--- title: "A 144% jump in one small revenue line is the real SiC story at StarPower" date: 2026-11-09 category: Chips & Compute site: NeuroAI canonical: https://neuroai.site/a/na-chips-starpower-sic-module-squeeze language: en --- # A 144% jump in one small revenue line is the real SiC story at StarPower > StarPower's SiC modules jumped 144% in 2025, yet profit fell — a clear look at what module-scale silicon carbide actually earns. On a line in Jiaxing, a robot places a silicon-carbide die onto a copper substrate, and a few steps later the whole assembly disappears into a grey plastic housing. What comes out is a power module roughly the size of a paperback, destined for the inverter of an electric car. For most of the past decade, this was a sideshow for StarPower Semiconductor (斯达半导, 603290.SH). The company built its business on silicon IGBT modules — the workhorse switch of electric drivetrains — and treated silicon carbide (碳化硅, SiC) as a research line with a long fuse. In 2025 the fuse lit. The company's "other products" revenue, which is mostly automotive SiC modules, rose 144% to RMB 654 million (≈ US$92.1M / HK$719M). That is still a small slice of a RMB 4.01 billion (≈ US$565M / HK$4.41B) business, but it is the first year the line has been big enough to change the shape of the income statement rather than just the slide deck. ## The awkward combination: more sales, less profit StarPower's 2025 results read like two different companies. Revenue grew 18.3% to a record, while net profit fell 20.2% to RMB 405 million (≈ US$57.0M / HK$446M). Gross margin slipped, and R&D spending rose 35.9% to about RMB 481 million (≈ US$67.7M / HK$529M). The explanation is not a demand problem. IGBT module shipments reached 15.22 million units, generating RMB 3.35 billion (≈ US$472M / HK$3.69B), or 84% of the business. New energy applications — vehicles, solar, storage — brought in RMB 2.55 billion (≈ US$359M / HK$2.81B), up 27%, with the solar-and-storage slice growing 55.7%. The squeeze came from the price side. Mature IGBT modules have become a crowded market in China, with several domestic suppliers chasing the same vehicle platforms. Building a SiC business at the same time means carrying two cost structures: the old line defending price, the new line absorbing ramp costs and qualification expenses. ## Why SiC modules are a different business from SiC wafers China's SiC conversation has been dominated by substrates — the crystalline wafers that everything else is built on. That is a materials story, with its own brutal price cycle. Modules are a manufacturing and qualification story. A carmaker does not buy a wafer; it buys a part that must survive a decade of thermal cycling under the hood, pass automotive qualification, and match a specific inverter design. Once designed in, it is hard to design out. That is why the module line matters more than its revenue share suggests. StarPower says it has built capacity for 60,000 6-inch-equivalent automotive SiC MOSFET wafers a year, alongside 300,000 wafers of high-voltage silicon capacity. In early 2026 it launched a fundraising project for a plant that would add 2.8 million SiC modules a year, at a stated investment of RMB 1.0 billion (≈ US$141M / HK$1.1B), with full-capacity revenue projected at RMB 1.8 billion (≈ US$254M / HK$1.98B). The arithmetic in that disclosure is worth sitting with: the new plant, at full output, is designed to generate more revenue than the entire SiC line does today, at an average selling price of roughly RMB 640 per module. ## The second demand curve nobody planned for StarPower spent 2025 telling investors that its third-generation SiC chips were aimed at 800V architectures — the higher-voltage electrical systems that let cars charge faster and run thinner wiring. The same voltage class is now turning up in data centers. AI racks have crossed the point where 48V distribution stops making sense, and the industry is moving toward 800V direct-current architectures, solid-state transformers and higher-density power shelves. Those are SiC problems, not silicon problems. StarPower has said its SiC platform is being positioned for exactly that: 800V DC power in AI facilities, plus fast-charging equipment. It has also started developing GaN HEMT devices and low-voltage MOSFETs aimed at power supplies, with customer sampling expected from 2026 — an admission that the module maker intends to sell into the rack, not only into the car. ## What a module maker actually controls The company runs a hybrid model: chips designed in-house, many fabricated at foundries such as Hua Hong and GTA, modules assembled and tested in its own plants. That gives it control over the part customers care about — the packaged, qualified, warrantied module — without carrying a full wafer fab for every product. It also means the SiC ramp depends on two things it does not fully control: foundry wafer supply and automotive qualification timing. Both are slow. Both explain why the revenue line took years to move and why it can stall just as easily. ## Key takeaways - StarPower (斯达半导) grew 2025 revenue 18.3% to RMB 4.01 billion (≈ US$565M / HK$4.41B) but net profit fell 20.2%, as IGBT price pressure and SiC ramp costs hit margins. - SiC modules — the "other products" line — jumped 144% to RMB 654 million (≈ US$92.1M / HK$719M), the first year the business was large enough to matter. - The next leg is not cars alone: 800V DC power for AI data centers is being positioned as a second market, with GaN and low-voltage devices sampling from 2026. ## Honest limitations Revenue, profit, unit shipments and capacity figures come from StarPower's 2025 annual report and its summary published in Shanghai Securities News, cross-checked against the Shanghai-listed disclosure and trade-press summaries of the same filing. The 60,000-wafer and 2.8-million-module capacity numbers are company statements about installed or planned capacity; I have not seen independent verification of utilization or yield. The RMB 640-per-module figure is my arithmetic from the company's own project disclosure, not a disclosed price. I did not verify how much of the 144% growth came from price versus volume, and the AI data-center positioning is a stated intention, not booked revenue. ## What readers can do now - Judge SiC suppliers by module revenue and design wins, not by wafer capacity announcements; capacity without a qualified part is inventory. - When a company grows revenue and shrinks profit in the same year, check whether the cause is price or ramp cost — the two have very different half-lives. - Watch 800V DC power designs in AI facilities as a demand signal for power semiconductors that has nothing to do with electric vehicles. --- Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-chips-starpower-sic-module-squeeze Free to quote with attribution and a link to the original.