Stealing data living on one VM directly from another VM is not a realistic attack scenario currently. While there are side channels which allow a small information leakage it is currently more a theoretical and expensive thing to exfiltrate data that way.
Much more likely are attacks directly against the VM containing the data (i.e. SQL injection, SSH brute force....). There is nothing special VM or cloud related with these attacks, i.e. once you have system privileges or even kernel privileges you can access the memory of any process.
Less likely but still realistic are attacks against the hypervisor, like breaking out of a neighbor VM or attacking the system where the hypervisor runs on. Once you got system or kernel privileges on the hypervisor this way you get full access to all VM's in control of the hypervisor. This includes full access to the memory of these VM.