Random Access Memory is the fast, volatile memory that holds data a computer is actively using. RAM is orders of magnitude faster than persistent storage (SSDs, hard drives) and orders of magnitude slower than on-chip CPU cache. The amount of RAM in a system is typically the binding constraint on how much data it can process without spilling to disk.
For AI workloads, GPU RAM (often called VRAM or HBM) is frequently the more important quantity. Frontier AI model inference is often memory-bound rather than compute-bound — the ability to fit the model weights and the KV cache in GPU memory determines what batch sizes and context lengths are possible. The competitive differentiation between GPUs (H100 80GB, H200 141GB, B200 192GB) often comes down to memory capacity and bandwidth rather than raw FLOPS.
For enterprise deployment planning, both CPU RAM (for general workloads) and GPU RAM (for AI) are capacity planning quantities that deserve explicit architectural attention rather than being left to defaults.