Semiconductor Insider(@SemiconductorsX):Substrate capacity is the real bottleneck. The client list is not all signed. Google and MediaTek on TPUv9 with EMIB-T for 2028 looks solid. AWS testing EMIB for a later Trainium variant is possible. Qualcomm “Rolex” and Microsoft Cobalt 300 still have no primary confirmation. Those two rows read like leaks, not booked sockets. Intel is prepaying Japanese and Taiwanese substrate makers and using Amkor to scale assembly. That can cover the Google volume. The rest of the table needs more evidence before anyone treats it as a pipeline. Boards and prepayments are real. A full foundry book built from rumor rows is not. $INTC $GOOGL $AMZN

2026.09.21 11:51

Substrate capacity is the real bottleneck. The client list is not all signed. Google and MediaTek on TPUv9 with EMIB-T for 2028 looks solid. AWS testing EMIB for a later Trainium variant is possible. Qualcomm “Rolex” and Microsoft Cobalt 300 still have no primary confirmation. Those two rows read like leaks, not booked sockets. Intel is prepaying Japanese and Taiwanese substrate makers and using Amkor to scale assembly. That can cover the Google volume. The rest of the table needs more evidence before anyone treats it as a pipeline. Boards and prepayments are real. A full foundry book built from rumor rows is not. $INTC $GOOGL $AMZN

Jukan@jukan05

2026.09.21 11:25

I’m not sure Intel has secured enough substrate capacity to support all these customers. And… Qualcomm’s CPU is codenamed Rolex? That’s pretty cool. $INTC