At a glance. A record quarter in which servers and networking reached 66% of sales. The 800V-relevant number is capacity, not voltage: liquid-cooling modules are being taken from 200,000 to 1 million sets a month. The call never said 800 V.
What changed this quarter
The mix, not just the scale. Thermal products were 63.6% of sales; chassis 20.2%. Servers and networking at 66.4% is why margins moved — gross margin 29.77%, up 3.9 points year on year.
NVIDIA's six tray designs were claimed as a full set. Management said the new Vera Rubin / Blackwell tray family looks fully liquid-cooled, and that Auras is developing against all six with different partners.
ASIC was given a design-role, not just a customer list. On NVIDIA work, NVIDIA draws the architecture and Auras builds it. On ASIC work — software and cloud companies — Auras says it draws the thermal architecture itself, which is why ASIC programmes are described as higher-margin.
The numbers
| Q1 2026 | Value | Change |
|---|---|---|
| Revenue | NT$49.04bn (~US$1.56bn) | +110% YoY, +2.6% QoQ |
| Gross margin | 29.77% | +3.94pp YoY, +3.38pp QoQ |
| Operating margin | 24.51% | — |
| Net income to parent | NT$7.92bn (~US$251m) | +146% YoY |
| EPS | NT$20.17 (~US$0.64) | +144% YoY |
Q1 capex was more than NT$3.6bn (~US$114m). Inventory days fell from 194 to 132. Full-year capex is budgeted at NT$15bn (~US$476m) in 2026 and NT$17bn (~US$540m) in 2027.
Product mix: thermal 63.6%, chassis 20.2%, systems and peripherals 7.7%, hinge subsidiary 8.5%.
NVIDIA 800V read-through
Not on NVIDIA's power-component list. The whitepaper's cooling block is the relevant one: 45°C loop, CDU, sidecar, liquid-cooled busbar.
What this call actually gives that block:
- Liquid-cooling module capacity 5x, 200,000 to 1 million sets a month, with chassis and manifold lines in Vietnam and China moving with it.
- Most ASIC deployments from 2026 are liquid, and the loop is spreading off the GPU/CPU onto memory and switches on the same tray — more plates per cabinet, not just a hotter GPU.
- CDU is already a custom, in-server unit for at least one hyperscaler. Management wants it to become a real revenue line over the next few years, not a side item.
- A rumoured drop of gold plating on Vera Rubin, if it happens, was described as outsourced anyway and not material to Auras's 2H–2027 path.
- Micro-channel lids were dismissed as a dead end on reliability and serviceability. Two-phase liquid cooling is in joint development, tested about 15% better than single-phase and marketed as a leak-risk fix; copper cold plates could reuse current fabrication.
None of that is an 800 V disclosure. It is the cooling half of the same rack.
Segments that moved
Servers and networking at 66% is the company. Management said US and Chinese hyperscalers are both on ASIC thermal programmes, and that it is already working with every US hyperscaler's ASIC project plus almost all of the Chinese ones — without naming them.
Chassis is growing with the loop. Racks are being expanded more slowly: a two-metre cabinet eats floor space, so capacity is rationed toward higher revenue per square metre.
Phone and tablet VC still contributes, but cannot match AI-server dollars.
Guidance
- Full year 2026: sequential growth through the year, with the steeper step in the second half as new CSP programmes start.
- ASIC revenue still expected to overtake GPU in 2027.
- ASIC gross margin above NVIDIA programmes, because of the design work.
- Second-half seasonality is described as steeper on customer forecasts, with the caveat that Vera Rubin is a harder system integration than GB200 to GB300, so the slope is not locked.
Management Q&A
Q: How should Q2 2026 and the gross-margin trend be viewed?
A: Sequential growth is expected through 2026. Q2 is still running programmes that extend from the second half of last year; the sharper change is the second half, when new projects with leading CSP customers come out. Servers and networking remaining the fastest-growing segment should keep supporting mix and margins, though a quantitative margin figure was not given.
Q: How is this year's momentum split between the first and second halves?
A: No quantitative split was given, but management said it is very excited about second-half new programmes. It is working with hyperscalers across the United States and China on thermal solutions for the next generation of AI.
Q: At GTC, NVIDIA showed several rack designs — Vera Rubin, Blackwell LPX, CPO switch and others. Will they all be liquid-cooled, and what is Auras's role?
A: NVIDIA described a six-tray ecosystem, which Auras reads as proof that AI infrastructure is more than one chip architecture. Those new designs currently look fully liquid-cooled, and Auras said it is already working with different companies on all six trays, on the customer's timeline rather than a published Auras date.
Q: How does the recent thermal-design change (gold plating) affect the business?
A: The gold-plating spec change on Vera Rubin has not been finally confirmed. It would hit other names in the chain harder than Auras, because Auras had already planned to outsource that process, so dropping the spec would not materially change the second-half-through-next-year trajectory. Gold plating was tied to a possible switch from graphene to liquid-metal TIM; liquid metal is a viable option, not an inevitable one.
Q: What is the view on micro-channel lids and on next-generation GPU thermal technology?
A: Auras remains firm that micro-channel lids will not be used on any future AI-server generation, because of reliability, serviceability and downtime constraints. Single-phase liquid cooling is still being improved generation by generation — as air cooling went from tens of watts to as much as 1,300 W — and two-phase liquid cooling is in active R&D, with about 15% better thermal performance versus like-for-like single-phase and hyperscalers already co-developing it.
Q: Does two-phase liquid cooling need an entirely new equipment set?
A: Work is still in the research phase. Cold plates are still likely to be copper, with manifolds as inlets and outlets, so existing copper machining and welding should still apply. What extra system components — and therefore which new manufacturing steps — will be needed has not been defined yet.
Q: What is liquid-cooling penetration at ASIC customers now, and where does it go over three years?
A: From this year, the vast majority of solutions developed and deployed for ASIC customers will be liquid-cooled; some air-cooled ASIC chips remain, but they are already a minority and should keep shrinking. The larger growth is liquid cooling spreading from GPU/CPU to memory, switches and other auxiliary parts, as in Vera Rubin cooling the whole tray rather than only the processors.
Q: How is ASIC work progressing with new hyperscalers such as G or M?
A: No single customer was named. Starting this year, Auras said it is already working on every US hyperscaler ASIC programme and on practically all of the Chinese hyperscaler ASIC programmes as well.
Q: Is ASIC still expected to overtake GPU customers in 2027?
A: Yes — that view is unchanged on the company's own capacity plan and forecast. ASIC customers as a group keep rising in importance, without NVIDIA or AMD ceasing to matter.
Q: How do NVIDIA and ASIC cold plates differ, and where is the custom-design value?
A: The analogy used was a tailored suit: the stitching is similar, but every programme is designed from scratch around chip size, heat, tray count, space and flow. NVIDIA is a hardware company, so the working model is closer to NVIDIA as architect and Auras as civil engineer; ASIC customers, often software or service firms, need Auras to design the thermal architecture as well, which is the extra value-add and why design-standardisation risk is described as mainly an NVIDIA issue, not an ASIC one.
Q: Will China ASIC growth be stronger than US ASIC next year (2027)?
A: Ramp timing is hard to predict because it is the customer's call. Auras is already first source for most of its Chinese CSP accounts, including the largest, and said it will be ready when they want to ramp.
Q: How intense is price competition in China?
A: Auras's edge in China is design capability, which local manufacturers are said to lack, so no serious AI-product price war is visible yet. China AI build-out is still early; the risk is not expected to be large for this year and the next several years, though it could rise if AI infrastructure becomes more of a commodity.
Q: Will second-half seasonality be stronger than last year?
A: Last year's steep second half partly reflected a weak first half when GB200 ramped poorly; this year's first half is sturdier, while customer forecasts show a steeper second-half curve, so the two effects pull against each other. Vera Rubin is also a harder system integration than GB200 to GB300 — GB300 was about 90% the same architecture — so there is still second-half execution risk and no locked seasonality figure.
Q: Do ASIC and NVIDIA programmes have different gross margins?
A: Yes — ASIC programmes usually carry higher gross margin than NVIDIA programmes, because Auras supplies more thermal architecture and design on ASIC, whereas NVIDIA has a strong in-house system and thermal team.
Q: What is the manifold opportunity in the Vera Rubin generation?
A: Auras positions itself as a full thermal-and-mechanical supplier, so tray manifolds and rack manifolds are being expanded alongside liquid-cooling modules, racks and chassis. Those parts are designed together, which is the point of offering the whole loop.
Q: What is the growth outlook for chassis and racks?
A: Chassis and mechanicals are growing well and presence is still being expanded. Racks are also being added, but a rack is a 2-metre product that uses a lot of floor space, so on value per square metre the company still prioritises higher-profit, higher-revenue components first.
Q: What is the latest CDU development progress?
A: Before CDUs became standalone racks they sat in the server, top or bottom, and Auras has already shipped custom in-server CDUs to selected hyperscalers. Progress continues, in a field with strong system-integration competitors; CDU is hoped to become a proper revenue line in a few years, not immediately.
Q: How is 2026 smartphone vapour-chamber momentum?
A: No named customer. The product type is spreading into more models and form factors, including phones and tablets, out of the Wuhan plant. The contribution will not match AI servers, but it should still be a meaningful piece of 2026 revenue.
Q: What is this year's capacity plan and capex budget?
A: 2026 capex is NT$15bn and 2027 is NT$17bn. Liquid-cooling module capacity is being raised from 200,000 units a month to 1 million; that 5x is for liquid-cooling modules only, with additional expansion in chassis, manifolds and other parts in Vietnam and China.
Q: If cold-plate capacity rises 5x, does related revenue also rise 5x?
A: Not as a straight line. Some future liquid-cooling modules will cost more than today's GB300 and some less, and next-generation architectures are too varied for a linear conversion from capacity to revenue.
Disclaimer
English notes on Auras Technology's investor conference of 14 May 2026, covering Q1 2026. 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.