Currently the only way a Worker node say deployed in Azure can talk to a Control Plane or Worker node  deployed in AWS is by deploying two different clusters (one in Azure and  the other in AWS), and then using RH service connect and the Submariner add-on to get the cluster in AWS connected with the cluster in Azure.  

This is similar to the  the native  Multi-cloud Kubernetes Cluster in AWS and Azure.
Deploying a new RHOCP Cluster for just adding a single Worker node with special resources  is an over-kill and waste of resources. 
In a true multi-cloud model, a Worker node must have the ability to talk to the control plane deployed in any of the CSP that Red Hat OpenShift  is supported on.  

For this example, let us assume that a customer:
a) Has subscription to both Azure and AWS
b) Has provisioned OpenShift v4.15 cluster with 3 Master nodes and couple of Worker Nodes provisioned in AWS
c) Has set-up network connection between Azure enclave and AWS enclave  using  mTLS
d) Has a specific AI workload that requires a GPU with significant amount of  GPU memory. 
Unfortunately  such resources are NOT available in AWS for immediate consumption and  and it takes 6 weeks of lead time. 
However such resources  are readily available in Azure. 
Customer wants to add just a Worker node with GPU in Azure and make that a part of AWS cluster for handling  the AI workload.