Compute
Cold Start
The delay when a serverless function or idle container has to spin up from scratch before handling its first request.
A cold start is the lag you pay when a serverless function or scaled-to-zero container has to boot before it can respond. The platform spins up a fresh runtime, loads your code, and only then handles the request. Depending on the language and package size, that's anywhere from tens of milliseconds to a few seconds.
Warm invocations reuse an already-running instance and skip the penalty, so cold starts mostly bite the first request after a quiet period, or the extra instances a sudden spike forces online. Heavier runtimes and large dependency bundles make them worse. Cloudflare Workers minimize the problem with a lightweight isolate model. Traditional container-based functions feel it more.
For user-facing paths where latency matters, cold starts are a real design constraint. Teams work around them by keeping functions warm, trimming dependencies, or choosing an always-on service instead of scale-to-zero for the hot path.