There is no VMConnect capability for console access to the VM as there is with Hyper-V.

So in the situation where I lost connectivity, by misconfiguration, or system failure, is there only one very time and bandwith consuming way to fix it - download the VHDs on-premise, boot it in Hyper-V and VMConnect to repair VM, then upload all GBs back to the Azure...

I requesting console access, also with ability to mount some image for repair and boot from it (as we do it in standard environment).

When we try to create a virtual machine from a custom image the VHD's for the new virtual machine are created in the same storage account as the custom image. This is not ideal because most customers have different environments in different subscriptions so they are forced to copy the image into storage account in each subscriptions. It is not easy to copy the images between the storage accounts either.

Customers have been asking for this every since IAAS went into preview. There are many customer migrating their on premise applications to Azure IAAS. Some of these applications have hard references to D: drive. It is possible to manually move temp drive from D: to another drive letter but it is hard to automate this. A simple fix for this is to move this temp drive to a Z: drive. A better fix may be to allow Azure Portal and PowerShell to select the drive letter for the temp drive during VM creation process.

This will simplify migrations.

Ability to move virtual machines to a different region

Provide a way to group VMs into production / staging environment. Somehow like the production / staging environments of cloud services.

Is there a way that CAD processing intensive applications can use worker roles that can access GPUs in the same way as CPUs.
That would mean to extend the Azure HW-Spec in some way to provide Multi-GPU-Cores in additon to Multi-CPU-Cores.

This would save a lot of time for users that manage multiple subscriptions as it can take a very long time to upload images. This is already supported in Amazon AWS. It would be great if you could share VM images across regions as well.

Currently Azure doesn't provide SLA for single VM. In each month, Azure will have a scheduled maintenance. During the maintenance window, all the VM will be rebooted. This approach brings great trouble to those applications which rely on memory heavily, such as online game, desktop app, etc. Normal HA solution, such as db persistent or sharing memory can't help here. A better approach is to let customer select maintenance window, or initiate reboot by customer himself. So the maintenance is controllable at customer side.

The ability to move a existing VM into a existing Virtual Network.

I have 10 plus developers in my group. All have MSDN licenses and we are exploring the use of Azure to do our SharePoint development. In order to setup SharePoint properly you need 3 Virtual Machines, a DC, a SQL Server and a SharePoint server. It would be nice if between the 10 developers they could share a DC and SQL Server and only have a SharePoint server unique to them. Of course, doing this in a secure fashion is key.

In order to develop graphic intensive applications or run graphic intensive applications more powerful video processing is required.

Right now you cannot resize your VHD-disks once attached to a VM directly from the portal. Edit should be possible to at least increase the size of VHD-disks on demand.

Both AWS and GCE allow runtime meta data calls from inside a VM , both of them using a similar mechanism of a call to http://169.254.169.254/something/

GCE:
https://developers.google.com/compute/docs/metadata

AWS: (and openstack and eucalyptus)
http://docs.aws.amazon.com/AWSEC2/latest/UserGuide/AESDG-chapter-instancedata.html

These provide ways to have the instance do introspection on where it is, what account it is in, info about its VLANs, ip's, etc.. WITHOUT needing the whole API tool chain and importantly without have credentials to the account in the instance.

A support for uploading VHDX images would be great. With that it wouldn't be necessary anymore to migrate a VHDX to VHD before.

It is a great feature of Azure to stop costs if the VM is stopped. But shutdown the VM needs close all the running programs and applications in the current session. It will be great if the VM can have a sleep mode which can stop all the costs (or partially). So users can quickly resume the work in the new session without restarting the VM.

Add the ability to rename a VM. The VM name is what appears in Azure, not the DNS name.

Changing a VM's name should not affect the VM in any way. If you need a unique identifier for a VM that does not change, I suggest using a GUID rather than the VM's name in Azure.