ASMPT Limited (0522.HK): Deep Dive Research Report

Research cut-off: 8 October 2026. ASMPT has a December year-end and reports quarterly. Its most recent release is Q2 / 1H 2026, published on 29 July 2026. Q3 2026 results would normally come in late October 2026, so they are not due yet.

1. What the Company Does

ASMPT makes the machines that take a finished silicon chip and turn it into something that can go inside a phone, a car or an AI server. A chip fabricated in a wafer fab is a fragile sliver of silicon with no way to connect to anything. Someone has to cut it from the wafer, pick it up and place it precisely. It then has to be bonded to a substrate or to another chip, wired or bumped so electricity can flow, and sealed. After that, the packaged chip has to be soldered onto a circuit board alongside hundreds of other components. ASMPT sells equipment for both stages. Its Semiconductor Solutions (SEMI) segment builds the back-end packaging tools that assemble the chip itself. Its SMT Solutions segment builds the printers and pick-and-place machines that populate circuit boards. Few equipment companies span both. That position, "from wafer to board", is the company's oldest selling point, and it is also the source of its current strategic question (Section 2).

How it got here. ASMPT began in Hong Kong in 1975 as the Asian arm of the Dutch equipment maker ASM International. According to the company's history page, it was set up as ASM's Asian marketing division, initially reselling semiconductor materials and equipment in the region. Wikipedia's account attributes the founding to ASM International founder Arthur del Prado together with Patrick Lam, an electrical engineer. Through the 1980s it moved from distribution into manufacturing its own wire bonders, die bonders and lead frames. It listed on the Hong Kong Stock Exchange in 1989. Three pivots explain the company you see today:

1. Becoming a back-end champion (1990s-2000s). Low-cost, high-volume assembly tools built in Asia for Asian assembly houses made ASMPT one of the largest back-end equipment suppliers in the world. ASM International's stake was once a majority. It has been sold down over time and now sits just under a quarter (Section 10).
2. Buying the board-assembly business (2011-2014). In January 2011 ASMPT bought Siemens' Electronics Assembly Systems business, home of the SIPLACE placement machines made in Munich. In 2014 it bought the UK screen-printer maker DEK from Dover. These two deals created the SMT segment almost overnight, and around 2015 ASMPT became the largest SMT equipment supplier in the world. Singapore became group headquarters in 2011.
3. A string of technology tuck-ins, then a pruning (2014-2026). ALSI (laser dicing, Netherlands), AMICRA (ultra-precision die bonding, Germany), NEXX (deposition, US) and Critical Manufacturing (factory software, Portugal) were added in 2014-2018, followed by several smaller Chinese acquisitions. Under the current strategy the direction has reversed. NEXX was sold to Applied Materials, closing on 3 June 2026. In January 2026 the board opened a review that could see SMT divested, spun off or listed separately (Section 7). The company that is emerging is a narrower back-end packaging specialist built around AI-era bonding tools.

The core value proposition. As transistor shrinking slows, chipmakers get more performance by packaging several chips together. They stack memory dies into high-bandwidth memory (HBM), place GPU and memory dies side by side on an interposer, or attach very large chips to substrates. Each step demands that tiny solder bumps, tens of microns apart and thousands in number, are aligned and joined without warping the silicon. ASMPT sells the bonders that make these joints at production speed and acceptable yield. On the board side, it sells placement lines that can mount the dense mix of components on an AI server motherboard or an optical transceiver at high throughput.

What makes it hard. A thermo-compression bonder (TCB) must heat a die and substrate, align them to within a micron or two, and press them together in a few seconds. Heat expands and warps both parts while this happens, and the tool has to compensate. ASMPT also removes surface oxides with a plasma ("active oxide removal") so no flux residue is left behind. This is precision mechanics, thermal control and process chemistry in one machine. It also has to be proven inside a customer's production line before anyone buys it in volume (Section 4).

A concrete example. Picture an outsourced assembly house (OSAT) building packages for an AI accelerator. The compound die, a logic chip plus its neighbours on an interposer, is now so large that the conventional "mass reflow" method warps it, and bumps fail to connect. The OSAT installs ASMPT chip-to-substrate TCB tools. Each tool picks one compound die, aligns it over its substrate and bonds it under heat and pressure, one unit at a time, with closed-loop control of force and temperature. The finished package then goes to an electronics manufacturer, which may well use ASMPT SIPLACE lines to mount it onto the server board alongside power stages, memory and connectors. In July 2026 OSAT customers placed bulk orders for more than 50 of these C2S tools in one go. That order is the clearest picture of the business ASMPT now wants to be.

2. Business Segments

ASMPT reports two segments. Inside SEMI, it also reports an "Advanced Packaging" (AP) product grouping that cuts across both segments, and it is the AP lens management uses when it talks about growth.

2.1 Semiconductor Solutions (SEMI), roughly 57% of 1H 2026 revenue (about 55% in FY2025)

What it does. SEMI sells back-end assembly and packaging equipment to chipmakers (IDMs), foundries, memory makers and OSATs. The product range runs from high-volume "mainstream" tools to advanced packaging tools:

• Mainstream: wire bonders, die bonders, encapsulation and optoelectronic and image-sensor assembly. These serve power devices, automotive chips, LEDs and consumer ICs.
• Advanced packaging: TCB, flip-chip bonders for 2.5D packages, hybrid bonders, laser grooving and dicing, panel-level packaging tools, and ultra-precision photonics assembly.

Core capability. SEMI's hard-to-copy skill is sub-micron placement under heat at production throughput, along with the process know-how to make it yield: plasma oxide removal, flux-free joints, and warpage control on very large dies. It also has decades of installed base in Chinese and Southeast Asian OSATs, which gives it the service footprint and the reference customers a new entrant lacks.

Why it is separate. SEMI sells into semiconductor capex, which follows wafer starts, packaging technology changes and chip-design cycles. Its customers are a small number of very large chip and assembly companies. Its tools are qualified one process step at a time, often over a year.

Role in the group. SEMI is the growth engine and, on management's own statements, the higher-margin business. Every strategic action since 2025 (the NEXX sale, the SMT review, and hiring a CEO from a foundry and advanced-packaging background) points to SEMI as the future company. Advanced packaging was about 30% of group revenue in 1H 2026. That share was about 39% in 1H 2025, before NEXX was split out. The drop is not AP shrinking: SMT and SEMI's mainstream business both grew faster in the period.

Main rivals (one line): BE Semiconductor (BESI) and Kulicke & Soffa in logic and hybrid bonding; Hanmi Semiconductor, Hanwha Semitech and Samsung's SEMES in HBM TCB; K&S and Chinese domestic makers in wire bonding.

2.2 SMT Solutions, roughly 43% of 1H 2026 revenue (about 45% in FY2025)

What it does. SMT sells the equipment that builds printed circuit board assemblies: DEK stencil printers that lay down solder paste, SIPLACE placement machines that mount components at tens of thousands of placements per hour, inspection systems and line software. Customers are electronics manufacturing services (EMS) firms, automotive tier-1 suppliers, industrial and telecom equipment makers, and increasingly AI-server board builders and optical-transceiver makers. The segment also sells system-in-package (SiP) and "chip assembly" tools, which apply SMT-style placement to very small modules and blur the line with SEMI.

Core capability. SMT's edge is flexible, high-mix placement accuracy on large-format boards. It also brings German-engineered machine reliability and software that connects a whole factory line. The high-flex machines that management says are gaining share in AI server boards in 2026 rely on handling oversized boards with a mix of tiny passives and large processors.

Why it is separate. It came from acquisitions (SIPLACE and DEK). Its R&D and manufacturing are centred in Europe (Munich, Weymouth). It sells to a different buyer set, electronics assemblers rather than chipmakers. Its demand followed automotive, industrial and consumer electronics cycles until AI servers arrived.

Role in the group. On management's own framing, SMT is the lower-margin business. Its recent history was a drag: automotive and industrial weakness drove a sharp decline in 1H 2025 SMT revenue. In 2026 it became an AI beneficiary through server boards, optical modules and Chinese EV electronics, and set record bookings in Q1 and again in Q2 2026. Since January 2026 it has been under strategic review, so management's stated priority is SEMI. SMT is being run as an asset whose future ownership is open.

Main rivals (one line): Fuji Corporation, Yamaha Motor's robotics division, Panasonic Connect, JUKI, Hanwha and Mycronic.

2.3 Geographic mix and export history

ASMPT has always sold mostly into Asia because that is where chips are packaged and boards are stuffed.

• China was about 41% of FY2025 revenue and 42% of 1H 2026. That is the largest single country, and it has been central since the 1989 move into Shenzhen manufacturing. Chinese demand is weighted to SEMI mainstream tools (wire and die bonders for domestic OSATs, power devices and EV electronics) and to SMT. Management describes Chinese advanced packaging as smaller but promising.
• Asia outside China (Taiwan, Korea, Malaysia, Singapore, Vietnam and Japan) rose from about 24% to 34% of revenue in FY2025, driven by TCB shipments to logic packaging and HBM customers.
• Europe fell from about 20% to 13% and the Americas from about 15% to 11% in FY2025. Together they were 21.3% of 1H 2026 against 23.7% a year earlier. These are mostly SMT regions (European automotive and industrial electronics, North American EMS), so their decline mirrors SMT's automotive and industrial slump.

The mix has rotated twice in two years: toward non-China Asia as TCB ramped, then back toward China as Chinese mainstream and SMT demand recovered in 2026.

3. Products and Operations

3.1 The products that matter to the investment case

Thermo-compression bonding (TCB): the FIREBIRD family. This is ASMPT's most important product line. TCB bonds one die at a time under controlled heat and force. It is used where conventional mass reflow fails: large dies that warp, fine bump pitches that bridge, and tall memory stacks that must stay flat. ASMPT sells TCB into three uses:

1. Logic chip-to-substrate (C2S). Bonding very large compound AI dies onto organic substrates. Management describes ASMPT as the "process of record" and primary supplier here. In mid-2025 it described itself as the sole C2S supplier. It had passed 500 installed TCB units worldwide by the first half of 2025.
2. Logic chip-to-wafer (C2W). Bonding chiplets onto a wafer or interposer at ultra-fine pitch. This uses ASMPT's plasma active-oxide-removal (AOR) process, which removes oxides without flux and so removes a cleaning step and a source of voids. The process was qualified at a leading foundry in Q3 2025. Four tools were booked from a leading advanced-logic customer in Q1 2026, and a bulk order for advanced CPUs came from a leading IDM in Q2 2026.
3. HBM memory stacking. Stacking 8, 12 and soon 16 DRAM dies. ASMPT installed a bulk order for HBM3E 12-high at a leading HBM maker in 1H 2025, was first to win HBM4 12-high orders from multiple HBM makers, and has a flux-based tool sampling for HBM4 16-high with a fluxless AOR version in qualification. ASMPT's tools are in HBM4 high-volume manufacturing at multiple memory makers. Memory orders have nonetheless been lumpy (Section 8.1).

The difference between flux-based and fluxless matters. Fluxless bonding avoids residue that is hard to clean from the narrow gaps in a 16-high stack. HBM4's tighter dimensions and the sub-micron placement accuracy it requires favour suppliers with precision optics and plasma process capability. That is the basis on which ASMPT has taken HBM share from Korean incumbents.

Hybrid bonding: LITHOBOLT and LITHOBOLT G2. Hybrid bonding removes solder bumps entirely and fuses copper-to-copper and oxide-to-oxide directly, which enables pitches below 10 microns. ASMPT's second-generation platform shipped in 2H 2025, received customer buy-offs, and is at sampling with key logic and memory customers. ASMPT does not do wafer-to-wafer hybrid bonding, only die-to-wafer. Management puts the volume inflection in the 2029-2031 window. Hybrid bonding is both an option for ASMPT and a long-term threat to TCB (Section 8.2).

Flip chip. High-throughput flip-chip bonders for large-format 2.5D AI packages and panel-level fan-out. This line grew strongly in 1H 2026.

Photonics. Ultra-precision die attach, much of it descended from the AMICRA acquisition, used to assemble lasers, photodiodes, lenses and fibre arrays in pluggable optical transceivers (800G and 1.6T). Further out, it is aimed at co-packaged optics (CPO), where optical engines sit next to the switch chip. ASMPT also supplies SMT lines and AP tools into the same transceiver factories, so it can offer the whole assembly chain for an optical module. Photonics was roughly 7% of group revenue in 1H 2026.

Mainstream SEMI. Wire bonders and die bonders are the volume business. They are less glamorous, but they carry the China exposure and the power-device demand linked to AI data centre power delivery and EVs.

SMT. SIPLACE placement platforms (including the high-flex machines used for large AI server boards), DEK printers, inspection, SiP and chip-assembly tools, and line software.

3.2 Manufacturing and delivery

ASMPT is an assemble-to-order equipment maker. It designs the machines, buys precision components (motion stages, optics, lasers, vision systems, heaters) from a supplier base, and assembles, calibrates and tests each tool before it ships. A customer purchase order is followed by:

1. Configuration of the tool to the customer's process (bond head, heating profile, AOR module).
2. Component procurement. In 2026 this is the bottleneck, with longer lead times for precision materials affecting both segments.
3. Assembly and factory acceptance test at ASMPT.
4. Shipment, installation and site acceptance at the customer.
5. Process qualification and buy-off, after which repeat orders follow.

Management said in July 2026 that this conversion from booking to billing now takes 6-9 months, and longer for advanced packaging tools. That is why the large July 2026 C2S order ships mostly in Q1 2027.

Sites. Main production is in Singapore (two Tech-Park buildings in Yishun), Malaysia, Huizhou in China, Munich (SIPLACE) and Weymouth, UK (DEK). A notable plant event: in Q3 2025 ASMPT put a Chinese subsidiary, ASMPT Equipment (Shenzhen) Co., Ltd. (AEC), into voluntary liquidation. The restructuring and inventory write-off caused a reported loss for the quarter, with a stated annual cost saving to follow. Production consolidated into the remaining Asian sites. The NEXX deposition business, a US operation, left the group in June 2026.

Value chain, for visualisation: wafer fab → [ASMPT SEMI: laser grooving/dicing → die bond / flip chip / TCB / hybrid bond → wire bond → encapsulation] → packaged chip → [ASMPT SMT: paste printing → component placement → inspection] → finished board → server, phone, car or transceiver.

4. Customers

Who buys. There are five distinct buyer groups:

1. OSATs (outsourced assembly and test houses in Taiwan, China and Southeast Asia) buy mainstream wire and die bonders in volume and, increasingly, C2S TCB for AI packages. The July 2026 order for more than 50 C2S tools came from OSAT customers.
2. Leading-edge logic makers. A leading foundry qualified ASMPT's C2W plasma-AOR process in Q3 2025. A "leading advanced logic customer" booked four C2W tools in Q1 2026. A "leading global IDM" placed a bulk C2W order for advanced CPUs in Q2 2026. ASMPT does not name them.
3. HBM makers. The company says it supplies "multiple HBM players". Korean trade press (The Elec, via TrendForce, December 2025) reported that SK hynix ordered seven dual-head TC bonders from ASMPT for HBM4 and that roughly half of SK hynix's HBM4 TC bonders at the time were ASMPT's. ASMPT itself does not name SK hynix.
4. Optical module makers, for photonics assembly and SMT.
5. Electronics assemblers: EMS firms, automotive tier-1s, industrial and telecom OEMs, AI-server board makers and Chinese EV supply chains, all for SMT.

Who decides and how. For advanced packaging tools the decision sits with the customer's packaging process-development team. It is then ratified by manufacturing and procurement. The criteria, roughly in order, are:

• whether the tool can hit the yield target on the specific package (placement accuracy, warpage control, void-free joints);
• throughput (units per hour per head);
• cost of ownership, including consumables such as flux and cleaning steps;
• the supplier's ability to ship dozens of tools quickly when a ramp is approved.

The cycle runs from engineering evaluation to a tool buy-off on one line, then low-volume production, then a bulk order. That can take 12 to 24 months and is tied to the customer's product roadmap: a new GPU generation, HBM generation or CPU tile architecture.

For mainstream and SMT tools the buyer is usually the factory engineering or capex team, buying on throughput, uptime, footprint, price and local service. Cycles are short, a few months, and track utilisation. When OSATs or EMS factories are full, they order.

Why customers choose ASMPT.

• It is often already the process of record. In C2S logic, ASMPT was in the line first, so the customer's package design, recipes and yield learning are built around its tools.
• It is ahead on fluxless processing. Plasma AOR lets customers drop a flux-clean step, which matters most at fine pitch and in tall memory stacks.
• It offers breadth. One supplier covers flip chip, TCB, hybrid bonding sampling and the SMT line downstream. For photonics customers it covers the whole optical-module assembly flow.
• Its service footprint is local in China and Southeast Asia, where most OSAT capacity sits.

Switching costs. These are high in advanced packaging and moderate elsewhere. Once a TCB recipe is qualified for a package, switching supplier means requalifying the process and possibly re-validating reliability with the end customer, such as the GPU designer. Fabs avoid doing this mid-generation. The lock-in resets at each new package generation, though, and that is exactly when competitors attack. SK hynix's HBM history shows it: Hanmi held the TC bonder business at SK hynix for HBM3E, yet SK hynix opened the door to ASMPT for HBM4. In mainstream tools and SMT, switching costs come mainly from operator familiarity, spare parts and software integration on the factory floor. They are real but not decisive against a cheaper, good-enough tool.

Concentration. Low by equipment-industry standards. The top five customers were about 16% of FY2025 revenue and about 19% of 1H 2026. The rise reflects bigger lumpy TCB orders. This spread across hundreds of OSATs, IDMs and assemblers is a structural cushion. Advanced packaging is the exception: within AP, a handful of customers (one or two GPU supply chains, three HBM makers, a few OSATs) drive most of the orders.

Contract structure. This is almost entirely capital-equipment sales by purchase order, with no long-term supply agreements disclosed. There is a smaller recurring stream from spare parts, service and software. Revenue predictability comes from the order backlog, which was very high after a 1H 2026 book-to-bill of 1.43 and a 6-9 month conversion time. Beyond the backlog there is no contractual visibility. Bulk orders tied to a customer's ramp can be pushed out if the ramp slips.

5. Competitive Landscape and Moat

ASMPT competes in four distinct arenas, each with a different structure.

1. Logic TCB (C2S and C2W). ASMPT's strongest position. In C2S, the main alternative for most of the last generation was mass reflow, a process step rather than a rival tool. As compound dies grew past the size reflow could handle, the market moved to TCB, and ASMPT was the incumbent. Kulicke & Soffa (US) has competed with its own fluxless TCB and is the most credible logic challenger. In C2W, ASMPT's plasma-AOR approach competes with BESI and K&S. ASMPT wins on qualification history and fluxless process maturity. It could lose if a customer designs a new package around a competitor's tool at the next node, or around hybrid bonding.

2. HBM TCB. Here ASMPT is the challenger. Hanmi Semiconductor (Korea) built a near-monopoly at SK hynix by focusing on HBM TC bonders early, when ASMPT and BESI were not prioritising them. Industry data cited by Korean analysts put Hanmi at about 71% of cumulative HBM TC bonder revenue through Q3 2025, SEMES (Samsung's in-house equipment arm) at about 13%, and ASMPT at about 6%. SemiAnalysis reported in August 2025 that Hanmi held 100% of SK hynix's TC bonder share until late 2024. It also reported that a dispute over SK hynix's orders with Hanwha led Hanmi to withdraw field-service engineers, and that ASMPT and BESI were rushing to improve their HBM-specific bonders. ASMPT's opening came from HBM4's tighter placement requirements and from SK hynix's wish for a second source. Hanmi has not gone away: it won its first HBM4 order for its new Griffin bonder from SK hynix in June 2026 (Tech Times). Samsung buys largely from SEMES. ASMPT wins on accuracy and fluxless capability. It loses on incumbency, on Korean vendor relationships and on Samsung's in-house preference.

3. Hybrid bonding. BESI (Netherlands) is the recognised commercial leader in die-to-wafer hybrid bonding, with more than a hundred systems shipped, and it works with Applied Materials on an integrated process. ASMPT is a follower with a second-generation tool at sampling. It has no wafer-to-wafer capability, where EV Group and Tokyo Electron are the reference suppliers.

4. Mainstream back-end and SMT. In wire bonding, ASMPT and Kulicke & Soffa have long been the two global leaders, while Chinese domestic makers are winning share in local, lower-end applications. In SMT placement, ASMPT, Fuji, Yamaha, Panasonic and JUKI hold most of the market. ASMPT became the largest SMT equipment supplier around 2015 by virtue of SIPLACE and DEK. The Japanese makers compete strongly in Asian high-volume consumer lines. ASMPT's edge is high-mix, large-board flexibility and European automotive accounts.

Barriers to entry. In advanced packaging they are high but not permanent. A new entrant needs several things at once:

• precision motion and optics at sub-micron level under heat;
• a proprietary bonding process (AOR, flux handling, warpage control) backed by patents and years of yield data;
• a reference production line at a top customer, the most important barrier because no fab wants to be first;
• the balance sheet and supply chain to deliver dozens of tools in a quarter when a ramp is approved.

Hanmi shows that a focused, well-connected local player can break in when it bets early on a niche the leaders ignore. SEMES shows that a captive buyer can build its own. In mainstream wire bonding and SMT, barriers are moderate. Chinese makers are already climbing the ladder in China, helped by procurement preferences for domestic tools.

Market share. ASMPT's own estimate is roughly 30% of the TCB market in 2025, with a target of 35-40% by 2028. This share is concentrated in logic C2S and C2W, while memory is the smaller part of ASMPT's TCB business. It is consistent with the third-party HBM-only data above because the two use different denominators. Share in each sub-market reflects who was qualified first at each customer for each generation.

Structural shifts.

• Consolidation and focus: ASMPT sold NEXX to Applied Materials, and Applied is also a BESI partner and shareholder, so a fab-equipment giant is moving into packaging.
• Korean localisation: Hanmi, Hanwha and SEMES sit beside SK hynix and Samsung.
• Chinese domestic substitution in mainstream tools.
• Technology substitution: hybrid bonding replacing TCB at the finest pitches. JEDEC's relaxation of HBM stack height delayed this, and further relaxation would delay it again.

Moat: Narrow - switching costs, intangible assets (process know-how and patents), qualification barriers

Qualification lock-in at each package generation (Section 4) and ASMPT's process-of-record position in logic C2S are real. The fluxless AOR process is a defensible piece of know-how, not a commodity. The moat does not qualify as wide, because it has demonstrably not held where a determined rival was already in place. Hanmi kept ASMPT out of SK hynix's HBM3E line for years, and SEMES largely keeps it out of Samsung. In mainstream tools and SMT the advantages are service and installed base, not structural protection. The single thing most likely to erode the moat is the move to hybrid bonding at the finest pitches, where BESI and Applied Materials, not ASMPT, hold the reference position.

6. Industry

What drives demand. Back-end equipment demand is the product of three forces:

1. Unit volumes of chips and electronics. Phones, PCs, cars, industrial gear and servers drive wire-bonder, die-bonder and SMT demand through factory utilisation.
2. Packaging intensity. AI accelerators need advanced packages: large interposers, many chiplets and HBM stacks. Each needs more bonding steps per chip and slower, more expensive tools per unit of output. This is the force that has decoupled advanced-packaging tool demand from overall unit volumes since 2023.
3. Technology transitions. Each new HBM generation (HBM3E → HBM4 → HBM4E/HBM5) and each new GPU or CPU packaging architecture triggers a new tool qualification and often a new buying round.

Size and growth. SEMI's mid-2026 forecast raised global semiconductor equipment sales to a record of about US$166bn for 2026, up about 23%, on AI-driven logic, memory and advanced-packaging spending. Assembly and packaging equipment is a small slice of that total, in the mid-single-digit billions of dollars, and SEMI expects it to grow for a third consecutive year in 2026. ASMPT's own estimate of the TCB market is about US$760m in 2025, rising to about US$1.6bn by 2028, a roughly 30% annual growth rate. That is one of the fastest-growing niches in back-end equipment, though still small next to front-end lithography or etch.

Position in the global supply chain. Back-end equipment makers sit between component suppliers (precision motion, optics, lasers) and the chip and assembly companies. The industry has a three-tier structure:

• global Western and Asian specialists (ASMPT, BESI, K&S, Disco, Towa);
• Korean and Japanese specialists close to their domestic memory and electronics champions;
• a fast-growing Chinese tier focused on the domestic market.

ASMPT straddles the first and second. It is global, but its manufacturing and customer base are Asian.

Import substitution. China is pushing domestic back-end equipment, as it is for front-end tools. In mainstream wire and die bonding, local suppliers are gaining share at Chinese OSATs. Korea's memory makers prefer local suppliers where performance allows. Both trends cap foreign suppliers' share in those two markets over time.

Regulation. US export controls on advanced computing chips and semiconductor equipment to China apply mainly to front-end tools, but they shape which advanced packaging capacity can be built in China and for whom. Tariff policy affected sentiment through 2025, and management repeatedly cited it as an uncertainty. There are no product-approval regimes of the medical or aerospace kind. Qualification is commercial, not regulatory.

Cyclicality. Back-end equipment is among the most cyclical parts of semiconductors. OSATs and EMS firms buy when utilisation is high and stop abruptly when it falls, so equipment orders swing far more than chip unit volumes. The 2023-2024 downturn in consumer, automotive and industrial electronics cut mainstream and SMT demand sharply while AI packaging boomed, a split cycle. In 2026 both halves are rising together for the first time in years. Historically that coincidence has been followed by an inventory and capex digestion phase.

Tailwinds:

• AI accelerators and custom ASICs moving to larger, multi-die packages;
• HBM content per accelerator rising each generation;
• chiplet architectures spreading to CPUs;
• optical interconnect bandwidth (800G → 1.6T → CPO);
• data-centre power delivery driving power-device packaging;
• a recovery in automotive and industrial electronics in China.

Headwinds:

• concentration of AI capex in a handful of buyers whose plans can shift;
• hybrid bonding gradually replacing bump-based bonding at the leading edge;
• domestic substitution in China and Korea;
• component supply tightness extending lead times industry-wide;
• tariff and export-control uncertainty.

7. Growth Triggers

Ranked by potential to change the business, most material first.

• Bulk C2S TCB order from OSATs ships into 2027. In July 2026 OSAT customers ordered more than 50 chip-to-substrate TCB tools for AI compute packaging. Most will ship in Q1 2027 and some in 2H 2026, and management describes ASMPT as the primary supplier and process of record for C2S. C2S repeat orders have come up at every call since Q1 2025. (Q2 2026 results call and order announcement, 29 Jul 2026)

• TCB market roughly doubling by 2028, with a share target of 35-40%. Management raised its TCB market estimate from over US$1bn by 2027 to about US$1.6bn by 2028 and set a 35-40% share target. The earlier US$1bn view was stated in July and October 2025, so this target has been repeated and raised. (FY2025 results call, 4 Mar 2026; Q2 2025 call, 23 Jul 2025; Q3 2025 results, 28 Oct 2025)

• Chip-to-wafer TCB for chiplet CPUs and GPUs. After four C2W tools booked from a leading advanced-logic customer in Q1 and a bulk order from a leading IDM for advanced CPUs in Q2, management expects C2W to become "more meaningful" in 2027 if next-generation GPUs move to chiplet architectures. (Q1 2026 call, 22 Apr 2026; Q2 2026 call, 29 Jul 2026)

• HBM4 16-high and HBM5. A key memory maker is sampling ASMPT's flux-based TCB for HBM4 16-high while qualifying its fluxless AOR process. In Q2 2026 ASMPT set up an exclusive joint evaluation programme for HBM5 with a key memory player. Management argued that stack heights beyond 900 microns would extend TCB's life through HBM5. The memory-order shortfall behind this trigger is covered in Section 8.1. (Q1 2026 call, 22 Apr 2026; Q2 2026 call, 29 Jul 2026)

• Co-packaged optics after the pluggable ramp. Management expects photonics bookings to keep growing in Q3 2026. It puts the CPO inflection at "2027-2028, more likely 2028", with ASMPT supplying SMT, advanced packaging and sub-micron photonics assembly tools across the optical value chain. 1.6T transceiver bulk orders were cited in Q1 2026.

  "We have built a strong position across the entire photonics value chain, extending from pluggable optical transceivers to CPO." - Robin Ng, then Group CEO (Q2 2026 call, 29 Jul 2026)

• Outcome of the SMT strategic review. On 21 January 2026 the board began assessing options for SMT, including a divestiture, joint venture, spin-off and listing, or keeping it, with Morgan Stanley Asia advising. In April management said it had received interest but made no decision. A transaction would leave ASMPT as a pure back-end packaging company. (Announcement, 21 Jan 2026; Q1 2026 call, 22 Apr 2026)

• Breaking into top hybrid-bonding customers. Management's 2026 goal is to break into leading hybrid-bonding customers in logic and HBM with the second-generation platform, now at sampling. It frames the volume opportunity as a 2029-2031 event. (Q2 2026 call, 29 Jul 2026; repeated from FY2025 call, 4 Mar 2026)

• Near-term order momentum and a panel-level entry. Management guided Q3 2026 bookings to grow sequentially, led by TCB and photonics, and said OSAT and PCB-maker capex should stay strong through 2026. It shipped a chip-on-panel TCB tool for qualification, an entry point into panel-level packaging. (Q2 2026 call, 29 Jul 2026)

8. Key Risks

8.1 The HBM opportunity proves smaller or later than the TCB plan assumes. Mechanism. Memory is the largest slice of the TCB market in management's own model until logic matures. ASMPT is the challenger there (Section 5). Orders arrive in bulk only when a memory maker freezes the tool set for a new generation. If HBM4 qualification drags, if SK hynix spreads its HBM4 and HBM4E volume back toward Hanmi, or if Samsung keeps buying SEMES, ASMPT's memory TCB stays lumpy and small.

Evidence. Management said in April 2026 that the last bulk memory TCB order came in Q4 2025. Memory TCB shipments were muted in 1H 2026 because customers were taking longer to meet HBM4 specifications. SK hynix was reported to be planning an order of about 100 TC bonders around March 2026, and Hanmi announced its first HBM4 Griffin contract with SK hynix in June 2026. Management's position in April:

"well-positioned to receive future orders as memory makers are ready." (Q1 2026 call, 22 Apr 2026)

Calibration: high probability of continued lumpiness, moderate effect. Logic TCB carries the TCB plan regardless.

8.2 Hybrid bonding arrives sooner and ASMPT is not the supplier. Mechanism. Hybrid bonding replaces the solder bump, and with it the step TCB performs, at the finest pitches. If HBM makers adopt it for HBM5 or later, or leading logic packages move from microbump C2W to hybrid-bonded stacks, TCB's growth runway shortens. BESI and Applied Materials lead die-to-wafer hybrid bonding. ASMPT has no wafer-to-wafer offering.

Evidence. Management places the inflection at 2029-2031 and argues relaxed HBM stack-height limits favour TCB. SemiAnalysis notes that JEDEC already relaxed stack height once (to 775 microns) and that this pushed hybrid bonding out.

Calibration: low near-term probability, high long-term impact. This is the single structural threat to the TCB thesis.

8.3 An AI capex digestion phase hits an unusually full order book. Mechanism. The 1H 2026 book-to-bill of 1.43 is the highest since 1H 2021, and conversion now takes 6-9 months. Bookings are being pulled forward by three things at once: AI packaging, AI server SMT and a Chinese mainstream recovery. Back-end equipment orders overshoot and then collapse (Section 6). A pause by a few GPU or ASIC programmes, or OSATs digesting the C2S tools they ordered in bulk, would turn into push-outs and possibly cancellations, at the same time as the mainstream cycle rolls over.

Evidence. The last time book-to-bill was this high, 2021, was followed by the 2022-2023 downturn. Management itself flagged limited visibility into 2H 2026 at the April call.

Calibration: moderate to high probability over 2-3 years, moderate to severe effect.

8.4 China concentration and geopolitics. Mechanism. China is about 42% of revenue (Section 2.3) and mostly mainstream SEMI and SMT. Three things could cut that revenue: tighter US or allied export controls extending to back-end or advanced-packaging tools; Chinese procurement shifting to domestic wire and die-bonder makers; or retaliatory tariffs. ASMPT's own Chinese production footprint shrank with the AEC liquidation (Section 3.2), so service and local presence rely on fewer sites.

Calibration: moderate probability, moderate effect. It falls hardest on the mainstream business that is currently recovering.

8.5 The SMT review destroys value or distracts. Mechanism. SMT is being reviewed just as it turns into an AI-server and optical-module business with record bookings. The risks are:

• a sale at a cyclical peak that later looks cheap, or one that fails to attract a credible buyer and leaves the segment in limbo, with key engineers and customers unsure who will own it;
• loss of the "wafer to board" cross-selling that management promotes in photonics, where SMT and AP tools sell into the same transceiver factories;
• stranded group costs after separation.

Evidence. There is no timetable and no guarantee of a transaction (21 Jan 2026 announcement). Management reported interest but no decision in April 2026.

Calibration: moderate probability of execution friction, moderate effect.

8.6 Leadership transition mid-cycle. Mechanism. Robin Ng, who led the strategy that produced the current portfolio, retired on 11 August 2026. His successor, Bassel Haddad, comes from outside the company and the region. Per ASMPT's August 2026 announcement, he spent 14 years in senior roles at Intel and then led SkyWater Technology's advanced packaging business, later becoming its SVP and GM of Foundry Solutions and Technology Platforms. The announcement gives no dates. A new CEO may re-scope the SMT review, change the hybrid-bonding investment or reset guidance practice (Section 9). He will also need to manage the HBM customer relationships Ng and his team built.

Calibration: moderate probability of strategic change, low to moderate effect. Most of the operating team is unchanged.

8.7 Supply-chain lead times cap delivery. Mechanism. Longer lead times for precision materials stretch booking-to-revenue conversion (Section 3.2). If they lengthen further, customers with urgent AI ramps may split orders with competitors who can deliver sooner. Each new HBM or packaging generation is effectively a race for slots.

Calibration: moderate probability, low to moderate effect.

9. Walk the Talk

Results events used (six quarters):

1. Q1 2025: results and call, 30 Apr 2025
2. Q2 / 1H 2025: results and call, 23 Jul 2025
3. Q3 2025: results, 28 Oct 2025
4. Q4 / FY2025: results and call, 4 Mar 2026
5. Q1 2026: results and call, 22 Apr 2026
6. Q2 / 1H 2026: results and call, 29 Jul 2026 (71 days before this report)

The commitments that tell you something.

The guidance drift. Through 2025 ASMPT guided ranges and landed near the middle. In 2026 it started landing at or above the top: Q2 2026 beat the top of its range, and the April call expected SMT bookings to fall just before they set a record. On the July 2026 call management also stressed that revenue and EPS had come in above consensus. That pattern suggests management sets deliberately cautious near-term numbers in an upswing, which reduces the information in each quarterly guide. It also means a quarter that lands only in the middle of its range would be a meaningful signal.

Hybrid bonding: the tool arrived, the market did not. In July 2025 the second-generation hybrid bonder's shipment to an HBM customer was a concrete promise for the next quarter, and it was delivered. Twelve months later the timing of the market it serves had moved out materially:

"2029, 2030 kind of timeframe, or even 2031." - management on the hybrid bonding inflection (Q2 2026 call, 29 Jul 2026)

The engineering commitments were kept. The adoption timeline is the customers' to set, and it has moved out. For ASMPT this cuts both ways: it delays a business in which ASMPT is a follower, and it extends one (TCB) in which it leads.

HBM: confident framing, thin orders. The October 2025 claim of being first to win HBM4 orders from multiple players was true as stated. Since then, though, the memory order flow has not matched that framing (Section 8.1). Management never put a number on 2026 memory TCB orders, so this is not a broken promise. It is a gap between tone and order flow, and investors should watch it.

Assessment. On commitments it controls (shipping a tool, closing a divestment, landing a quarter inside its range), this management has done what it said for six straight quarters, and its near-term guidance has become conservative in the upswing. Its weaker area is framing outcomes it does not control, chiefly HBM order timing, more confidently than events have borne out. Execution credibility is high. Strategic timelines deserve a discount, and both the CEO and the strategic review are now in transition.

10. Ownership, Governance and Shareholder Friendliness

Part 1: Ownership and control. ASMPT has one ordinary share class, with about 418 million shares in issue as of May 2026 (HKEX announcement, 5 May 2026). The largest holder is ASM International N.V. (Euronext Amsterdam), the company's former parent. It held 24.56% at the end of Q2 2026, down from 24.65% in Q1 2026 (ASM International Q1 and Q2 2026 results releases, 21 Apr and 28 Jul 2026). The small decline is consistent with new shares issued under employee award schemes rather than sales, but that is an inference, not a disclosed fact. BlackRock, Inc. is a disclosed substantial shareholder above the 5% threshold (HKEX disclosure of interests, per MoatMap). No dual-class shares, golden share or pyramid structure has been identified. ASM International is not a controlling shareholder under the Hong Kong 30% takeover threshold. It is nevertheless the decisive single block, with two board seats. Free float is roughly three-quarters of the shares. For a minority shareholder, this means one strategic holder can strongly influence board-level decisions, such as a sale of SMT, without controlling them. Any change in ASM International's stake would be a material overhang or catalyst.

Part 2: Governance. The board comprises:

• four independent non-executive directors, including the chairman, John Lok Kam Chong;
• two non-executive directors from ASM International: Dr Hichem M'Saad, ASM International's CEO since May 2024, and Paulus Antonius Henricus Verhagen, its CFO since June 2021 and an ASMPT non-executive director since 18 May 2021 (ASM International management board biographies; ASMPT leadership page);
• executive directors, now Bassel Haddad (Group CEO from 11 August 2026) and Guenter Walter Lauber.

This is per the FY2025 results announcement of 4 March 2026 and the CEO announcement of August 2026. Half the board is independent and the chair is independent, which is better than many Hong Kong-listed peers.

The one governance flag is pay design. The May 2026 grant of about 1.52 million award shares to 1,073 employees, about 0.36% of shares in issue, carries no performance targets and no clawback provisions, per the company's HKEX announcement of 5 May 2026. These are broad staff awards, and the scheme predates Hong Kong's revised Chapter 17 share-scheme rules. Even so, equity awards here are not tied to measurable performance. No material related-party transactions with ASM International, no auditor change and no modified audit opinion were identified in the documents reviewed. Apart from the pay design, governance is unremarkable.

Part 3: Capital returns.

Dividends. Dividends per share over the last three full years:

The dividend was halved in the 2024 trough and restored in 2025 (2023 interim and final per the FY2023 results; 2024 per the FY2024 results; 2025 per the FY2025 results, 4 Mar 2026). The 1H 2026 interim was raised to HK$0.97 (results, 29 Jul 2026). Management described a policy of distributing about 50% of profit (Q2 2026 call). The pattern is a payout ratio that moves with earnings, plus specials in good years, rather than a progressive dividend.

Buybacks and dilution. MoatMap's database records no share buybacks in the last ~90 days (since 10 July 2026). For the earlier part of the three-year window (late 2023 to mid-2026), searches of company announcements, results releases and financial news found no general share repurchase programme announced or executed. The FY2025 results contain no buyback. The share count has edged up slightly through new shares issued to satisfy employee awards (for example, 56,650 new shares vesting on 30 April 2026), and some awards are met by trustee purchases of existing shares. The exact net change in shares outstanding over three years could not be verified from the filings reviewed. The evidence points to a broadly flat count with minor dilution.

Verdict: Returns Capital. It pays out about half of earnings plus specials, though through cyclical dividends rather than buybacks. It also holds a net cash balance sheet, boosted by the US$120m NEXX proceeds, that it has so far chosen not to use for repurchases.

11. Insider Activities

In the 12 months to 8 October 2026, MoatMap's HKEX disclosure-of-interests data shows three filings, all by one substantial shareholder and all classified as "Other" rather than open-market purchases or sales:

No price or value is disclosed for any of them. The data shows no director or executive dealings in the window: no buys or sells by the board, the outgoing CEO or the new CEO. BlackRock's filings at this size are typical of a large index and active-fund complex crossing disclosure thresholds through portfolio rebalancing, securities lending or derivative positions. They are not a view expressed by people with inside knowledge. Read: neutral. There is no insider buying signal and no insider selling to explain.

12. Scenarios

Bull case. The C2S bulk order is the first of several as AI accelerators keep growing in size and OSATs add capacity. C2W becomes a second logic leg in 2027 when next-generation GPUs and server CPUs move to chiplets. An HBM maker freezes HBM4E or HBM5 on ASMPT's fluxless TCB through the exclusive evaluation programme, and taller stack-height limits push hybrid bonding into the 2030s. Photonics moves from pluggables into CPO around 2028 with ASMPT supplying the whole assembly flow. The SMT review ends in a clean sale or listing to a credible owner, leaving a focused back-end packaging company with net cash, an independent board and a new CEO with a foundry and packaging background to sell it to Western logic customers. By 2028-2029 ASMPT is seen as a core AI packaging tool supplier rather than a cyclical Asian assembly-equipment maker. Its TCB share reaches the 35-40% target, led by logic.

Base case. Management keeps delivering what it controls, as its six-quarter record suggests. The backlog converts through 2026 and into 2027 as tools ship. Logic TCB grows with C2S repeat orders and early C2W volume. Memory TCB stays lumpy, with occasional bulk orders shared with Korean rivals. Photonics grows on pluggables while CPO arrives slowly. The mainstream and SMT recovery continues into 2027 and then plateaus as OSAT and EMS capex digests. The SMT review takes longer than hoped, and its result may be a partial separation or a decision to retain the business. The new CEO spends his first year on customer relationships and portfolio choices rather than a strategic reset. ASMPT ends up a bigger advanced-packaging business inside a still-cyclical group, with dividends that rise and fall with the cycle.

Bear case. 8.3 and 8.1 combine. AI capex pauses in 2027 just as the C2S bulk order is delivered, so OSATs push out follow-on purchases, while memory makers take HBM4E volume from Hanmi and SEMES. The record order book drains through push-outs instead of follow-on orders. 8.4 adds pressure as Chinese OSATs switch mainstream bonders to domestic suppliers when the cycle turns. 8.5 compounds it: SMT bookings fall from their record, the review stalls without a buyer at an acceptable value, and the segment spends a downturn in limbo. Then 8.2 arrives early. A leading logic customer designs its next package around BESI and Applied Materials' hybrid-bonding flow, cutting into ASMPT's C2W runway. A new CEO (8.6) inherits a smaller opportunity than the TCB market plan promised, and the dividend falls back as it did in 2024.

Sources: ASMPT 2026 Interim Results · Q2 2026 earnings call transcript (Investing.com) · Q2 2026 call highlights (Globe and Mail) · ASMPT Q1 2026 Results · Q1 2026 call highlights (Yahoo Finance) · ASMPT 2025 Annual Results · FY2025 call highlights (Investing.com) · ASMPT Q3 2025 Results · Q3 2025 restructuring (Tiger) · ASMPT 2025 Interim Results · ASMPT Q1 2025 Results · CEO announcement: Bassel Haddad · Semiecosystem on the new CEO · SMT strategic review (TipRanks) · SMT review (Bamboo Works) · NEXX sale to Applied Materials (TipRanks) · 50+ C2S TCB order · ASMPT history · ASMPT (Wikipedia) · ASM International Q2 2026 results · ASM International management board biographies · ASMPT leadership: P.A.H. Verhagen · Employee share award, May 2026 (TipRanks) · Dividend history (StockAnalysis) · SK hynix HBM4 TC bonder order (TrendForce) · Hanmi Griffin HBM4 order (Tech Times) · HBM TC bonder shares (mytenbagger) · BESI, ASMPT, Hanmi and Hanwha (Lumen Alpha) · SemiAnalysis: Scaling the Memory Wall · SEMI equipment forecast (via Dan Nystedt) · ASMPT manufacturing sites (thebuildout.ai) · BESI market cap · Kulicke & Soffa market cap · Hanmi market cap · Hanwha Vision market cap · Fuji Corp market cap · Yamaha Motor market cap · Mycronic market cap