Cloud savings

Paying for what the product needs and nothing it doesn't. On a first look we nearly always find a test server someone forgot, a database two sizes too big and logs on expensive storage. Once that's fixed, tags and alerts stop the number drifting back.

Start where it's easiest. Delete unattached volumes and old snapshots, switch off anything idle, shrink servers that sit at low CPU and stop test environments running overnight. If something runs around the clock, a Savings Plan will usually make it cheaper. Cross-region data transfer is worth a look too, since it hides well.

It's sizing servers from what they really use. Pull a few weeks of CPU and memory figures, pick the size that fits with a bit of room, and change it. Most instances were chosen in a hurry at launch and never revisited. Databases and caches get the same treatment.

Both swap a one- or three-year commitment for a lower hourly price. Reserved Instances apply to a specific instance setup. Savings Plans are a commitment to an hourly spend, and Compute Savings Plans also apply to Fargate and Lambda, which gives you room to change instance types later.

Through AWS Activate. You can apply yourself, but the bigger tiers usually come through an accelerator, incubator or investor that's an Activate provider. Amounts and conditions change, so go by what the Activate page says today.

Google Cloud, Microsoft and a number of AI and dev tool companies run programmes as well, each with its own rules. Expect to need a registered early-stage company with a real product and little history on their platform. Larger tiers often want investor or accelerator backing. Credits expire, so decide what they'll cover first.