diff --git a/docs/zalando.org_elasticsearchdatasets.yaml b/docs/zalando.org_elasticsearchdatasets.yaml
index 86e8aa82..b6dc390d 100644
--- a/docs/zalando.org_elasticsearchdatasets.yaml
+++ b/docs/zalando.org_elasticsearchdatasets.yaml
@@ -102,6 +102,10 @@ spec:
                     format: int64
                     minimum: 0
                     type: integer
+                  scaleDownDiskUsagePercentBoundary:
+                    format: int32
+                    minimum: 0
+                    type: integer
                   scaleDownThresholdDurationSeconds:
                     format: int64
                     minimum: 0
@@ -114,6 +118,10 @@ spec:
                     format: int64
                     minimum: 0
                     type: integer
+                  scaleUpDiskUsagePercentBoundary:
+                    format: int32
+                    minimum: 0
+                    type: integer
                   scaleUpThresholdDurationSeconds:
                     format: int64
                     minimum: 0
diff --git a/docs/zalando.org_elasticsearchdatasets_v1beta1.yaml b/docs/zalando.org_elasticsearchdatasets_v1beta1.yaml
index ab73c012..071c9a90 100644
--- a/docs/zalando.org_elasticsearchdatasets_v1beta1.yaml
+++ b/docs/zalando.org_elasticsearchdatasets_v1beta1.yaml
@@ -103,6 +103,10 @@ spec:
                   format: int64
                   minimum: 0
                   type: integer
+                scaleDownDiskUsagePercentBoundary:
+                  format: int32
+                  minimum: 0
+                  type: integer
                 scaleDownThresholdDurationSeconds:
                   format: int64
                   minimum: 0
@@ -115,6 +119,10 @@ spec:
                   format: int64
                   minimum: 0
                   type: integer
+                scaleUpDiskUsagePercentBoundary:
+                  format: int32
+                  minimum: 0
+                  type: integer
                 scaleUpThresholdDurationSeconds:
                   format: int64
                   minimum: 0
diff --git a/operator/autoscaler.go b/operator/autoscaler.go
index 5b473140..7a854cdf 100644
--- a/operator/autoscaler.go
+++ b/operator/autoscaler.go
@@ -126,6 +126,23 @@ func (as *AutoScaler) scalingHint() ScalingDirection {
 			as.logger.Info("Not scaling up, currently in cool-down period.")
 		}
 	}
+
+	// check if disk usage is below threshold
+	// TODO(Radu) is there a way to avoid calling _cat/nodes all the time? maybe just once per loop?
+	// or just store it somewhere with a timeout?
+	esNodes, err := as.esClient.GetNodes()
+	if err != nil {
+		as.logger.Errorf("Can't get ES Nodes: %v", err)
+	} else {
+		maxDiskUsagePercent := as.getMaxDiskUsage(esNodes)
+		scaleDownDiskUsagePercentBoundary := float64(scaling.ScaleDownDiskUsagePercentBoundary)
+		if maxDiskUsagePercent < scaleDownDiskUsagePercentBoundary {
+			as.logger.Debugf("Scaling hint DOWN because max disk usage %.2f is lower than the threshold %.2f",
+				maxDiskUsagePercent, scaleDownDiskUsagePercentBoundary)
+			return DOWN
+		}
+	}
+
 	return NONE
 }
 
@@ -268,6 +285,30 @@ func (as *AutoScaler) scaleUpOrDown(esIndices map[string]ESIndex, scalingHint Sc
 		}
 	}
 
+	// independent of scaling direction: if we run out of disk, we try to scale up
+	esNodes, err := as.esClient.GetNodes()
+	if err != nil {
+		as.logger.Errorf("Can't get ES Nodes: %v", err)
+	} else {
+		maxDiskUsagePercent := as.getMaxDiskUsage(esNodes)
+		scaleUpDiskUsagePercentBoundary := float64(scalingSpec.ScaleUpDiskUsagePercentBoundary)
+		if maxDiskUsagePercent > scaleUpDiskUsagePercentBoundary {
+			// TODO(Radu) compute how many nodes we need instead of increasing by one?
+			newDesiredNodeReplicas := as.ensureBoundsNodeReplicas(currentDesiredNodeReplicas + 1)
+
+			scalingMessage := fmt.Sprintf("Scaling up to %d nodes because %.2f is higher than %.2f",
+				newDesiredNodeReplicas, maxDiskUsagePercent, scaleUpDiskUsagePercentBoundary)
+
+			as.logger.Debug(scalingMessage)
+
+			return &ScalingOperation{
+				ScalingDirection: UP,
+				NodeReplicas:     &newDesiredNodeReplicas,
+				Description:      scalingMessage,
+			}
+		}
+	}
+
 	// independent of the scaling direction: in case there are indices with < MinIndexReplicas, we try to scale these indices.
 	if len(newDesiredIndexReplicas) > 0 {
 		return &ScalingOperation{
diff --git a/pkg/apis/zalando.org/v1/types.go b/pkg/apis/zalando.org/v1/types.go
index 1aa50315..d16dccf7 100644
--- a/pkg/apis/zalando.org/v1/types.go
+++ b/pkg/apis/zalando.org/v1/types.go
@@ -155,6 +155,9 @@ type ElasticsearchDataSetScaling struct {
 	ScaleUpCPUBoundary int32 `json:"scaleUpCPUBoundary"`
 	// +kubebuilder:validation:Minimum=0
 	// +optional
+	ScaleUpDiskUsagePercentBoundary int32 `json:"scaleUpDiskUsagePercentBoundary"`
+	// +kubebuilder:validation:Minimum=0
+	// +optional
 	ScaleUpThresholdDurationSeconds int64 `json:"scaleUpThresholdDurationSeconds"`
 	// +kubebuilder:validation:Minimum=0
 	// +optional
@@ -164,6 +167,9 @@ type ElasticsearchDataSetScaling struct {
 	ScaleDownCPUBoundary int32 `json:"scaleDownCPUBoundary"`
 	// +kubebuilder:validation:Minimum=0
 	// +optional
+	ScaleDownDiskUsagePercentBoundary int32 `json:"scaleDownDiskUsagePercentBoundary"`
+	// +kubebuilder:validation:Minimum=0
+	// +optional
 	ScaleDownThresholdDurationSeconds int64 `json:"scaleDownThresholdDurationSeconds"`
 	// +kubebuilder:validation:Minimum=0
 	// +optional