You are correct. The pricing model guarantees this. Pay per compute vs pay for uptime (during which you could have more compute for cheaper)