Taylor needs about 300 pellicles a year, call it 400 with mask turnover. At full 25 layer coverage it is 750, call it 900 to 1,000. At 40 thousand wafers a month and Canatu’s 15,000 lifetime it is again around 1,000. So a mature Taylor consumes somewhere between 400 and 1,000 CNT pellicles a year. That is the whole fab.
What one reactor makes
Three ways to estimate it, and they agree.
From Canatu’s targets: 10 to 20 reactors by 2030 supporting half of global EUV wafers at 15,000 wafers per pellicle works out to roughly 10,000 CNT pellicles a year across the fleet, or 500 to 1,000 per reactor. From the revenue target: 100 million, mostly recurring, from 10 to 20 reactors is 4 to 5 million per producing reactor per year; at 2,000 to 5,000 euros of royalty and consumables per pellicle that is 1,000 to 2,500 pellicles per reactor. From the physics: floating catalyst CVD deposits a membrane in minutes to tens of minutes and the reactor runs continuously, so raw output could exceed 10,000 membranes a year; what limits finished output is yield through transfer, framing, coating at FST and inspection, and on a brand new process that is not high.
Call it 1,000 to 2,500 finished pellicles per reactor per year at mature yield, with 1,000 as the conservative planning number. A reactor is demand limited, not capacity limited.
So reactor two is not for Taylor
One reactor covers Taylor several times over. FST ordering a second one, before the first has shipped a commercial pellicle and before a euro of royalty has been paid, only makes sense if the end customer intends to pull CNT pellicles into Korea.
Samsung’s Korean footprint is a different scale. Foundry at 3nm and 2nm in Pyeongtaek and Hwaseong is perhaps 20 to 40 thousand wafers a month at 25 layers, which is 600 to 1,200 pellicles a year at Canatu’s lifetime assumption. DRAM is the bigger number. Samsung runs EUV on its leading DRAM nodes across something like 200 to 300 thousand wafers a month with about five EUV layers each, which is 12 to 18 million passes a year and 800 to 1,800 pellicles depending on lifetime. Whole of Samsung at full adoption is 2,000 to 4,000 CNT pellicles a year. Two reactors is exactly the right count for that. It is also why “memory” was promoted to a strategic milestone this week: the customer that is furthest along makes more DRAM wafers than anyone.
What it means in euros
The FSB 2025 confirmed that royalties are charged per CNT pellicle sold by FST to its end customers, so the model is per unit. Nobody outside Canatu and FST knows the rate. Bounding it from the top, Canatu’s 2030 targets against its own market size imply a take of around 2,000 euros per pellicle all in. Bounding it from the bottom, a 10 to 20% royalty on a pellicle priced in the low tens of thousands plus consumables is 3,000 to 6,000. Use 2,000 to 5,000.
Taylor alone at 400 to 1,000 pellicles is then 1 to 5 million a year to Canatu at maturity, most likely 2 to 3. Samsung wide at 2,000 to 4,000 pellicles is 4 to 20 million. That is a real business, and it is the first recurring revenue Canatu will ever have from pellicles. It is not 100 million.
The reactor count is the whole story
Samsung at full adoption is two to four reactors. To get to 15 producing reactors by 2030, which is what the targets need, Canatu has to be inside the TSMC ecosystem, where leading edge EUV wafer volume is three to four times Samsung’s. That is the unnamed second reactor, shipped in December 2024 to what the FSB called “a global semiconductor leader” rather than the “Korean semiconductor company” it uses for FST. Different phrasing, and I do not think it is accidental. The report gives us nothing new on that reactor beyond “ongoing,” twenty months after shipment.
The targets
Over 100 million euros of revenue in 2030, at least 20% annual growth from there to 2035, revenue per employee above 400 thousand, and capex under 6 million a year from 2027. The March range of 100 to 150 million is gone at the top, and the 25 to 30% EBIT margin target is gone entirely. The company says it prioritises growth in this phase and profitability will follow scale.
Two things to say about this. One, dropping the upper bound is either the new CEO knowing the number is closer to 100 than 150, or the visibility on the upside being worse than the previous management let on. Two, the margin target is not really gone, it is hidden. At 100 million and 400 thousand per employee you are capped at around 250 people, and if you run that cost base against a normal gross margin you land in the 20 to 25% EBIT range anyway. They just do not want to promise it.
The company has also stopped publishing annual operative targets. The stated reason is negotiating position with customers and suppliers. The unstated reason is that the previous set of operative targets was being missed in public and had become a stick to beat the stock with. You can dislike the loss of transparency and still understand the choice.
Memory shows up
The strategy release lists “expansion of CNT based EUV pellicles to memory chip production” as a named milestone, separate from qualification for device manufacturing generally. Memory was in the March material as a small slice of the pellicle market with upside. Now it is a priority. The reason is not mysterious: FST is Korean and its end customers are Korean, and the report notes that the remaining steps toward mass production are the end customers’ approvals in pilot and risk production, plural. The path that actually has a reactor, a licence and a repeat order runs through Korea and memory. The strategy now says so.
Robotics, Mobility and Defence is a different business than it was in March
The Automotive unit was renamed in June and the report makes clear why. The market commentary is about robotaxis, Chinese L3 licensing, Waymo volumes and defence budgets. The CEO letter talks about “safe 24/7 autonomous driving.” And the outlook contains a new risk factor that was not there in March: the approval of operating autonomous vehicles depends on regulations implemented by local governments.
That sentence tells you the ADAS camera heater lead customer’s volumes are tied to autonomous fleets, not to mass market passenger cars. Full windshield heaters and solar cells, which were priority applications in March, no longer appear in the strategic milestones. The new milestones are ADAS heater ramp, defence and robotics expansion, and a 3D heater product. The unit’s own leader is now the Semiconductor SVP on an interim basis, and it is one of the three areas covered by the change negotiations. This is a unit being narrowed to the things that can ship, with the moonshots parked.
Medical Diagnostics keeps its milestones
Proof of concept for total testosterone was reached, an alpha prototype system is targeted for end 2026, ISO 13485 was obtained, and the strategy names the start of an FDA pathway and a joint development agreement with a leading diagnostics company as milestones. It also got layoffs. Milestone driven development with a tighter budget is the phrase.
The membrane customer
The Semiconductor revenue collapse in H1 was not only about reactors. The report says a major inspection membrane customer was working down high inventory levels during the half. Inspection membranes have been Canatu’s most reliable revenue line, and this customer is very likely the single largest account in the company. The sentence is written in the past tense, which is meant to tell you the destocking is done and H2 normalises. Whether it does is one of the two or three things that will decide the full year number.
What was not said
No problems with the CNT technology itself. No slippage in the FST relationship, the opposite. No mention of a third reactor customer beyond “opportunities with existing and potential new customers have progressed.” No change to the balance sheet story. And nothing at all, in either direction, about who the second reactor customer is.
Next up: my thoughts, and the key information said in the webcast.