One cabinet, 72 GPUs, as much power as a hundred homes. Stand with about three metres of clear floor ahead of you, then send it a question and follow a single token through it.
Tap any part of the rack to learn what it does. Turn off silent mode to hear the tokens sing.
A simplified visualization of AI inference: real providers route requests and assign hardware in many ways, and live answers come from a small open model on Cloudflare. Figures are approximate.