Questions to ask a serverless GPU provider in 2026

Start with your traffic

Steady traffic suits reserved nodes. Spiky or unpredictable traffic suits per-second billing. Most products have both, so ask whether a provider can mix them.

Questions worth asking

How long is a cold start at P95 with my image? Which regions can I pin to? What happens to open connections during a deploy? Is billing per second, per minute or per hour?