TSMC's conversion of 4nm/5nm capacity to 3nm to clear a backlog stretching to 2028 shows 3nm is the bottleneck, and TSMC controls the queue. Broadcom flagged the constraint publicly. The mix shifts toward 3nm premium pricing, with AI accelerator growth guided in the mid-to-high 50s% over five years, so every conversion is margin expansion disguised as logistics. TSMC is trading handset volume for HPC and AI margin, leaving Qualcomm, MediaTek and other non-Apple customers to absorb the squeeze. x.comÂ

