At a glance. Nine-month sales were up 52%, with thermal products up 77% to 73% of the company. GPU liquid-cooling modules for AI servers started shipping in 2025 and were given a full-marks 2026 outlook. The technical story is a ten-year microchannel programme that removes TIM2 — and that revenue is not in this year's scores. The call never said 800 V.
What changed this quarter
Thermal became the company. In the first nine months of 2025, thermal was 73% of sales, +77% YoY. Vapor chambers / lids were +82% on advanced packaging and AI servers. Liquid-cooling modules were +48%, the first year of AI-server GPU water product.
Q3 margin was called the trough. Gross margin fell from a Q1 peak of 44.91% to 38.97%, on a stronger Taiwan dollar and a mix shift away from high-margin lids toward lower-margin consumer parts. Management said Q3 should be the bottom.
Capacity in Taoyuan is being doubled. Dayuan Plant 1 expansion is due in Q1 2026 (+4,735 ping). Dayuan Plant 3 at the Aerotropolis is due in Q1 2027 (+11,340 ping). Together 16,075 ping, described as a doubling of the Taiwan footprint.
The numbers
| Q3 2025 | Value | Change |
|---|---|---|
| Revenue | NT$5.07bn (~US$161m) | — |
| Gross profit | NT$1.976bn (~US$63m) | — |
| Gross margin | 38.97% | down from Q1 44.91% |
| Operating margin | 29.28% | — |
| Net income | NT$1.367bn (~US$43m) | — |
| EPS | NT$9.49 (~US$0.30) | — |
Nine-month sales +52% YoY. Expense ratio down to about 9% even as absolute opex rose, because sales grew faster. Receivables NT$5.2bn (~US$165m), mostly longer-tenor Taiwan packaging-house accounts, all said to be inside normal terms.
Convertible bonds issued in 2025 have mostly converted; the remaining NT$600m (~US$19m) is expected to convert by year-end, which will enlarge the share count.
Capex: NT$900m in 2025, NT$1.2bn in each of 2026 and 2027 (~US$29m then ~US$38m), split between tools and buildings. Cash is described as enough; no raise is planned.
NVIDIA 800V read-through
No 800 V, no Kyber, no sidecar power rack. Jentech is a cooling-and-mechanicals name on the same AI cabinet.
What the call actually adds:
- GPU water-cooling modules are shipping. That is 2025 revenue, not a 2027 slide. The 2026 outlook for the line is 5/5 because a platform of GPU and surrounding-loop products, at several customers, is going to production — not because of one SKU.
- Microchannel liquid cooling is the next-gen claim: ten years of work, assembled at the advanced-packaging house, removing TIM2. Simulated and packaged-chip data are said to match. The numbers used on stage were 3,000 W to 7,000 W dies held around 70°C, module height about 22 mm (0.5U). Customer timing was refused — it follows the next chip, not Jentech's calendar. Importantly, none of that potential revenue is in the 2025 outlook scores.
- Vertical integration (tooling, stamp, plate, assemble) is the stated reason a US$6,000 AI chip buyer would rather have a US$10 lid that yields than a cheaper one that does not. No customer has asked Jentech to share the design or patents.
Read it next to Auras and AVC: those two sell the tray loop. Jentech is trying to sit on the package, then sell the loop as well. Until microchannel is a customer-dated programme, the 800V file is "liquid cooling for high-TDP GPUs, voltage unsaid."
Segments that moved
| Line | 9M 2025 share | Notes |
|---|---|---|
| Thermal | 73%, +77% YoY | Lids / vapor chambers +82%; liquid-cooling modules +48% |
| Leadframes | 9%, −9% YoY | Auto, LED, 5G; Japanese competitor exited; 2026 back to growth |
| Electronic parts | 7%, −2% YoY | Server ILM and sockets; sockets start in 2026, conservative |
| Other | 10%, +49% YoY | Tooling and scrap |
The self-image is a mechanical-plus-thermal solution house, not a cold-plate vendor. Fasteners, backplates and sockets are how it says it can cut six months off a customer's thermal-stress problem.
Guidance
Outlook scores (5 is full marks), excluding microchannel revenue:
- Liquid-cooling modules: 5/5 — several customers' programmes enter production next year; ASP is higher than lids.
- Vapor chambers / lids: 4/5 — still AI-server demand. Last year's 4/5 on this line beat the score.
- Leadframes: 4/5 — share gain from a competitor exit.
- Server ILM and sockets: 2/5 — end-market demand is weak.
Management Q&A
Q: Why is the liquid-cooling-module outlook scored 5/5? Is that one successful product?
A: It is not one product. Over the past year Jentech built a full liquid-cooling platform covering the GPU and the surrounding loop. Several different customers' programmes go to production next year, which is why the overall view is very optimistic.
Q: What is the progress, the hard part, and customer willingness on microchannel liquid cooling?
A: The company has spent ten years on the technology. Customer interest rose sharply last year once existing cooling could no longer handle the next chips. The claimed advantages are very-high-TDP cooling and a much thinner module. Thermal simulation through packaged-chip measurements is described as very positive; Jentech is helping customers with system-integration issues, and that work is going smoothly.
Q: Why is liquid cooling scored 5/5 and vapor chambers 4/5? Is that because microchannel sits inside the water-cooling line?
A: Potential microchannel revenue is not included in these outlook scores. Liquid cooling scores higher because ASP is higher and several programmes will start to contribute. The scores are only a reference: last year vapor chambers were also a 4, and the outcome beat that by a wide margin.
Q: If a customer asked Jentech to share the design or patents with other suppliers, what would the company do?
A: It has never received that request. An AI chip may cost US$6,000 while Jentech's lid is perhaps US$10. In that setting the customer wants quality, technical lead and yield, not the lowest price. Management's line is that the AI era values technology and quality, not cheap CP value.
Q: When might customers adopt microchannel?
A: That depends on the customer's next-generation chip schedule; Jentech will not answer for them. Every sign, it said, is that the technology is effective and necessary for future high-TDP dies.
Q: What is the three-year capex plan?
A: Split between tools and buildings. 2025 total NT$900m (NT$600m tools + NT$300m buildings); 2026 NT$1.2bn (NT$600m + NT$600m); 2027 also NT$1.2bn (NT$600m + NT$600m).
Q: How does assembling microchannel at a packaging house differ from assembling a cold plate at a system house?
A: The main advantage is assembly at the advanced-packaging house, so TIM1 bond quality and coverage can be inspected by scan. A conventional cold plate built at the system house needs a thick TIM2 layer to take up substrate warpage, which hurts heat transfer and cannot be inspected well. The design idea of microchannel is to remove TIM2, the thermal bottleneck.
Disclaimer
English notes on Jentech Precision Industrial's investor conference of 28 November 2025, covering Q3 and the first nine months of 2025. Figures are as presented by management and have not been independently verified. USD equivalents are approximate at about NT$31.5 to the dollar. For reference only; not investment advice.