Besi Books Its Best Order Quarter Ever, Then Asks You Not to Expect It in Q4

Orders nearly doubled and net margin doubled — but the HBM adoption clock quietly stretched again, and the sector tape is voting the other way.

BESI.AS · Earnings Call · 2026-10-22

An order book that has outrun the revenue line

Besi's third quarter of 2026 was, by the numbers, the quarter the hybrid-bonding thesis was supposed to produce. Revenue of €283.4 million rose 113.6% year over year. Orders of €342.6 million rose 96.1%. Net income of €107.2 million nearly quadrupled, and the net margin reached 37.8%, double the year-ago level. Year to date, revenue is up 69% to €718.1 million, orders have more than doubled to €905.2 million, and net cash has swollen to €246.6 million — up €254.4 million in twelve months, enough to lift the buyback to €100 million. Then management guided the fourth quarter to revenue that is flat, plus or minus 5%. That gap — a record order quarter, a flat revenue quarter — is the whole story. Morgan Stanley's Lee Simpson asked whether the issue was a customer push-out into the first half. Richard Blickman's answer was unusually blunt: “It's not a customer push-out.” — Richard Blickman, CEO · 2026-10-22 His explanation was mechanical and, on its own terms, credible — a bonder is one tool in a line, and delivery waits on every other tool and the building around it. But the phrasing matters, because it is the third consecutive quarter in which order intake has run well ahead of what the company will actually recognise:

Although we have received last quarter EUR 340 million something in orders, that doesn't mean that they are all shipped in the next week. Some are shipped in Q4, some are shipped in Q1, and also some are shipped in Q2. That's a normal pattern.
Richard Blickman, CEO · 2026-10-22

That is a push-out. The difference is that it is a push-out into the customer's H1 production capacity plan rather than a customer losing conviction — a distinction Besi would like the market to hold onto, because the alternative reading is that the order book is now a 2027 revenue story rather than a 2026 one.

The clock that keeps being reset

The more interesting shift is in the memory timeline. In February, management framed 2026 as the decisive year: “2026 will be a very important year to understand the adoption of hybrid bonding for HBM stacking.” — Richard Blickman, CEO · 2026-02-19 By April the schedule was still intact — “the time line is '26. And as we explained several quarters, that has not changed” — Richard Blickman, CEO · 2026-04-24 — with mainstream volume production landing in 2027. In July, the honest framing of the obstacle was qualification physics: “The major hurdles are simply cost and yield.” — Richard Blickman, CEO · 2026-07-23 Now, with ten weeks of 2026 left, the language has changed shape. The prepared remarks concede the near-term mechanism outright — “Severe memory shortages this year have extended the use of lower stack heights in current HBM production beyond prior expectations” — Richard Blickman, CEO · 2026-10-22 — and then reach much further out for the payoff:

We anticipate that hybrid bonding will be adopted in a range of stack heights over the next two years and in new innovative stacks combining HBM, custom advanced logic base dies, and the direct bonding of DRAM onto accelerator chips by decade end.
Richard Blickman, CEO · 2026-10-22

"A range of stack heights over the next two years" is a materially softer commitment than "production capabilities towards the end of this year for mainstream volume production in '27." And "by decade end" for direct DRAM-on-accelerator bonding is a 2030 sentence. Management's defence — repeated in four separate answers — is that stack height was never the driver, that energy, heat and bandwidth are, and that all three memory manufacturers continue developing at full speed. That may well be true. But it has now been the same answer for four quarters, while the date that matters has moved in one direction only. The fair counterweight is that the customer base really is broadening. Hybrid bonding adoption reached 23 customers, against 21 in July, 20 in April, 18 in February and roughly 16 a year ago, with cumulative unit demand above 215 systems and year-to-date hybrid orders more than double all of 2025. The logic and co-packaged optics legs of the story are compounding. It is specifically the memory volume that keeps slipping.

The vocabulary changed — and that is the signal

The words management chose this quarter are as informative as the guidance. TC Next is the single highest-ranked theme in Besi's own keyword set, displacing the memory-customer framing that dominated the second quarter. It is no longer a research curiosity: seven customers now, developed for bond pad pitches below the 20-micron market standard, fluxless, and explicitly positioned to extend the life of the reflow process in the gap before hybrid bonding. The Q3 memory order was for both hybrid bonding and TC Next — a hedge placed by the customer, and a hedge sold by the supplier. Two genuinely new items appeared. First, a strategic review — a sixteen-week deep dive with key customers, four weeks in, concluding in early January. The last such exercise preceded an upgrade to Besi's hybrid-bonding TAM numbers, so this is a live catalyst rather than housekeeping. Second, an unsolicited denial: “We do not view ASML as a competitor, contrary to recent analyst reports.” — Richard Blickman, CEO · 2026-10-22 Blickman spent two full answers on it, describing joint die stress work and the EPIC Center programme with Applied Materials. A CEO rebutting a sell-side note on a call is a company that has been told the narrative is under attack. What has not changed is the operational reality beneath it: market share gains in flip chip, multi-module attach and hybrid bonding, visibility lengthening from a quarter to one-to-two quarters, and a fab footprint able to support the ramp. The addressable market argument is intact — TechInsights now models 37% assembly-equipment growth in 2026 and a $12 billion market by 2028, and management claims it will grow at more than double that rate.

The tape is voting the other way

The uncomfortable contrast is that the market has spent the last quarter marking down precisely the complex Besi just sold into. Across the global theme tape, AI data center is down 5.6% over thirty days with six advancers against forty-eight decliners; high bandwidth memory is down 6.9% over ninety days with eleven decliners and no advancers; co package optics is down 5.1%; and design win activity is the single worst thirty-day decliner at -10.3%. Every one of those themes was a top-360-day advancer. The AI packaging complex is not broken on a one-year view — it is being de-rated on a one-quarter view. So either the tape is early and Besi's €342.6 million order quarter is the confirmation, or Besi is late and the flat Q4 is the first quiet admission that the digestion has started. Blickman has himself raised the overcapacity question before, in July, with the fatalism of a man who has watched six cycles: an up-cycle typically runs six to eight quarters. What the January strategic review, and the Q4 print, will not be able to avoid is which of those two readings the first half of 2027 delivers.