Is it possible to create a VM that survives outage of host?

What you describe is certainly possible.  Tandem Computers, Inc., now a division within Hewlett Packard, developed a redundant, fault-tolerant computer system called NonStop.  It was used in banks, stock exchanges, telephone switching centers, and other similar commercial transaction processing applications requiring maximum uptime and zero data loss.  Representatives claimed that the only ways a NonStop system would crash would be massive physical damage (e.g., fire, flood, or earthquake) or a power failure.  (And of course those risks could be mitigated with distributed data centers and on-site backup electrical generators.)  NonStop would laugh at the failure of a single component (CPU, storage device, bus, etc.)

If you could virtualize a cluster of NonStop processors, using multiple hosts, you should be able to achieve the same non-stop availability characteristics as the original Tandem system.  This, per se, is easier said than done — NonStop used a proprietary CPU, and, AFAIK, the operating system is not available for free download.  But, in principle, it should be possible to do something like this.