Kabelo Dike / Writing & Notes

AI Stack Mental Model

· Drafting

The AI stack shows the layers that work together to power AI applications, from energy and facilities at the base to applications at the top. Click a layer to explore its role and how it connects to the rest of the stack.

The AI stack Five AI stack layers, from energy and facilities through chips, serving infrastructure, and foundation models to applications, with representative examples in each layer. Pages are available for L3, L4, and L5. ABSTRACTION ↑ L5 Application Agent harness · planning loops · tool use Prompt templates · UI L4 Foundation models Pre-trained neural weights · next-token prediction Fine-tuning · RLHF (one post-training method) L3 Infrastructure · serving Inference engineering · API keys · rate limits KV caching · request/worker routing L2 Chips (Pending) GPUs · TPUs · High-Bandwidth Memory (HBM) Tensor / matrix-multiplication units L1 Energy + facilities (Pending) Grid power · gigawatt-scale capacity · substations Data-centre facilities · liquid cooling FIVE LAYERS · ENERGY → APPLICATION