At a glance. A loss-making first half on idle compound capacity and new-tool depreciation — and, underneath it, the demand story management wants you to underwrite: AI racks heading for 600 kW in 2027, the industry moving to 800V HVDC, and that architecture pulling in the SiC and GaN wafers Episil runs. GaN is already growing. SiC is still the hole in the P&L.
What changed this half
GaN and silicon grew; SiC did not. On the foundry line, GaN was +11% on AI servers and related uses, silicon-base +8% on a consumer bounce, and SiC −33% — a smaller decline than before, but still a decline. Compound's share of group sales fell from 49% to 41%. Humanoid robots were named as a further GaN demand pocket, not as the thing that produced the 11%.
Gen4 planar SiC MOSFET cleared customer reliability. Versus Gen3: +20% performance, −20% die size, 1,000-hour reliability done. Small-volume production is guided for 2025Q4, becoming a mainstay next year. Trench MOSFET, co-developed with customers, is aimed at 2026Q2 volume, on a port-in model so Episil does not put its own trench platform into a patent thicket.
An 8-inch SiC/GaN pilot is being installed at Vanguard, not in Episil's own fab, at 1,500 wafers per month. Trial production is guided for the first half of 2026, qualification in the second half. The point of using Vanguard's building is cost.
The numbers
| H1 2025 (consolidated) | Value |
|---|---|
| Revenue | NT$2.662bn (~US$84.5m) |
| Gross profit | −NT$131m, margin −5% |
| Operating profit | −NT$464m, margin −17% |
| Net income | −NT$407m (~US$12.9m) |
| EPS | −NT$1.02 (~−US$0.03) |
| Book value / share | NT$18.58 (30 Jun 2025) |
Foundry alone: revenue NT$990m (~US$31.4m), −6.8% YoY.
The loss is utilisation plus depreciation on new tools, not a demand collapse across the board. SiC customers stayed conservative on inventory after a weak solar and EV year and a China price war. Management's other price point: SiC versus IGBT has compressed from 2–3× to 1.2–1.5×, which it calls the cost-performance "sweet spot" for faster substitution.
NVIDIA 800V read-through
This is a foundry demand thesis, not a shipping 800V product.
Management's chain is explicit: AI server rack power is expected to break 600 kW in 2027 → the industry is pushing 800V HVDC → that architecture "will significantly lift" demand for SiC and GaN power devices. On the foundry line it already names data-center DC-to-DC converters as an application, alongside EV on-board chargers, solar, a push into EV main inverters, industrial equipment, and parts above 3300 V.
What the half actually printed is the split inside that thesis. GaN +11% is the AI-server (and robot) line working today — which may still be 48 V / 54 V iron, not 800 V. SiC −33% is solar, EV and China pricing, not a data-center miss. There is no named hyperscaler, no wafer start attributed to an 800V design, and no mix figure.
Treat 800V as the reason Episil wants you to fund 8-inch compound capacity and wait for SiC to rebound — not as evidence it is on NVIDIA's bill of materials.
Segments that moved
By process (H1 2025): compound SiC & GaN 41% (from 49%); TVS/ESD 25% (from 17%); automotive MOSFET 19% (from 14%); fast-recovery diodes 1%, being displaced by SiC diodes.
By end market: industrial 42%; consumer 30%; automotive 22% (from 28%); green energy 6%.
Capacity: 6-inch SiC 5,000 wpm, GaN 2,000 wpm. Non-China IDM volumes are expected to keep rising; China fabless remains on the book.
Guidance
- H2 2025 group revenue +10% to 20% versus H1, driven by a SiC rebound as customers restock. Silicon-base to grow slowly. GaN H2 is conservative — geopolitics and tariffs, visibility poor.
- 2026 double-digit growth versus 2025, with compound as the engine and silicon-base flat to slightly down.
- Gen4 planar SiC: small revenue from 2025Q4, a mainstay in 2026. Trench: volume 2026Q2.
- 8-inch pilot: trial H1 2026. Longer-term capex will follow actual IDM demand; no number given.
Management Q&A
Q: Views on 12-inch SiC substrates and their use?
A: 12-inch SiC substrates are mainly for advanced packaging, using their thermal conductivity, and have no direct link to Episil's power-device business. The trend is still a good opening for non-China substrate suppliers, because the AI chain is likely to exclude mainland sources and that would rebalance a China-dominated substrate market.
Q: Outlook for the second half and next year?
A: H2 2025 revenue is expected to grow 10–20% versus the first half, driven by compound semiconductors. 2026 is still expected to grow at a double-digit rate year-on-year; silicon is seen as flat to slightly down, with compound carrying the growth.
Q: Current SiC and GaN capacity plans, main customers and applications?
A: 6-inch SiC is 5,000 wafers a month (full MOSFET) and GaN 2,000; the 8-inch pilot, now installing, is planned at 1,500 wafers a month, with a first-half-next-year pilot run and second-half qualification. Customers are mainland fabless houses plus non-China IDMs, and non-China IDM volume is expected to keep rising from this second half into next year. Applications have moved from EV onboard chargers and solar into the EV main inverter, data-center DC-to-DC, industrial equipment and devices above 3,300 V.
Q: Order status for the fourth-generation planar SiC MOSFET?
A: It has passed customers' 1,000-hour reliability tests, and customers are satisfied with performance and yield. Several customers have already taken it up; small-volume production is expected from Q4 this year, and it should be one of the main products next year.
Q: Funding for the 8-inch line and future capex?
A: The near-term focus is standing up the 1,500-wafers-a-month pilot and using it to engage large IDMs. Further expansion will follow actual customer demand, so a concrete long-term capex number is hard to give now; that 1,500 wafers of 8-inch capacity is expected to be ready next year.
Q: GaN in EV main traction, and trench-technology patent issues?
A: What is entering the main traction inverter is SiC, not GaN; GaN in the car is mainly OBC or DC-to-DC, with new demand from data centers, drones and humanoid robots. Because trench has many patent issues, Episil will not offer a standard trench platform, but it will foundry for customers who bring their own unencumbered IP.
Q: How does Episil-Precision compete with Chinese epi houses?
A: Chinese epi firms have already entered international accounts, but many customers — Japanese names in particular — still want a non-China supply-chain solution for geopolitical reasons. Episil-Precision is one of the few non-Chinese compound-epi houses and can go after that market.
Q: Product-price trend, and the 8-inch versus 6-inch price gap?
A: GaN prices are stable to slightly up. SiC prices fell sharply over the past year — by about 60% — and have now stabilised, with limited room to fall further. Early 8-inch pricing may track area gain (customers may look for about 1.8× at the same generation), but later nodes such as Gen5 and Gen6 should carry more value than a simple area multiple.
Q: Is the 8-inch SiC line being built at Episil or at Vanguard, and what is the future cooperation model?
A: The line is being built at Vanguard, using its existing buildings and tools to keep cost down. Initially all of the business will be led by Episil; if a large IDM later needs a very large investment, the customer might work with Vanguard directly, but the two say they will split roles so they do not compete internally.
Disclaimer
English notes on Episil Technologies' investor conference of 17 September 2025, covering the first half of 2025. 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.