HBM Boom Reaches the Machine Makers: TOWA Orders Up 125.8%

TOWA booked ¥32.97 billion in orders for the July–September 2026 quarter, up 125.8% year over year, against sales of just ¥17.53 billion. The order-to-sales ratio of roughly 1.88 says customers are committing to packaging equipment nearly twice as fast as the Kyoto-based manufacturer can ship it. The AI memory boom, in short, has reached the people who make the machines.

The industry has mostly measured the artificial intelligence investment cycle through Nvidia, Micron and SK hynix revenue. High-bandwidth memory is now producing a second, quieter story in semiconductor capital equipment. HBM is not ordinary DRAM: dies must be thinned, pierced with through-silicon vias, stacked, bonded and inspected before a finished package ever meets an AI accelerator. Every added step is another machine sold, another process cost, another yield risk.

Two demand sources are working at once. Memory makers are adding capacity, and successive HBM generations demand more sophisticated deposition, etch, bonding and process control. SK hynix, Samsung and Micron are all investing in HBM capacity, and equipment suppliers on three continents are reporting the consequences.

How are the U.S. equipment majors exposed?

Lam Research posted September-quarter 2026 revenue of $8.22 billion, up 54.4% year over year and 22% sequentially, with a GAAP operating margin of 40.0%, versus 37.4% the prior quarter and 34.4% a year earlier. Lam's route into HBM runs through deep silicon etch, thin-film deposition and cleaning for TSV structures, plus conventional DRAM upgrades. The company credited foundry and memory demand alike, not HBM alone.

KLA's exposure is different in kind. As stack heights rise and interconnects shrink, one defective die can scrap an entire stack, which makes inspection and metrology progressively more valuable. KLA reported September-quarter revenue of $4.14 billion, up 28.9%. Its demand grows with process complexity, not merely wafer count, and those two curves can diverge even at the same customer.

Applied Materials, with July-quarter revenue of $9.12 billion (up 25.0%), covers deposition, CMP and interconnect. Its more interesting move came on October 26, 2026: an agreement to raise its stake in BE Semiconductor Industries to roughly 20% of shares outstanding, extending a relationship that began in 2020. The Dutch company builds hybrid bonding equipment. Note the wording — the deal awaits regulatory clearance, so 20% is a target, not a completed position.

Camtek offers the most concentrated play. September-quarter revenue reached about $161 million, up 28%, and management projects more than $800 million in 2027 against roughly $595 million expected in 2026 — roughly 35% growth, well above the broader equipment market. The small base cuts both ways: faster percentage growth, sharper sensitivity to customer qualification schedules.

What do the Asian suppliers confirm?

TOWA raised its fiscal-year forecast, ending March 2027, from ¥64 billion to ¥71 billion in revenue and from ¥10.24 billion to ¥13.15 billion in operating profit. The arithmetic is telling: ¥7 billion of extra revenue carries ¥2.91 billion of extra profit, an implied incremental margin of 41.6%. The company cites AI and data center investment, HBM included, though it serves other packaging applications too.

Wonik IPS, headquartered in Pyeongtaek, supplies the front end: PECVD, ALD and thermal processing for DRAM and NAND. Preliminary third-quarter revenue was about ₩337.8 billion, up 25.8%, with operating profit of ₩66.7 billion, up 142.0%. Its semiconductor equipment segment jumped 61.7% sequentially, from ₩189.6 billion to ₩306.6 billion.

On divisional estimates, Wonik's semiconductor equipment revenue should climb from ₩707.3 billion in 2025 to ₩1.049 trillion in 2026 and ₩1.408 trillion in 2027 — cumulative growth of 99.1%. Display equipment, by contrast, falls from ₩202.6 billion to ₩123.3 billion before recovering to ₩178.1 billion. Semiconductors rise from 77.7% of revenue to 89.5%. These figures cover all memory equipment, not HBM exclusively; they track the broader cycle, and they show front-end and back-end investment rising together.

The supply chain is genuinely global, per data from The Information Network: Lam and Wonik in front-end processing, KLA and Camtek in inspection, TOWA in molding, Hanmi Semiconductor of Incheon in thermocompression bonding for HBM stacking, DISCO of Tokyo in grinding and thinning, and BESI in hybrid bonding.

Is hybrid bonding the next pivot?

Current HBM relies on established bonding processes, but higher stacks and finer pitches strain package thickness, electrical performance and yield. Hybrid bonding — direct copper-to-copper bonding of prepared surfaces — could replace microbumps at finer pitches. It demands exceptionally flat, clean surfaces and tight defect control, which itself feeds spending on surface preparation and metrology. Timing, however, remains uncertain, and thermocompression and hybrid bonding may coexist for years.

The discipline worth keeping: equipment orders can run ahead of production needs. If HBM capacity eventually outgrows demand, memory prices soften and capital spending moderates, as it always has. Complexity spending cushions the cycle; it does not repeal it. The likely winners are not the loudest HBM names but the suppliers whose tools stay indispensable across successive memory generations.