Blog

• AI's energy and water bill, and where hybrid models fit

  September 25, 2026 · Beyond attention, part 1

  What the public record says about the electricity, water, and carbon cost of AI, what model design has to do with it, and how much hybrid models can realistically help.

• What makes hybrid AI different from transformer AI

  September 25, 2026 · Beyond attention, part 2

  A plain language look at how attention layers and state space layers read the past, why pure state models struggle to recall, and why most new efficient models mix the two.

• Why hybrid AI costs less to serve

  September 25, 2026 · Beyond attention, part 3

  A walk through the memory arithmetic of serving language models, from the KV cache to batch size to cost per token, and what changes when most attention layers are replaced by a fixed size state.

• Samba, Mamba, and what hybrid models mean for chip design

  September 25, 2026 · Beyond attention, part 4

  How Mamba and Samba remember the past, why a bounded window makes per stream memory fixed, and what published research says that means for AI chips.