Inference clouds focus on serving models rather than creating the underlying model family. They compete on hardware, model optimization, cold starts, throughput, deployment control, and support for custom weights.

Some expose a large serverless catalog; others are closer to deployment infrastructure for a model your team already chose.