AIXTRON's Inflection Landed Early — and Not Where It Was Promised
Orders doubled to €275m on indium phosphide lasers and, abruptly, GaN. Management still won't say what any of it is worth.
AIXA.DE · Earnings Call · 2026-10-29
The order book outran the script
For five straight quarters, AIXTRON have been the people at the party telling you the good stuff is coming later. In October 2025 the message was that the real volume pull in power semis would not arrive until 2027–28. In February 2026, asked whether the 800-volt data-centre GaN orders were in the backlog, Felix Grawert was blunt: “I don't think there is still 800-volt orders in.” — Felix Grawert, Chief Executive Officer · 2026-02-26 And on the same call he said the company needed roughly €300m of new orders in the first half to make its 2026 guidance, and was “fully on track” to get them.
Then Q3 happened, and a single quarter delivered almost that entire half-year requirement.
We have received strong new orders of EUR 275 million, more than double the prior year level, and 40% above Q2, predominantly driven by optoelectronics and increasingly also Power Electronics.
The whole of 2026 revenue is guided to €560m ±30, meaning one quarter of intake now equals roughly half a year of shipments.
That is the story. Not the top line — revenue of €184m landed inside the guided €180m ±20 band, and full-year guidance was simply reconfirmed. The story is that the order book has decoupled from the model, and it has decoupled because two things happened that management spent a year describing as “too early.”
The demand came from the wrong places, in the right way
Optoelectronics took 68% of equipment order intake, and tools for indium phosphide optical applications now make up the majority of backlog, giving revenue visibility well into 2027. The structural driver is the one everybody knows — AI data center build-out pulling lasers into high-speed optical interconnect — but the evidence got harder this quarter. Grawert cited LightCounting's May report raising its 2030 forecast for 800G-and-above transceivers plus LPO and CPO ports from roughly 198 million to 279 million units annually, about 40% up. This is the same co-packaged optics complex that peers flagged this week: Marvell-adjacent and Celestica both talked integrated optics at hyperscale volumes, and Monolithic Power referenced 800V data centre architectures directly.
What is genuinely new, and genuinely counterintuitive, is the wafer-size story. The received wisdom is that compound semi migrates to bigger wafers and each generation needs fewer tools. AIXTRON's own data says the opposite for 2027: “In fact, in 2027, the 6-inch goes down from around 30% down to 25%.” — Felix Grawert, Chief Executive Officer · 2026-10-29 Even 3-inch is doubling from 15% to 30% of systems. The reason is substrate scarcity — customers take every wafer they can get regardless of diameter, which means more millimeter fab capacity is built out of smaller, less productive wafers, which means more tools per unit of output. The migration they once treated as a headwind is now a multiplier.
GaN is the second surprise, and here the company's own script has been explicitly edited. The new demand is low- and medium-voltage devices for the 800V DC architecture NVIDIA standardised with its partners — intermediate bus converters, battery backup, auxiliary supplies — not the 650-volt consumer charger wave of 2021–23. And silicon carbide, the segment the market spent two years waiting on, has been demoted in public: “I see SiC coming slower than GaN. You see, I've also changed the order and put GaN upfront.” — Felix Grawert, Chief Executive Officer · 2026-10-29 That is a rare admission of reordering priorities mid-cycle, and it matters because SiC was the swing factor in every 2027 model built over the last year.
The refusal to size it, and what that is worth
Annualise €275m and you get €1.1bn of revenue. Consensus for 2027 sits near €800m. Grawert declined the arithmetic, and not out of coyness: “we are currently looking at lead times on the order of 12 months, plus or minus” — Felix Grawert, Chief Executive Officer · 2026-10-29 — up from the “7 to 10 months” he quoted in February. The binding constraint, repeated four times on the call, is ramping speed: hiring, supplier capacity, operational throughput. Unlike customers, AIXTRON has no clean room that fills up. Formal guidance waits until 4 March 2027.
There is a harder tell than the order number, and it is in the contract fine print. Customers are now signing capacity reservations and slot reservations with staged prepayments — Advanced payments hit €309m at quarter-end versus €44m at year-end 2025, driving working capital down €254m and nine-month operating cash flow to €311m against €128m a year ago. Order intake is a soft signal; customers paying cash to hold a delivery slot four quarters out is a hard one.
Two things temper the enthusiasm. First, gross margin hit 48% in Q3 and management spent the call disowning it — Q4 will be lower on a heavier G4/G5 and China mix, and 2027 should add only “a few percentage points” on operating leverage, not repeat the constellation-of-stars quarter. Second, and more importantly for the equity story, the CFO signalled a significant step-up in R&D into 2027 to fund the next tool generation — a real change from February's answer that €90m of R&D was simply the number and beyond that he “wouldn't want to predict.” Gross profit will not all fall to EBIT.
On the balance sheet, €945m of liquidity and €591m of net financial assets drew the inevitable buyback question, and got the same answer as before — “too early.” The nuance is worth holding onto: strip out the contract liabilities for unearned prepayments and the convertible, and the genuinely free number is closer to €275m.
The tape and the order book are telling different stories
Here is the tension worth carrying forward. Across the market, the AI optics complex has been cooling even as AIXTRON's order book inflected: co-packaged optics names are up roughly 73% over 360 days but down about 5% over the last 90, and the broad AI data center basket shows the same shape — up over a year, negative on the recent window. Either the tape is right and this is a capex pull-forward into 2027 that fades, or the tape is late and a European equipment maker is quietly selling into a demand pool the market has stopped paying for.
“I keep asking my marketing team, and I keep not trusting the numbers anymore.” — Felix Grawert, Chief Executive Officer · 2026-10-29 Grawert said that about his own near TAM estimate, which he has already publicly conceded was too low. Refusing to replace a wrong number with a bigger one is discipline. But it also leaves investors holding a doubled order book, a single end-market concentration in datacom lasers, and no framework to value either.