AI computing power is surging, making power semiconductors a "rigid demand" - 36 Kr

As generative AI computing power demand surges, the power density of AI data centers is approaching the megawatt level, driving demand for power semiconductors. Xinlian Power, a subsidiary of Corelink... Read original →

The surge in AI computing power—now approaching megawatt-scale per rack—is forcing a fundamental redesign of power delivery architectures. Xinlian Power’s 3300V SiC MOSFET isn’t just plugging a domestic gap in solid-state transformers; it slashes system BOM costs by up to 35%, accelerating 800V HVDC adoption in data centers. Technologically, this triggers co-innovation in magnetics and thermal management due to higher switching frequencies. Geopolitically, tightening U.S. export controls on wide-bandgap semiconductor tools are pushing Chinese firms to localize 8-inch SiC fabrication, though yield stability remains a bottleneck. Competitively, Infineon and Wolfspeed may respond with IP litigation or aggressive pricing, while NVIDIA could deepen partnerships with local SiC suppliers to de-risk its AI infrastructure. Over the next 18 months, high-voltage SiC modules will become a strategic chokepoint—driven by converging demand from AI clusters and 800V EV platforms—making domestic supply not optional, but essential for infrastructure resilience.