video1.0<iframe src="https://www.loom.com/embed/85f3a658b8a643a28a3646b48372e927" frameborder="0" width="1662" height="1246" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>12461662Loomhttps://www.loom.com12461662https://cdn.loom.com/sessions/thumbnails/85f3a658b8a643a28a3646b48372e927-13997fd35cf90d4f.gif183.361This Loom introduces SpotterPerch, a source of truth for choosing GPU spot instances with live pricing and eviction risk. Anna explains that AI teams overpay on GPUs because pricing and eviction rates are scattered or not published, and SpotterPerch can save about 60 to 80% versus on-demand. She demos Spottiker, a live AWS and Azure matrix showing spot price per GPU per region and color-coded eviction risk thresholds (green under 5%, orange 5 to 15%, red over 15%). The agent recommends spot options like T4 for cost-sensitive inference with low eviction risk and highlights that the system can estimate restart expectations (for example, about one restart). She also outlines plans to add workload guidance and roll out GCP next.