At a glance. A real earnings recovery — gross margin 18% versus 6% a year ago, and a return to profit — plus the most direct 800V sentence in this cohort of epi and foundry names: GaN is already in 800V HVDC systems, at small-volume trial. Silicon MOS is in AI; whether that is 800V, management does not know. SiC IDMs will not say.
What changed this quarter
The P&L turned. Q1 revenue NT$1.058bn, +18% YoY; operating margin 11% versus −2% a year ago; EPS NT$0.35. Three named causes: mix (8-inch up), utilisation (8-inch near 90%), and cost control.
Silicon photonics started. Germanium/silicon photodetector epitaxy was qualified last year at what management called an international first-tier customer — a US firm with more than 15 years in the field — and entered small-volume production in February. Margin is above company average; volume is still small. Capex for the year is being tripled versus the original budget, mostly 8-inch silicon epi and SiPho tools — SiPho is a slow process and needs a lot of kit.
Calls go to twice a year, as a governance change.
Order books lengthened from 1–2 months to 2–3 quarters, and some large accounts are already talking 2027–2028 capacity.
The numbers
| Q1 2026 | Value | Change |
|---|---|---|
| Revenue | NT$1.058bn (~US$33.6m) | +18% YoY, −0.5% QoQ |
| Gross profit | NT$188m (~US$6.0m), margin 18% | vs 6% a year ago |
| Operating profit | NT$112m (~US$3.6m), margin 11% | vs −2% a year ago |
| Net income | NT$100m (~US$3.2m) | — |
| EPS | NT$0.35 (~US$0.01) | vs NT$0.01 a year ago |
6-inch GaN turned profitable in the quarter. 4-inch and 5-inch utilisation is below 60%, which is why those tools are being retired to free cleanroom for 8-inch and SiPho.
NVIDIA 800V read-through
Asked directly how the company is positioned in 800V HVDC systems, management split the three materials:
- GaN — yes. Product "has definitely entered" 800V HVDC systems. Small-volume trial, not a mix number, not a named rack.
- Silicon MOS — in AI, unknown rail. Already in the AI supply chain; management cannot say whether those parts sit on an 800V bus.
- SiC — unknown. IDM customers are conservative about end-application disclosure.
That is a cleaner answer than most second-tier names give, and it should be read at face value. A trial GaN wafer in an HVDC system is a real data point. It is not 800V revenue. Silicon epi is still ~90% of sales, and the thing that actually moved the quarter is 8-inch silicon for AI power and PMICs, plus utilisation — the same AI power recovery everyone else is printing, which does not require 800 V.
The Ge/Si photodetector line is an AI optics story. Do not fold it into the HVDC read.
Non-China supply is the commercial edge management wants on this stack: AI customers asking for a non-red epi source, versus Innoscience and other mainland houses. China stays China-for-China.
Segments that moved
| Line | Q1 2026 share | Notes |
|---|---|---|
| Silicon epitaxy | ~90% | 8-inch now 70% of this line, from 55% |
| Compound (GaN + SiC) | ~10% | 6-inch GaN profitable; 8-inch GaN in small production |
| Ge/Si (SiPho) | not split out | Small-volume from February; may become its own line later |
SiC: a new Japanese IDM, volume aimed at Q2 2026, automotive and high-end industrial. EV is choppy near term; management still likes the longer autopilot/EV case.
GaN beyond AI servers: humanoid-robot motor drives and aerospace (radiation tolerance). Future GaN expansion is 8-inch; tools are tight, and a plan with shareholder HMC is hoped for within one to two years.
Guidance
- Q2 2026 revenue +20% or more sequentially; silicon epi and compound both expected to grow.
- Full year 2026 described as relatively optimistic; second half better than first. No H2/H1 ratio given.
- Gross margin and utilisation can hold Q1 levels, with a chance to go higher once SiPho contributes.
- Full-year capex at 3× the original plan. Expansion toward 2028 demand is being paired with customer capacity commitments.
Management Q&A
Q: Industry and company outlook?
A: AI infrastructure spending is strong, and order visibility has extended to 2–3 quarters. Large silicon-epi customers are already discussing capacity for 2027–2028. The full year is described as optimistic, and the momentum is expected to continue for the next 1–2 years.
Q: HVDC (800V high-voltage DC) positioning?
A: For silicon MOS, a customer is confirmed in the AI supply chain, but whether those parts sit in an 800V system is uncertain. For GaN, entry into 800V HVDC is confirmed, still in trial production. For SiC, IDM customers are conservative about end-use, so information is limited.
Q: Pricing plans and raw-material shortage risks?
A: Costs are rising, especially materials, but only minor price adjustments have been made for some customers. Further adjustments may come in the second half, depending on costs and what customers can absorb — not an across-the-board hike. Chemicals and rare-earth controls are not causing shortages in the current plan.
Q: Revenue mix of silicon, GaN and SiC?
A: Silicon epitaxy is about 90% of revenue. Compound semiconductors (GaN + SiC) together are about 10% and are not split further, because substrate costs differ. Silicon photonics (germanium-on-silicon) is still very small; it may be broken out as its own line later.
Q: Q2 and full-year outlook?
A: Q2 revenue is expected to grow 20% or more sequentially. For the full year, the second half should be better than the first; no 45/55 split was given. Overall the year is described as optimistic. Gross margin and utilisation can hold Q1 levels and may still go higher.
Q: Silicon-photonics equipment purchases and customer type?
A: Capex is being raised to about three times the original board budget, mainly 8-inch silicon epi and SiPho tools. MOCVD tool counts will be high because the process is slow, but the specific number was not disclosed. The customer is a US company with more than 15 years in the field. Qualification finished last year; small-volume production began in February. Gross margin is above the company average; ASP was not disclosed.
Q: Capacity utilisation for 8-inch and 4/5-inch?
A: 8-inch is close to 90%. 4-inch and 5-inch are below 60%, and small-size capacity is being phased out so the cleanroom can take 8-inch and SiPho tools.
Q: How high can the compound-semiconductor mix go, and where do GaN and SiC grow?
A: As high as possible, depending on the market. AI helps PD (silicon photonics), GaN and SiC. SiC is tied to EVs and autonomous driving over the longer term. GaN fits high-frequency, low-loss motor drivers in humanoid robots. Low-voltage GaN in the 800V-to-48/54V rack conversion is a fit, but the ramp timing is for NVIDIA to call.
Q: GaN expansion plan?
A: Volume today is 6-inch; 8-inch has been sampled and some lots have passed qualification. Future expansion is meant to be 8-inch first. 8-inch tools are scarce; the company is working with major shareholder HMC and hopes to have a clearer plan within one to two years.
Q: Facing competition from China's Innoscience?
A: Innoscience is a large 8-inch mainland producer. AI-supply-chain customers generally require a non-China source. Management reads NVIDIA's posture as China-for-China, with non-China still served by non-China suppliers — the intended separation on GaN now, and expected on SiC later.
Disclaimer
English notes on Episil-Precision's investor conference of 29 April 2026, covering Q1 2026. 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. For reference only; not investment advice.