At a glance. ATM advanced packaging is now running the group: LEAP 2026 is guided above the old US$3.5bn and to double in 2027, and CapEx is US$10.5bn. Jason Chang's CoWoS sentence is the honest one for an OSAT: keep raising CoWoS-related capacity, and if the customer brings EMIB or another substrate, assemble that too — "not a zero-sum game." ASE does not run TSMC's CoWoS brand. The call never said 800 V.
What changed this quarter
The raise was the third of the year. Full-year CapEx goes to US$10.5bn (+US$2bn: US$1bn facilities, US$1bn equipment), a record. Most of the add is LEAP; mainstream advanced packaging and test are being expanded too.
LEAP was sized twice. 2026 revenue will beat US$3.5bn, with room for several hundred million dollars more. 2027 is a double versus 2026. Demand is "not the risk"; execution — buildings, tools, yield — is.
ATM margin is being allowed to break its own cap. Q2 ATM gross margin 27.3%; Q3 28–29%; Q4 has a chance to go through the 30% structural ceiling, after which the long-term band would be reviewed up, not down.
The numbers
| Q2 2026 | Value | Change |
|---|---|---|
| Group revenue | NT$191.06bn (~US$6.07bn) | +10% QoQ, +27% YoY |
| Gross margin | 21% | +1pp QoQ, +4pp YoY |
| Operating margin | 11.1% | +1pp QoQ, +6.8pp YoY |
| Net income | NT$21.07bn (~US$669m) | +49% QoQ, +180% YoY |
| EPS | NT$4.80 (~US$0.15) | second-highest quarter; highest excluding a 2021Q4 asset sale |
| ATM revenue | NT$126.15bn (~US$4.00bn) | +12% QoQ, +36% YoY |
| ATM gross margin | 27.3% | +1.3pp QoQ, +5.4pp YoY |
First half 2026: revenue NT$364.73bn (~US$11.6bn), +22%; gross margin 20.6%; operating margin 10.6%; net income NT$35.2bn (~US$1.12bn), +134%; EPS NT$8.03.
ATM application mix: communications 41%, computing 30%, automotive / consumer / other 29%. Q2 equipment CapEx US$1.695bn; first-half equipment US$2.698bn.
Q3 guidance assumes NT$31.9 per dollar: group revenue +21% to +22%; group gross margin 20.5–21.5%; operating margin 11.5–12.5%. ATM revenue +11% to +13%, ATM gross margin 28–29%. EMS revenue +~40%, EMS operating margin 3.2–3.4%.
NVIDIA 800V read-through
ASE is not on NVIDIA's 800 VDC list and this call never said 800 V. The read-through is the package under the GPU, not the rail.
Hold the OSAT boundary:
- Foundry versus OSAT. The most advanced wafer-level steps, customer IP and system design stay with the foundry. Other packaging, test and full-process can be split on technology, efficiency and what the customer wants. ASE has no interest-conflict with the foundry or the substrate houses, so it can buy the substrate the customer chooses.
- CoWoS and EMIB are not substitutes in ASE's mouth. First job is still to raise CoWoS-related capacity, scale and efficiency. If the customer picks EMIB or something else, that goes on the roadmap. Pure-play OSAT can purchase the substrate and do the back-end. "The world is large."
- Glass is not 2026–27 volume. In evaluation with substrate vendors, the foundry and customers. No production in the next 12 months.
- Panel-level 310×310 mm, fully automated, production 2027 Q1. Complementary to the foundry's approach; same class of customer; similar reticle and line/space. Who wins is capacity, performance and time-to-volume.
- CPO small volume from end-2026; revenue and 2027 outlook in about two more quarters. In-house work for about 20 years.
Full-process (turnkey) is guided about US$300m (~NT$9.6bn) in 2026 — real, still small versus the group, and not yet in the margin. CPU, GPU and ASIC are in it; names in one to two quarters.
Segments that moved
ATM is the AI engine and the margin engine. Q2 utilisation about 80–85%; existing tools near full, so sequential growth is gated by install and buildings. Q3 ATM +11–13% means roughly that much more effective capacity has to come in. General (non-LEAP) ATM growth was raised from 13% to 20% for the year; overall ATM +35%. General demand is industrial, power, interconnect and storage into EV and AI data centers.
LEAP and test are the mix that can take ATM gross margin through 30%. Equipment CapEx mix: about 56% packaging, 40% test, remainder EMS and materials; about 70% of tool capacity is leading-edge.
EMS Q3 +40% is partly seasonal and partly memory and other component prices. That inflation is why the 3.2–3.4% operating-margin guide sits below a "normal" 3.7–3.8%. Q4 EMS revenue is guided similar to Q3; full-year EMS growth below 20%.
Thirteen new-fab projects and eight acquisitions / retrofits are sized through 2028, some into 2029. Free cash flow stays negative while that spends; the balance sheet is described as still healthy, with several funding channels.
Guidance
- Q3 2026: group revenue +21–22% sequentially at NT$31.9 / US$1; ATM +11–13%, ATM GM 28–29%; EMS +~40%, EMS OP 3.2–3.4%.
- Full year 2026: LEAP above US$3.5bn (plus several hundred million possible); general ATM +20%; overall ATM +35%; EMS below +20%; CapEx US$10.5bn.
- 2027: LEAP revenue double 2026; full-process "significant" growth; panel-level production from Q1.
- ATM gross margin: chance to break 30% in Q4, then revisit the structural band upward.
- US / geography: process development and high-volume manufacturing stay Taiwan-first; US sites are design, test-development and packaging-architecture engineering. Replication abroad "when the time is right."
Management Q&A
Q: LEAP will beat US$3.5bn this year — why is a 2027 double credible?
A: Progress is already ahead of the old US$3.5bn, with room for several hundred million dollars more in 2026. Customer demand, buildings and tool-move-in are in hand. Demand is not the risk; whether ASE can install, yield and ramp on plan is. Yield and execution so far are why the year was raised. The next 12 months of factories and tools are what have to support 2027. "Business demand is not what we worry about. The key is our own execution."
Q: Where does the US$10.5bn CapEx go?
A: Another US$1bn of buildings and US$1bn of tools on top of the prior plan. Of tools, about 56% packaging, 40% test; about 70% leading-edge. The add is mostly LEAP, but mainstream advanced packaging and test are growing too, so more than one product line is being expanded. Thirteen new fabs and eight buy-and-retrofit sites cover demand through 2028, some to 2029. Free cash flow will stay negative for a period; funding channels are in place.
Q: When can ATM gross margin break 30%?
A: Q3 is guided 28–29%. As LEAP and test scale, ATM may go through the 30% structural cap. If it does, the long-term band will be reviewed up. Mix, automation, utilisation, scale and past investment coming through all sit in that, not only LEAP share. Pricing is still described as relatively favourable; materials inflation can be discussed with customers against supply, demand and return on the new kit.
Q: How tight is packaging and test, and why was general business raised too?
A: Q2 utilisation about 80–85%. New tools aside, most installed capacity is near full, so near-term growth is install- and building-limited. Q3 ATM +11–13% is also the extra effective capacity ASE has to bring in. LEAP is tight and general packaging and test customers are still asking for more. General full-year growth goes from 13% to 20%; overall ATM +35%. General demand is industrial, power, interconnect and storage into EV and AI data centers.
Q: What is the status of full-process, panel-level packaging and glass substrate?
A: Full-process is about US$300m in 2026, on plan, still too small to move group profit, guided to grow clearly in 2027. CPU, GPU and ASIC are in the mix; a finer customer split in one to two quarters. Panel-level is a 310×310 mm fully automated line, production 2027 Q1, complementary to the foundry, similar customers and similar reticle / line-space — the pick is capacity, performance and time-to-volume. Glass is in evaluation; no production in the next 12 months.
Q: When does CPO start to ship?
A: Small volume from end-2026, when bandwidth, system performance, thermal, yield and cost data will exist to judge the ramp. Revenue size and 2027 outlook are not being given now; more complete numbers in about two quarters. Electrical-optical integration is the direction; difficulty rises as it gets closer to the die and the substrate. ASE has been in the work about 20 years. "We have already done this for about 20 years; waiting another two quarters should be fine."
Q: How does ASE divide work with the foundry, including CoWoS versus EMIB?
A: ASE is a pure-play OSAT. No direct conflict with foundries or substrate houses, so it can work with whatever technology the customer brings. The most advanced wafer-level, IP and system-design steps stay with the foundry; other packaging, test and full-process can be split. Alternative backends that deliver yield, performance and efficiency are welcome — they relieve a capacity bottleneck. ASE will keep adding CoWoS-related capacity, and will put EMIB or another scheme on the roadmap if that is what the customer uses. "This is not a zero-sum game, and it is not that only one technology can be chosen. The world is large."
Q: Why is Q3 EMS revenue up about 40%?
A: Seasonal demand plus higher memory and other component prices, so the revenue jump is not all unit growth. That is why EMS operating margin is guided 3.2–3.4%, below a normal 3.7–3.8%. Q4 EMS revenue similar to Q3; full-year EMS growth below 20%.
Q: What about the US build-out versus Taiwan manufacturing?
A: Process development and high-volume, automated lines are being built in Taiwan with major customers. Replication to the US or elsewhere comes when that is ready. US sites (Fremont, San Jose, plus third and fourth buildings) are chip-design support, test development, R&D and packaging-architecture engineering — not the mass-production core.
Disclaimer
English notes on ASE Technology Holding's investor conference of 30 July 2026, covering Q2 2026, from contemporaneous quotes of management and the company's reported figures. Figures and statements are as presented by management, are unaudited, and have not been independently verified. USD equivalents are approximate at about NT$31.5 to the dollar except where ASE guided on its own FX assumption of NT$31.9. For reference only; not investment advice.