A concern when architecting for the cloud is dealing with elasticity. The pricing of Windows Azure on compute / VM roles had an affect on how you dealt with scaling for demand. Previously if you spun up an instance for 20 minutes to temporarily deal with additional load, you would be charged for a full hour. This lead to a 1 hour granularity when evaluating when you should scale. Your best bet was to evaluate on the hour if you need to spin more up, and then 10 mins before the hour was up, evaluate if this is a good time to take them down again.

Microsoft announced updated pricing, where they will now be charging per minute.

http://blogs.msdn.com/b/windowsazure/archive/2013/06/03/announcing-new-offers-and-services-on-windows-azure.aspx 
http://blogs.technet.com/b/firehose/archive/2013/06/03/windows-azure-introduces-pay-by-the-minute-feature.aspx

This is great news, as it allows finer granularity in how services are scaled. I’m sure a lot of people will be excited about this.