Ann: AGM Presentation, page-5

yes from what i am reading whilst the huge computing power to run these large models in the cloud is gaining most of the attention the interconnect and networking speeds are just as important as these models draw on huge amounts of data to support these models which will need to be accessed from many servers. from what . NVIDIAS current top end architecture is infiniband 400gbs , with their infiband slated for release next year coming in at 800gbs. and there is NVlink a high speed interface standard for closely connected computers. ethernet is cheaper though and these high end interfaces are more the go to too for supercomputers. but as these cpu's and gpu's become faster and faster the need for high performance neyworking does too.

if banks are implementing their own ai llm's (think an AI system that is dedicated to your internal datasets ) they will probably be looking to setup server clusters to support this.

dont know if this means anything for DDR but they do seem well positioned and are already gaining traction supporting what is undeniably a growing megatrend.