Two acts · 17 chapters · 7–8 minutes

Inside the Machine

How a prompt becomes tokens.

A question leaves your screen. Somewhere, a room full of machines turns it into an answer. Follow the journey from a datacenter aisle to the arithmetic inside a chip.

ACT I / THE MACHINEBefore you press enter.

Power, memory, and a model already waiting.

ACT II / YOUR REQUESTOne question. Many chips.

Read the prompt, choose experts, stream an answer.

Drag to look around · scroll to zoom · space to pause · ← → to skip

INSIDE THE MACHINE / FIELD NOTES

How a prompt becomes tokens.

One worked example, not a universal deployment: Kimi K3 on a GB300 NVL72 rack. Geometry and timing are illustrative; estimates carry ≈.

Primary references: Moonshot model card · NVIDIA rack specifications