The last two posts were about HBM/Memory and Power & Cooling. Both topics show that the critical points in the AI stack are shifting step by step.
Today it's all about Advanced Packaging.
The reason for this lies in the architecture of modern AI systems: memory, compute and other components are are moving ever closer together physically. Without this integration, current performance and bandwidth requirements would hardly be achievable.
This is also changing the role of packaging. It used to be more of a downstream production step. Now it is becoming a key technological key technological point.
This is particularly evident in HBM (High Bandwidth Memory). The memory is located directly next to the GPU, data must be transferred extremely quickly, while at the same time the requirements for precision, heat dissipation and yield increase considerably.
That's why I look at companies like $ASX (+2,41 %) (ASE Technology), $BESI (-0,19 %) (BE Semiconductor) or even $TSM (-1,58 %) (TSMC) are currently very close to this bottleneck.
I find one thing particularly interesting: the limiting factor is once again shifting a little further within the infrastructure. It is no longer just computing power that counts, but increasingly the ability to connect highly complex components efficiently at all.
Next, I'll take a look at energy/grid. Another bottleneck area that is currently becoming critical within the technology stack.
