A revolutionary cycle in bonding equipment: HBM's massive capacity expansion and the platformization of hybrid bonding technology open up a second growth curve!

Bonding devices are undergoing a value reassessment driven directly by AI computing power.

The global bonding equipment market size is estimated at approximately US$2.3 billion in 2025, with bonding accounting for 44% of the value of back-end packaging equipment, making it the single largest value category. In the short term, HBM capacity expansion and increased stacking layers make thermocompression bonding (TCB) the most certain equipment segment: ASMPT's new orders in the first half of 2026 reached US$1.631 billion, a year-on-year increase of 85.1%, with the order-to-shipment ratio rising to 1.43; Besi's orders in the second quarter reached €292.9 million, a year-on-year increase of 128.8%, and the number of hybrid bonding customers increased from 15 at the end of 2025 to 21.

In the medium term, the technological inevitability of hybrid bonding (HB) has been established, but its adoption has been delayed by two relaxations of JEDEC's packaging height standards. The key constraint is not technological feasibility, but the economics of single interconnect cost: HBM4 has a pad pitch of 10 micrometers, and switching to hybrid bonding at this pitch offers no cost advantage. Therefore, the mass production window for hybrid bonding on the memory side falls on HBM4E and HBM5, with a steepening of device shipments expected between 2027 and 2028.

TCB will determine the performance realization over the next two years, while hybrid bonding will determine the valuation center over three to five years. The real industry variable is not "whether to adopt hybrid bonding," but rather the equipment platformization driven by hybrid bonding, transforming the bonding machine from a single host into an integrated operating system that combines cleaning, plasma activation, and in-situ metrology, thereby spilling over value to intermediate links such as CMP, metrology, and cleaning. This is also the transmission chain most easily underestimated in this round of industry chain value revaluation.

I. What happened? TCB volume surged.

Bonding equipment: The single equipment category with the highest value in back-end packaging.

According to Besi's statistics, bonding equipment accounts for approximately 44% of the back-end packaging equipment market, making it the largest category, exceeding any individual stage such as etching, deposition, or testing. This structural fact leads to two inferences: first, the transmission of advanced packaging capital expenditure to mid-to-back-end equipment is amplified primarily through bonding; second, as the proportion of advanced packaging in the overall packaging market increases, the incremental growth of bonding equipment is not linear, but rather amplified by both its high value share and its increasing structural share.

Looking at the longer timeline, the bonding equipment itself has also been evolving. From wire bonding in 1975 and flip chip bonding in 1995, to the mass-production thermo-press bonding (TCB) jointly developed by Intel and ASMPT in 2012, and then to the hybrid bonding that took shape in 2018, the common direction of this technology sequence is the continuous improvement of interconnect density and the exponential growth of the number of interconnects per chip.

ASMPT holds 18%, Besi 17%, and K&S 15%, with only a 1-3 percentage point difference between the three leading manufacturers. EVG and SUSS account for 9% and 7% respectively, with the top five combined holding approximately 66%. The remaining 34% is dispersed among Japanese, Korean, and smaller manufacturers. This balance is rare in the equipment industry and requires a distinction between two meanings. In traditional segments like wire bonding and flip bonding, this balance signifies slow market share shifts and intense price competition, making it difficult for newcomers to break through with a single product. However, in the incremental segments of TCB and hybrid bonding, this balance represents an unpriced option —the technological barriers in high-end segments will redistribute market share. Manufacturers that complete customer process validation (POR/AOR approval) first can achieve a market share jump of far more than 1-3 percentage points within a few quarters. The final market share outcome is not determined by existing market share, but by who secures mass production line orders first. Referring to the order structure disclosed in Section 1.4 of this chapter, this market share redistribution is already underway.

The hard evidence from the order book: The financial reports of the two OEMs have already translated the booming business into numbers.

The most reliable way to determine whether the market conditions for a particular equipment are genuine is to look at the OEM's order book to order-to-shipment ratio, rather than looking at roadmaps. In the second quarter of 2026, both Besi and ASMPT released their strongest order data in the past four years.

Besi's second-quarter revenue reached €249.9 million, a year-on-year increase of 68.7% and a quarter-on-quarter increase of 35.2%; new orders totaled €292.9 million, a year-on-year increase of 128.8%. Profitability showed even greater resilience: gross margin was 65.7%, a year-on-year increase of 2.4 percentage points; net profit was €89 million, a year-on-year increase of 177.3%, with net profit margin climbing to 35.6%. First-half cumulative revenue was €434.7 million and net profit was €140.6 million, representing year-on-year increases of 48.8% and 121.1%, respectively. More significantly, the total order value for the past twelve months reached €987.6 million, a record high, and the proportion of AI-related system orders in first-half orders increased from approximately 50% in the same period last year to approximately 60%.

ASMPT's metrics are broader, but its structural information remains equally clear. Second-quarter revenue reached $630 million, a 52.1% year-over-year increase; new orders totaled $903.6 million, a 97.6% year-over-year increase and a 24.8% quarter-over-quarter increase. First-half cumulative revenue reached $1.138 billion, with orders totaling $1.631 billion, representing year-over-year increases of 42.5% and 85.1% respectively. The order-to-shipment ratio reached 1.43, the highest level since the first half of 2021. Advanced packaging revenue reached $339 million, a 17% year-over-year increase, accounting for 30% of the group's revenue; advanced packaging orders in the Semiconductor Solutions segment more than doubled year-over-year. Looking further ahead, ASMPT's TCB business revenue is projected to increase by 146% year-over-year in 2025.

Looking at the signals from both companies side-by-side yields a more precise conclusion: the surge in TCB production is no longer an expectation, but a reality reflected in orders and existing contracts. This is driven by repeat purchases from HBM manufacturers, large orders from advanced logic customers for chip-to-substrate (C2S) and chip-to-wafer (C2W) equipment, and demand for optical interconnect mounting equipment resulting from the expansion of 800G and above optical module production. The real question is what will absorb this second wave of demand, rather than whether the first wave of demand is genuine.

The significance of the number of customers lies in its depiction of the breadth of technology diffusion, rather than just the purchasing depth of a single large customer : 21 customers are distributed across four application categories—logic, memory, co-packaged optics (CPO), and consumer electronics—indicating that hybrid bonding has moved beyond the verification stage of a single scenario. Conversely, the number of customers only increased from 2 to 8 in the first year of growth (2021 to 2023), indicating that the commercial verification period for this technology was as long as three years. Any optimistic assumptions about the adoption pace should respect this historical context.

II. Why is it important? Technological inevitability and cost inflection point

Orders of magnitude of performance advantage: from device specifications to system economics

The technical definition of hybrid bonding is not complicated: it allows chips to be directly bonded together via copper-copper (Cu-Cu) bonding, while simultaneously achieving dielectric layer connections via oxide-oxide bonding, thus eliminating the intermediate structure of microbumps. The effect of eliminating microbumps is cascading—the interconnects are no longer arrays of solder points, but rather arrays of copper pads, with spacing that can be an order of magnitude lower, resulting in a simultaneous decrease in parasitic capacitance and resistance.

Solving the cost paradox: The cost of a single interconnect is only one-tenth that of a TCB.

The most common criticism of hybrid bonding is the high cost of the equipment. This criticism has some basis: hybrid bonding requires entirely new equipment platforms and higher cleanroom standards, and the initial capital expenditure does indeed far exceed that of traditional TCB solutions. However, using initial investment as the basis for decision-making will lead to incorrect conclusions in scenarios where the number of stacked layers is rapidly increasing, because what truly determines the economics of mass production is the cost per interconnect, not the equipment purchase price.

The core investment logic of hybrid bonding is: high initial capital expenditure, but low unit cost, with a cost curve that intersects as the number of stacked layers increases . Data from Besi shows that the cost of a single interconnect in hybrid bonding is only about 1/10 of that of the TCB solution. The mechanism is not mysterious: the TCB solution relies on microbumps, and as the number of stacked layers increases from 8 to 16 or even 20, the number of bumps grows exponentially, and each layer requires independent flux treatment, reflow, and cleaning processes; hybrid bonding's direct copper-copper bonding keeps the marginal cost of interconnects very low, and economies of scale become apparent.

So why hasn't there been mass production yet? The constraint isn't technology, but rather the economics of maintaining a 10-micron pad pitch.

If the performance and cost advantages are both valid, why has the adoption of hybrid bonding on the memory side been delayed? The answer lies in a JEDEC standard revision. In January 2026, JEDEC raised the upper limit of HBM package height from 720 micrometers to 775 micrometers. The extra space allows 16-layer stacked HBM4 to be assembled using mature microbump technology, without having to switch to hybrid bonding. According to TrendForce, JEDEC is considering further raising the upper limit to approximately 900 micrometers.

It postponed the introduction of hybrid bonding from HBM4 to HBM4E and HBM5, a delay of about one to two years. The pricing reason is the pad pitch. HBM4 has a pad pitch of 10 micrometers, at which hybrid bonding does not offer a cost advantage—the economics of hybrid bonding require even smaller pitches. SK Hynix's plans confirm this: the company continues to use advanced monolithic reflow and molding underfill processes on 16-layer HBM4, reserving hybrid bonding as a backup solution, while continuing to validate 12-layer hybrid bonding samples. The 16-layer HBM4 sample showcased by the company at CES 2026 did not use the previously expected fully hybrid bonding process.

The division of labor between the two process routes: W2W provides the basic framework, and D2W determines the flexibility.

Hybrid bonding has two engineering approaches. Wafer-to-wafer (W2W) involves bonding two complete patterned wafers face-to-face before dicing. Alignment is done only once at the wafer level, allowing for the narrowest pitch and highest production efficiency. The trade-off is that both wafers must contain chips of the same size, and defective chips cannot be removed, limiting yield. Die-to-wafer (D2W) involves flipping and aligning a pre-diced and tested single chip onto a target wafer. Known good chips (KGD) can be selected, supporting the hybrid integration of chips of different sizes and process nodes. The trade-off is that processing is done one chip at a time, resulting in lower output efficiency.

The application division between the two approaches is relatively clear. W2W is the foundation of mature applications, with Sony's three-layer stacked CIS (pixel layer-DRAM-logic layer) and 3D NAND being the main scenarios. As the number of layers in 3D NAND increases, the difficulty of etching from the top layer to the bottom layer in one go rises rapidly. It is necessary to use W2W to bond two NAND chips with fewer layers to form a multi-layer structure. Yangtze Memory's Xtacking architecture and Kioxia's CBA structure are examples of this. D2W corresponds to the most flexible incremental scenarios, including HBM stacking, 3D Logic (such as TSMC's SoIC), and Chiplet heterogeneous integration.

A panoramic view of the industry chain giants: the roadmap has converged, and the only difference lies in the timetable.

The most effective way to assess the industry consensus on a technology is to compare the publicly available roadmaps of major players. If the technological directions of each company are consistent but their timelines are scattered, it indicates that the technology has been finalized and the pace of its adoption is uncertain—this is precisely where hybrid bonding currently stands.

The three giants of storage: aligned in direction, but diverging in pace.

Samsung is the most aggressive of the three memory manufacturers in investing in hybrid bonding. The company is building a D2W hybrid bonding mass production line at its P5 plant in Pyeongtaek, South Korea, selecting Besi as its primary partner and introducing 50 D2W hybrid bonding machines (Model Datacon 8800 CHAMEO ultra plus AC), requiring structural customization of the standard equipment. Its subsidiary, SEMES, has also received orders to supply D2W hybrid bonding machines for its mass production line, planning to build two hybrid bonding mass production lines in Pyeongtaek. Hanwha Semiconductor is also on its shortlist of hybrid bonding machine suppliers. Technically, Samsung plans to apply hybrid bonding starting with 16-layer stacking of HBM4E to address heat dissipation issues, and at GTC 2026, it presented data showing that hybrid bonding can reduce thermal resistance by more than 20% compared to thermocompression bonding. Samsung is also advancing on the NAND side, planning to mass-produce V10 NAND using hybrid bonding technology in 2026.

SK Hynix's strategy is more conservative. The company insists on using advanced monolithic reflow and molding underfill processes for its 16-layer HBM4, reserving hybrid bonding as a backup solution, while continuing to validate 12-layer hybrid bonding samples; the 16-layer HBM4 samples showcased at CES 2026 did not employ a fully hybrid bonding process. Regarding equipment procurement, the company ordered mass-production hybrid bonding equipment in March 2026, expecting to apply it starting with next-generation HBM products. On the capacity side, SK Hynix is investing $3.87 billion in Indiana to build an advanced packaging plant, aiming for production to begin in 2028.

Micron is in the validation phase, researching hybrid bonding technology for HBM4, and has already started construction on a $7 billion HBM advanced packaging plant in Singapore, expected to be operational around 2027. The new capacity of both memory plants is designed for the ultimate volume required for hybrid bonding, with equipment commitments already ahead of the technology confirmation and mass production timeline—a typical example of industrial capital prioritizing technological development, and also constituting an implicit order backlog for equipment manufacturers.

Foundry and Logic Sides: Hybrid bonding has completed mass production verification

Unlike the waiting on the storage side, mass production on the logic side has already occurred, which explains why the customer structure for equipment orders is mainly logic manufacturers.

TSMC's SoIC platform leads in production volume. At the 2026 North American Technology Symposium, the company presented a pitch roadmap of 9 microns in 2023, 6 microns in 2025, and 4.5 microns in 2029. The second-generation SoIC adds face-to-face bonding to the back-to-back stacking supported by the first generation. The node stacking roadmap is progressing simultaneously, moving from the current N3P-on-N4 to N2P-on-N2P in 2028 and A14-on-A14 in 2029. On the capacity side, the company is building the Chiayi AP7 campus as its largest advanced packaging base, aiming to start production in 2026. Luke Lin, a senior executive at TSMC responsible for advanced packaging, made two important predictions: First, HBM5 may generate demand for hybrid bonding earlier than the market expects. HBM5 is expected to achieve mass production around 2028, and the demand for hybrid bonding equipment and supporting materials will appear as early as 2027 to 2028. Second, the company's SoIC capacity expansion in 2026 will be lower than initially planned, but it will return to an aggressive expansion pace from the end of 2027 to 2028. This is because the early customer implementation progress of CPO is not as expected, and the maintenance and warranty processes still need to match the actual needs of customers.

III. What should we focus on next? When will the volume increase and value spillover occur?

Timeline: Three time windows, three asset attributes

By dividing equipment demand by time, we can obtain three windows with different characteristics, corresponding to three different asset attributes.

The first window (2026-2027): TCB and traditional processes dominate. Due to JEDEC's relaxation of height standards, storage vendors still tend to use traditional hot-press bonding during this phase, while conducting early verification and small-scale orders for HBM4E. Equipment revenue is mainly supported by repeat purchases from HBM vendors and large C2S/C2W orders from advanced logic customers. The asset attribute of this window is "performance realization"—orders are already in hand, and the key variable is the delivery and acceptance schedule.

The second window (around 2028): the transition period between Fluxless TCB and hybrid bonding. As interconnect pitches shrink further and package layers increase, the difficulty of completely cleaning flux at pitches below 40 micrometers rises rapidly. Incomplete cleaning can lead to decreased yield and long-term reliability issues. During this phase, fluxless thermocompression bonding (Fluxless TCB) becomes a transitional solution—it continues the thermocompression framework of TCB, using only formic acid vapor or plasma cleaning instead of flux. Customer adoption costs and process migration difficulties are lower than a complete switch to hybrid bonding. Simultaneously, D2W hybrid bonding begins to be used for connections between the base die of HBM4E and custom HBM (cHBM) and DRAM. The asset attribute of this window is "order switching"—demand shifts from traditional TCB to Fluxless TCB and HBM, with manufacturers with complete product lines benefiting more.

The third window (2028 to 2030 and beyond): Hybrid bonding enters large-scale production. The 20-layer stacking of HBM5 and the high-density interconnect requirements of 3D Logic will jointly drive the mass production of D2W hybrid bonding. Initially, it is expected to use a combination of TCB and hybrid bonding, followed by a rapid increase in the penetration rate of hybrid bonding in high-end bonding. The asset attribute of this window is "valuation reassessment"—the revenue structure is shifting from cyclical equipment to platform-based systems.

According to Yole Group's forecast, the W2W hybrid bonding equipment market will reach approximately $510 million in 2027, while D2W will reach approximately $230 million, totaling approximately $740 million. W2W will account for 68.9% and D2W for 31.1%. However, the growth rates are inversely proportional to the scale: W2W's CAGR is approximately 16%, while D2W's is approximately 69% . This combination of "large and slow" versus "small and fast" implies different investment implications in the equipment industry. W2W applications are concentrated in CIS and 3D NAND, where demand is stable but the market structure is solidified—EVG holds approximately 82% of the market share in this segment, making it difficult for newcomers to enter. D2W, corresponding to HBM, 3D Logic, and Chiplet, represents the largest growth scenario over the next five years, and market share is still being redistributed. Based on D2W's $230 million base in 2027, if the 69% CAGR continues, its size will surpass W2W's around 2030.

Long-term incremental growth: CPO and optoelectronic integration

Beyond HBM and logic, there is an even further but faster-growing demand curve for hybrid bonding – co-packaged optics (CPO).

As AI clusters evolve from tens of thousands of cards to hundreds of thousands of cards, traditional copper interconnects and pluggable optical modules are approaching their physical limits. Some CPO solutions require hybrid bonding to achieve high-density integration of photonic and electronic integrated circuits. TSMC's COUPE platform leads in this field: compared to traditional copper traces, the COUPE solution achieves 4x power efficiency and 10x latency reduction at the substrate level; when further integrated to the silicon interposer, performance is improved to 10x power efficiency and 20x latency reduction. On the customer side, NVIDIA's Spectrum-X and Quantum-X photonic switches and Broadcom's Tomahawk CPO platform have already been adopted; optical component suppliers such as Coherent and Marvell are adjusting their product portfolios to meet CPO requirements.

Core judgment

Judgment 1: The increased volume of TCB is a confirmed fact, while the increased volume of hybrid bonding is a point in time yet to be confirmed. The two are different in nature and should not be confused. Besi and ASMPT both reported their strongest order data in the past four years in Q2 2026, with order-to-shipment ratios and order backlogs providing solid evidence. Hybrid bonding, on the other hand, is still in the customer validation and production line construction phase, with its revenue contribution expected to steepen between 2028 and 2030. Pricing hybrid bonding as a performance driver for 2026-2027 is inconsistent with the order structure observed in this report.

Judgment 2: The adoption of hybrid bonding is irreversible, but the timing is determined by the cost curve rather than technological capabilities. JEDEC has twice relaxed the upper limit of HBM height (from 720 microns to 775 microns, and is considering increasing it to approximately 900 microns), buying time for micro-bump solutions at the physical level; with HBM4's 10-micron pad pitch, hybrid bonding does not yet have a cost advantage. When these two constraints will be lifted depends on when the number of stacked layers exceeds 16 and moves towards 20 layers, and whether the upper limit of package height will continue to be relaxed.

Judgment 3: The real industry variable is platformization, not the replacement of individual devices. Hybrid bonding is pushing equipment from single mainframes to integrated systems that combine cleaning, plasma activation, and in-situ metrology (Applied Materials and Besi's Kinex, SUSS's XBC300 Gen2 all embody this form). This change extends the value of the bonding process to intermediate processes such as CMP, cleaning, and metrology. Therefore, the scope of beneficiaries in the industry chain is significantly broader than just the bonding machine manufacturers themselves.

Judgment 4: The competitors are not just each other; Fluxless TCB is an alternative to hybrid bonding within a specific timeframe. K&S has achieved mass production of fluxless copper-copper TCB and expanded its APTURA platform from advanced logic to memory; ASMPT has also secured orders for Fluxless TCB for next-generation HBM. This technology continues the hot-pressing framework of TCB, with lower customer adoption costs and migration difficulties compared to a complete switch, thus possessing realistic competitiveness within the timeframe where hybrid bonding is not yet mature. Any predictions of hybrid bonding penetration rates should be compared with the maturity of Fluxless TCB, not with traditional fluxed TCB.

Risk warning and disclaimerInvesting involves risk; please exercise caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Any investment decisions made based on this information are at your own risk.