Why a Legacy Cloud Mentality Breaks AI Infrastructure | Deep Geeks ep03

Two words nobody expected to hear together: legacy cloud. It describes how most technology leaders still think about AI, and why it does not work. https://www.weka.io/podcast/?utm_source=youtube&utm_medium=social In the cloud era you provisioned a large CPU instance, then optionally added memory, then optionally added storage. Composable, incremental, familiar. AI does not work that way. In this clip from DeepGeeks Episode 3, we unpack why. Performance requirements are too high and latency tolerances too low, so GPUs and high bandwidth memory get packaged together and provisioned as a single unit. That creates a real problem: training and inference want different memory to compute ratios. Training benefits from the memory that ships with the GPU. Inference wants far more memory and far less GPU. Today you largely cannot buy or deploy systems that way. Disaggregating compute from memory for inference is what changes that, and it returns AI infrastructure to a mental model cloud era technical leaders already understand. Expect to see a lot more of it through 2026. Learn more: Listen to the full episode: https://www.weka.io/podcast/?utm_source=youtube&utm_medium=social&utm_content=full-episode WEKA NeuralMesh: https://www.weka.io/product/neuralmesh/?utm_source=youtube&utm_medium=social&utm_content=product-page Subscribe for more AI infrastructure content: https://www.youtube.com/@WekaIO?sub_confirmation=1