1. Why can't you use something like AWS Fargate and spin up the required number of VMs as needed to keep your running costs completely variable? 2. Similar to m

1. Why can't you use something like AWS Fargate and spin up the required number of VMs as needed to keep your running costs completely variable? 2. Similar to many other comments here, you should price your service based on value offered to the customer (i.e., how much $ they are likely to save based on some engineering cost model) as opposed to a cost-plus-margin model. 3. Once you have a critical mass of clients and tests running, you can then optimize your Fargate pool with spot instances. Starting with #1 will give you a floor for your running costs and if you tack your other costs (overheads) and a desired margin on top of that you will have a floor price. Then you just need to figure out the value you are providing (and you can A/B test that with prices higher than the floor price you obtain).