TL;DR
- Supply Leverage: China’s reported silicon carbide position is becoming an AI infrastructure supply-chain concern.
- Power Demand: Wolfspeed’s 3.3 kV modules tie silicon carbide to AI data center and energy infrastructure power systems.
- Capacity Context: A Chongqing 8-inch facility began operations in February 2025, but current output remains unverified.
- Market Risk: Non-China suppliers need qualified 8-inch output before customers can treat the reported position as manageable.
AI data centers are pushing power systems toward higher voltages, and that is making silicon carbide supply harder to ignore. The material is used in power semiconductors, and the move to larger 8-inch wafers gives China’s manufacturing position more weight in the supply chain.
The commercial stakes are growing with the same power shift. One market estimate puts power electronics on track to grow at a 10 percent annual rate to more than $65 billion by 2036, raising the value of reliable silicon carbide wafer supply. That does not prove China’s current output or market share, but it explains why the reported leverage matters as AI data centers demand more high-voltage power hardware.
Why AI Power Hardware Is Turning to Silicon Carbide
Silicon carbide is a power-semiconductor material used when equipment has to handle high voltage, heat, and current. In AI data centers, that matters because electricity has to be converted and switched efficiently through dense computing and energy systems.
That demand is already visible in supplier products. On May 21, Wolfspeed, a supplier of silicon carbide power devices, introduced 3.3 kV silicon carbide power module families for AI data centers and energy infrastructure, using half-bridge and full-bridge designs for high-voltage conversion. Toshiba’s 1200 V silicon carbide MOSFET, a power-switching transistor, points to the same need for devices that can control higher voltages in AI data-center power systems.
High-voltage direct-current designs, including 800 V systems, are part of the electrical path that moves power through larger AI and energy installations. If the required silicon carbide devices or substrates are hard to source or qualify, the bottleneck can move upstream from a product order to the pace of power-system deployment.
That is why the 8-inch wafer transition matters. A wafer is the circular slice of semiconductor material used to make devices; a larger 8-inch format can improve scale, but only if suppliers keep yields, reliability, and customer qualification under control.
Chongqing Facility Gives China’s Capacity Claim Context
China’s reported leverage has a concrete historical marker in Chongqing. In February 2025, an 8-inch silicon carbide facility began operations there after project partners Sanan Optoelectronics and STMicroelectronics established it in 2023.
The timing matters because the project was already underway before the current 8-inch wafer discussion around AI power hardware. It supports the idea that China has been building relevant capacity, but it does not confirm current 2026 output, pricing, or market share.
The scale matters as well. Project materials pointed to a fourth-quarter 2025 mass-production target and planned output of about 10,000 automotive-grade wafers per week once fully operational, backed by about 23 billion yuan, or roughly $3.2 billion, in investment. Those figures show why 8-inch silicon carbide capacity is expensive to build, while remaining historical targets rather than proof of today’s production.
The broader context is that wafer size has become a scale marker across compound-semiconductor manufacturing. That makes China’s silicon carbide capacity relevant to AI power supply, especially as export controls continue to shape China-linked chip manufacturing, without turning the available evidence into proof of global control.
What to Watch Next
Because the current record does not verify 2026 China output, near-term proof points are current 8-inch silicon carbide output, substrate pricing, customer qualification, and whether non-China suppliers can secure enough material for AI data center power modules. Wolfspeed’s full-bridge module samples were available for select customers through direct sales representatives, making sample availability one concrete gate for product adoption.
Until those proof points arrive, AI infrastructure operators do not face evidence that every SiC device is suddenly unavailable. As more power systems move to demanding electrical designs, predictable material sourcing becomes a near-term supply-chain question. Non-China suppliers need customer-qualified 8-inch SiC output before AI data center power customers can treat China’s reported position as a manageable supply risk.
