Why AI’s next bottleneck is memory, not GPUs
from The Five Nine Hosted by The Five Nine
This week, we dig into why the AI infrastructure conversation is shifting from compute scarcity to the memory wall. As large language models get bigger and enterprises move from training to high-volume inference, memory, storage, networking and GPU utilization are becoming critical to AI performance and economics. Featuring insights from Jason Hardy of NVIDIA and Val Bercovici of WEKA, this episode breaks down why inference workloads are fundamentally different from training, how agentic AI increases pressure on memory systems, and why idle GPUs can turn into a costly problem for enterprises building AI factories.
Embed this episode
No transcript yet.