diff --git a/controllers/vmware/virtualmachinegroup_reconciler.go b/controllers/vmware/virtualmachinegroup_reconciler.go
index d075debebe..ab303e4bd0 100644
--- a/controllers/vmware/virtualmachinegroup_reconciler.go
+++ b/controllers/vmware/virtualmachinegroup_reconciler.go
@@ -476,13 +476,23 @@ func shouldCreateVirtualMachineGroup(ctx context.Context, mds []clusterv1.Machin
 		currentVSphereMachineCount++
 	}
 
-	// If the number of workers VSphereMachines matches the number of expected replicas in the MachineDeployments,
-	// then all the VSphereMachines required for the initial placement decision do exist, then it is possible to create
-	// the VirtualMachineGroup.
-	if currentVSphereMachineCount != expectedVSphereMachineCount {
+	// Gate initial VMG creation on having enough worker VSphereMachines for the sum of MachineDeployment replicas.
+	// If current < expected, wait. Initial placement is meant to wait for enough VSphereMachines to be created to form a batch.
+	// When current > expected (for e.g. MachineDeployment surge during a rolling update):
+	// (1) New clusters with Node Auto Placement enabled: worker VSphereMachines are created in a batch, and all are
+	//     included together in the same initial placement, and membership are maintained afterward.
+	// (2) Existing clusters to adopt Node Auto Placement: CAPV can set per-MachineDeployment zone annotations on
+	//     the VMG based on the placement result of exsting MachineDeployments, and pins new VSphereMachines to the right failureDomain. Proceed includes
+	//     all current VSphereMachines, and membership are maintained afterward.
+	if currentVSphereMachineCount < expectedVSphereMachineCount {
 		log.Info(fmt.Sprintf("Waiting for VSphereMachines required for the initial placement (expected %d, current %d)", expectedVSphereMachineCount, currentVSphereMachineCount))
 		return false
 	}
+
+	if currentVSphereMachineCount > expectedVSphereMachineCount {
+		log.Info(fmt.Sprintf("More VSphereMachines than expected replicas; proceeding to create VirtualMachineGroup with all current VSphereMachines (expected %d, current %d)", expectedVSphereMachineCount, currentVSphereMachineCount))
+	}
+
 	return true
 }
 
diff --git a/controllers/vmware/virtualmachinegroup_reconciler_test.go b/controllers/vmware/virtualmachinegroup_reconciler_test.go
index 4a63181a63..5de3e87188 100644
--- a/controllers/vmware/virtualmachinegroup_reconciler_test.go
+++ b/controllers/vmware/virtualmachinegroup_reconciler_test.go
@@ -108,6 +108,22 @@ func Test_shouldCreateVirtualMachineGroup(t *testing.T) {
 			},
 			want: false, // tot replicas = 4, 3 VSphereMachine exist
 		},
+		{
+			name: "Should create a VMG when more VSphereMachines than expected replicas exist",
+			mds: []clusterv1.MachineDeployment{
+				*createMD("md1", "test-cluster", "", 2),
+				*createMD("md2", "test-cluster", "", 1),
+				*createMD("md3", "test-cluster", "zone1", 1),
+			},
+			vSphereMachines: []vmwarev1.VSphereMachine{
+				*createVSphereMachine("m1", "test-cluster", "md1"),
+				*createVSphereMachine("m2", "test-cluster", "md1"),
+				*createVSphereMachine("m3", "test-cluster", "md1"), // surge: one extra for md1
+				*createVSphereMachine("m4", "test-cluster", "md2"),
+				*createVSphereMachine("m5", "test-cluster", "md3"),
+			},
+			want: true, // tot replicas = 4, 5 VSphereMachines (current > expected)
+		},
 		{
 			name:            "Should not create a VMG there are no expected VSphereMachines",
 			mds:             []clusterv1.MachineDeployment{}, // No Machine deployments
@@ -1110,6 +1126,45 @@ func TestVirtualMachineGroupReconciler_ReconcileSequence(t *testing.T) {
 				},
 			},
 		},
+		{
+			name:    "VirtualMachineGroup should be created when more vSphereMachines than replicas exist",
+			cluster: clusterInitialized,
+			mds: []clusterv1.MachineDeployment{
+				*createMD("md1", clusterInitialized.Name, "zone1", 2),
+				*createMD("md2", clusterInitialized.Name, "", 1),
+			},
+			vSphereMachines: []vmwarev1.VSphereMachine{
+				*createVSphereMachine("m1", clusterInitialized.Name, "md1", withCustomNamingStrategy()),
+				*createVSphereMachine("m2", clusterInitialized.Name, "md1", withCustomNamingStrategy()),
+				*createVSphereMachine("m3", clusterInitialized.Name, "md1", withCustomNamingStrategy()), // surge: extra worker for md1
+				*createVSphereMachine("m4", clusterInitialized.Name, "md2"),
+			},
+			existingVMG: nil,
+			wantResult:  ctrl.Result{},
+			wantVMG: &vmoprvhub.VirtualMachineGroup{
+				ObjectMeta: metav1.ObjectMeta{
+					Namespace: clusterInitialized.Namespace,
+					Name:      clusterInitialized.Name,
+					UID:       types.UID("uid"),
+					Annotations: map[string]string{
+						fmt.Sprintf("%s/%s", vmoperator.ZoneAnnotationPrefix, "md1"): "zone1", // failureDomain for md1 is explicitly set by the user
+					},
+					// Not setting labels and ownerReferences for sake of simplicity
+				},
+				Spec: vmoprvhub.VirtualMachineGroupSpec{
+					BootOrder: []vmoprvhub.VirtualMachineGroupBootOrderGroup{
+						{
+							Members: []vmoprvhub.GroupMember{
+								{Name: "m1-vm", Kind: "VirtualMachine"},
+								{Name: "m2-vm", Kind: "VirtualMachine"},
+								{Name: "m3-vm", Kind: "VirtualMachine"},
+								{Name: "m4", Kind: "VirtualMachine"},
+							},
+						},
+					},
+				},
+			},
+		},
 
 		// During initial placement
 		{