Video AI Cost Optimization — 25-Lever Checklist

One page: the 25 cost levers for AI in a video product, grouped into five families — process less video (frame sampling, downscale, crop, event-gating, prompt/output trimming); run a smaller model (routing, cascade, small open model, distillation, quantization); serve more efficiently (continuous batching, KV-cache reuse, speculative decoding, a purpose-built runtime, right-sizing); pay less per unit (batch API at 50% off, prompt caching at 90% off, semantic caching, spot GPUs, reserved capacity); and architect & govern (serverless scale-to-zero, right provider + egress control, hybrid edge+cloud, precompute once, per-feature cost budgets). Includes the worked-example waterfall ($135K → $4.7K, ~29×) and the pre-optimization questions. The lead-magnet asset for Phase 8.