The Boring Back-End of AI.Original analysisNot investment advice

The semiconductor industry loves glamorous bottlenecks. EUV. HBM. CoWoS. Blackwell. 2nm nanosheets. High-NA. The chips that get written about are the ones with the smallest features and the largest valuations. But chips do not become useful just because transistors exist. The die has to be connected, packaged, powered, cooled, tested, and turned into a reliable product. That is the back-end. And that is where Kulicke & Soffa lives.

K&S is not Nvidia, TSMC, or ASML. That is exactly why it is interesting. It sits in the invisible layer where chips become systems. In the AI era, that layer matters more than the marketing makes obvious. The question this essay asks is not whether K&S is a good stock. The question is what the company tells us about a part of the chip economy most people stop reading about after the headline.

AI is pushing the industry to scale sideways and vertically, not just shrink transistors.

---

1. The lazy view of K&S

The shallow market view of Kulicke & Soffa goes like this. K&S is an old wire-bonder company. Wire bonding is old packaging. Advanced packaging is the future. Therefore wire-bonder companies are structurally dead. The trade is short the bonders and long whoever makes CoWoS look impressive.

It is not a stupid story. Wire bonding really is old. Advanced packaging really is the leading edge. The lazy version is just missing two things. First, wire bonding is not disappearing in a hurry. It is still the high-volume default for mature logic, analog, power semiconductors, automotive electronics, industrial chips, sensors, memory at certain tiers, LEDs, and lower-cost packages. Second, K&S itself has been working for a decade to climb out of the “wire bonders only” box.

The better question is not whether wire bonding disappears. The better question is where wire bonding remains good enough, where it evolves, and where K&S can climb into higher-value packaging layers.

---

2. What the 2021 thesis got right

The starting point for this essay is the uploaded 2021 SemiAnalysis piece on K&S.¹ It is a useful historical anchor, not a final verdict. The piece treated K&S as a misunderstood compounder hiding inside the “old wire bonders” framing, and made an argument that is worth restating in its own words before grading it.

The 2021 piece was not simply “wire bonding is back.” The better version was: the market was underestimating the entire back-end. That framing has aged well. The specific claim that wire-bonding tightness would persist for years was right for a while and then faded as the cycle turned. The bigger claim that K&S would matter as a back-end story, not just a wire-bonder story, has lasted longer.

No SemiAnalysis text, charts, screenshots, or images are reproduced here. The original piece is treated only as cited historical framing.

---

3. What changed by 2026

Five years on, the K&S numbers tell a more layered story than either the bull or bear version of 2021. K&S’s own FY2025 10-K filing shows the picture of a company that is still cyclical, still exposed to ball bonding, but no longer only that.⁴

FY2025 revenue was USD 654.1 million, down 7.4% year over year. Ball Bonding Equipment revenue fell 18.1%. Wedge Bonding Equipment grew 4.5%. Advanced Solutions revenue grew 37.6%. APS (the aftermarket products and services segment) represented 23.9% of total revenue. The story is a soft year on legacy bonders and double-digit growth in advanced and aftermarket. That is the shape of a transition, not a collapse.

The more recent print sharpens the point. K&S Q2 FY2026 revenue came in at USD 242.6 million with gross margin of 49.3%, and the company guided Q3 FY2026 revenue at about USD 310 million plus or minus USD 20 million.² K&S also raised FY2026 capex from about USD 12 million to about USD 22 million to expand TCB production, and said that capex supports up to roughly USD 400 million in annual TCB system sales over time. None of that is a guarantee. It is a company making a capital decision about which technology it expects to scale.

The updated 2026 thesis is not that wire bonding magically became sexy. It is that the old back-end company is trying to bridge old packaging and new packaging at the same time, while the rest of the industry catches up to how important that bridge is.

---

4. Why AI makes packaging more important

The macro context is the part most readers already know, but it matters to spell out. AI is not only a transistor problem. AI systems need more compute, more memory, more bandwidth, more power delivery, and more thermal management than the previous generation of workloads. That pushes the industry toward chiplets, HBM, interposers, CoWoS, SoIC, advanced substrates, larger packages, and more complex bonding.

TSMC’s own 2026 framing makes this concrete. TSMC said it is in production with CoWoS at about 5.5 reticles of area. It said it is planning 14-reticle-size CoWoS by 2028, capable of integrating roughly 10 large compute dies and 20 HBM stacks in one package.⁵ Those are TSMC company claims, not independent benchmarks. The direction is what matters. The leading edge is no longer scaling only inside a die. It is scaling outward, on the package.

ASML’s Q4 FY2025 deck adds the other half of the picture. ASML said AI compute demand has outpaced Moore’s Law, and that Moore’s Law alone is not sufficient to meet future training compute requirements. It said AI demand for compute density and high-bandwidth memory is growing faster than overall semiconductor sales.⁶ Read together with TSMC’s roadmap, the message is that the industry is still scaling transistors, and it is also scaling sideways and vertically.

That is the climate K&S lives in. The more the package becomes the scaling surface, the more bonding, interconnect, and assembly matter as decisions in their own right.

---

5. The K&S memory packaging update

In March 2026, K&S formalised its memory-packaging story. The company said it is structuring its memory interconnect roadmap around Ball Bonding, Vertical Wire, Advanced TCB, and future Hybrid Bonding, and said TCB demand should grow approximately 70% sequentially in FY2026, with continued growth in logic and memory advanced packaging.³ That is a company claim, not an external forecast. Read it as direction.

Each of those four technologies plays a different role. Ball bonding is the traditional high-volume interconnect method, still dominant in many legacy and mature-node packages. Wedge bonding, which K&S also sells, is the heavier-current cousin used in power modules, automotive, and industrial. Vertical wire arranges wire in a more vertical, dense pattern, useful for stacked memory and higher-density interconnect than classic ball bonding can give you. TCB, thermo-compression bonding, uses heat and pressure to bond dies, and is one of the workhorse processes inside today’s advanced packages and heterogeneous integration. Hybrid bonding goes further: direct copper-to-copper, wafer or die, no bumps. It is the bet for future high-density 3D packaging.

None of this means K&S will dominate hybrid bonding. The honest read is that K&S is trying to be present across the bonding spectrum at the moment the package becomes the scaling surface. Competition is intense. Besi, ASM Pacific, and Applied Materials all want share in TCB and hybrid bonding, and each has its own franchise to defend. The 2026 update is that K&S has earned a seat at this table, not that it has won the table.

The better question is not whether wire bonding disappears. The better question is where it remains good enough.

---

6. Mini and micro LED: the part that aged unevenly

The 2021 piece was excited about PIXALUX and LUMINEX because mini LED and micro LED manufacturing are brutal placement problems. A typical iPad Pro display in the mini LED era used over 10,000 mini LEDs. Even a tiny defect rate hurts panel yield, which means placement tools that combine throughput with sub-micron-class precision become strategic.¹

The technology turned out to be real. K&S said in 2023 that LUMINEX with TSMT had reached 540,000 units per hour at 15µm 3-sigma accuracy, and TSMT said it was working to improve yield from 99.99% to 99.999%.⁷ Those are company numbers, but they describe a tool category that is no longer hypothetical.

The market thesis aged less cleanly. Apple shifted its iPad Pro to OLED, which weakened the “Apple drives mini LED tablets forever” case. The 2026 reframing is that the LED opportunity is broader than tablets. TVs, gaming monitors, automotive displays, signage, direct-emissive consumer displays, and eventually wider micro LED adoption are still on the table. Micro LED itself is attractive on paper because it offers brightness, efficiency, durability, and contrast advantages. The reason it has not won yet is manufacturing: transfer, placement, bonding, yield, repair, and cost. That is exactly the place back-end process equipment matters most.

The honest 2026 read is that LUMINEX is real, mini LED is a real volume opportunity in non-tablet displays, and micro LED is a longer-dated bet with serious yield problems. None of those are guaranteed to drive K&S earnings on any particular timeline.

---

7. Battery and power electronics as secondary optionality

The 2021 piece also flagged battery and power applications. This is where I would temper the enthusiasm most. The stronger argument is not that K&S becomes a battery company. The stronger argument is that electrification creates more places where precise, reliable interconnects matter.

K&S’s wedge and ribbon bonding expertise can apply to power modules, automotive sensor packages, DBC (direct bonded copper) substrates, and small cylindrical battery packs. Electrification raises demand for power semiconductor packaging and high-current interconnects. That is a real adjacency. It is not the strongest pillar of the 2026 thesis. K&S’s public reporting does not yet break out battery exposure cleanly, and treating it as a major driver without segment-level proof would be an overreach.

Battery is best read as optionality, not as a primary line of the argument.

---

8. Mature nodes still matter

The other reason the back-end stays relevant is that the chip economy is not one market. Mature logic, analog, power and discrete, automotive, and industrial chips are still enormous. The OECD’s 2025 chip landscape report says China leads in mature logic in-production capacity at about 4.23 million 8-inch-equivalent wafer starts per month, and also leads in power and discrete and in analog capacity.⁸ Those are not leading-edge AI accelerators. They are MCUs, automotive ICs, power devices, sensors, and the chips that run everything else.

SEMI’s 2026 outlook adds another data point. SEMI projects double-digit growth in global 300mm fab equipment spending in 2026 and 2027, driven by AI investment and regional semiconductor self-sufficiency.⁹ Some of that spending sits at the leading edge. A meaningful share sits at mature and trailing-edge nodes built to anchor domestic supply chains. Both ends of the stack need back-end assembly, and the mature end is still where wire bonding does the bulk of the work.

The chip economy is a stack. Leading-edge AI accelerators sit at one end. Mature-node power, analog, MCU, sensor and automotive chips sit at the other. Both need back-end assembly. K&S is unusually positioned to sit on both ends at once.

---

9. The K&S exposure map

If you want to compress the company into one picture, this is roughly what it looks like in 2026. Seven exposures, each with a different posture and a different risk profile. None of them is the whole story on its own.

That eighth tile is the “batteries” one, kept neutral on purpose. It is in the map because the 2021 piece flagged it, not because it is a load-bearing part of the 2026 thesis.

---

10. Risks and why this is not a simple bullish story

None of the above implies K&S is risk-free, and the company itself is explicit about that in its filings.⁴ A serious back-end thesis has to carry a serious risk list.

The point is not that K&S is risk-free. The point is that the back-end layer is strategically more important than most people realise, and a company that sits across it deserves to be analysed on more than the wire-bonder caricature.

---

11. Evidence ledger

Because this essay sits between the uploaded 2021 thesis and a 2026 reading, here is the load-bearing evidence in one place, paired with source and a one-line note on why it matters.

---

12. Final verdict

The 2021 thesis was right in two ways. It was right that K&S was more than an old wire-bonder company. It was right that back-end assembly was becoming more important than the market was pricing in.

The 2026 version should be sharper. K&S is a case study in the forgotten infrastructure layer beneath AI. The leading-edge story focuses on the transistor and the package. The mature-node story focuses on power, analog, automotive, and industrial chips. Both ends of that stack lean on the back-end. K&S sits in the middle of it.

Whether the company is a good investment is not the point of this essay. The point is that the layer where chips become systems is strategically more important than the headline coverage suggests, and the company that sells equipment, consumables, and services into that layer is interesting on its own terms.

The front-end makes the transistor. The back-end makes it usable. And in the AI era, that boring back-end is becoming harder to ignore.

---

Sources & footnotes

This essay treats the uploaded 2021 SemiAnalysis article as a cited historical anchor. Specific 2026 claims come from K&S investor releases and the FY2025 10-K. Macro framing is sourced from TSMC, ASML, OECD, and SEMI public material. No third-party charts, screenshots, or logos are reproduced. This is not investment advice.