If the bottleneck isn’t TSMC wafer starts but CoWoS, where exactly does that bottleneck come from? From what I understand, its the interposer connecting GPU and HBM wafers. Are they hard to make, is the yield bad, are there insufficient production lines, …?
