diff --git a/source/blender/draw/intern/draw_instance_data.c b/source/blender/draw/intern/draw_instance_data.c
index 9695643d707..dca112c85c6 100644
--- a/source/blender/draw/intern/draw_instance_data.c
+++ b/source/blender/draw/intern/draw_instance_data.c
@@ -102,6 +102,13 @@ static ListBase g_idatalists = {NULL, NULL};
 
 static void instance_batch_free(GPUBatch *batch, void *UNUSED(user_data))
 {
+	if (batch->verts[0] == NULL) {
+		/** XXX This is a false positive case.
+		 * The batch has been requested but not init yet
+		 * and there is a chance that it might become init.
+		 **/
+		return;
+	}
 	/* Free all batches that have the same key before they are reused. */
 	/* TODO: Make it thread safe! Batch freeing can happen from another thread. */
 	/* XXX we need to iterate over all idatalists unless we make some smart
@@ -194,12 +201,11 @@ void DRW_instancing_buffer_request(
 	/* Create the batch. */
 	ibuf = chunk->ibufs + new_id;
 	ibuf->vert = *r_vert = GPU_vertbuf_create_with_format_ex(format, GPU_USAGE_DYNAMIC);
-	ibuf->batch = *r_batch = GPU_batch_duplicate(instance);
+	ibuf->batch = *r_batch = MEM_callocN(sizeof(GPUBatch), "GPUBatch");
 	ibuf->format = format;
 	ibuf->shgroup = shgroup;
 	ibuf->instance = instance;
 	GPU_vertbuf_data_alloc(*r_vert, BUFFER_VERTS_CHUNK);
-	GPU_batch_instbuf_set(ibuf->batch, ibuf->vert, false);
 	/* Make sure to free this ibuf if the instance batch gets free. */
 	GPU_batch_callback_free_set(instance, &instance_batch_free, NULL);
 }
@@ -253,6 +259,9 @@ void DRW_instance_buffer_finish(DRWInstanceDataList *idatalist)
 				GPU_vertbuf_data_resize(ibuf->vert, size);
 			}
 			GPU_vertbuf_use(ibuf->vert); /* Send data. */
+			/* Setup batch now that we are sure ibuf->instance is setup. */
+			GPU_batch_copy(ibuf->batch, ibuf->instance);
+			GPU_batch_instbuf_set(ibuf->batch, ibuf->vert, false);
 			ibuf->shgroup = NULL; /* Set as non used for the next round. */
 		}
 		else {
diff --git a/source/blender/gpu/GPU_batch.h b/source/blender/gpu/GPU_batch.h
index c77f0707ee5..f4911d15e56 100644
--- a/source/blender/gpu/GPU_batch.h
+++ b/source/blender/gpu/GPU_batch.h
@@ -100,7 +100,7 @@ enum {
 
 GPUBatch *GPU_batch_create_ex(GPUPrimType, GPUVertBuf *, GPUIndexBuf *, uint owns_flag);
 void GPU_batch_init_ex(GPUBatch *, GPUPrimType, GPUVertBuf *, GPUIndexBuf *, uint owns_flag);
-GPUBatch *GPU_batch_duplicate(GPUBatch *batch_src);
+void GPU_batch_copy(GPUBatch *batch_dst, GPUBatch *batch_src);
 
 #define GPU_batch_create(prim, verts, elem) \
 	GPU_batch_create_ex(prim, verts, elem, 0)
diff --git a/source/blender/gpu/intern/gpu_batch.c b/source/blender/gpu/intern/gpu_batch.c
index d70752e9b2f..a2184df229f 100644
--- a/source/blender/gpu/intern/gpu_batch.c
+++ b/source/blender/gpu/intern/gpu_batch.c
@@ -114,19 +114,22 @@ void GPU_batch_init_ex(
 }
 
 /* This will share the VBOs with the new batch. */
-GPUBatch *GPU_batch_duplicate(GPUBatch *batch_src)
+void GPU_batch_copy(GPUBatch *batch_dst, GPUBatch *batch_src)
 {
-	GPUBatch *batch = GPU_batch_create_ex(GPU_PRIM_POINTS, batch_src->verts[0], batch_src->elem, 0);
+	GPU_batch_init_ex(batch_dst, GPU_PRIM_POINTS, batch_src->verts[0], batch_src->elem, 0);
 
-	batch->gl_prim_type = batch_src->gl_prim_type;
+	batch_dst->gl_prim_type = batch_src->gl_prim_type;
 	for (int v = 1; v < GPU_BATCH_VBO_MAX_LEN; ++v) {
-		batch->verts[v] = batch_src->verts[v];
+		batch_dst->verts[v] = batch_src->verts[v];
 	}
-	return batch;
 }
 
 void GPU_batch_clear(GPUBatch *batch)
 {
+	if (batch->free_callback) {
+		batch->free_callback(batch, batch->callback_data);
+	}
+
 	if (batch->owns_flag & GPU_BATCH_OWNS_INDEX) {
 		GPU_indexbuf_discard(batch->elem);
 	}
@@ -144,10 +147,6 @@ void GPU_batch_clear(GPUBatch *batch)
 		}
 	}
 	GPU_batch_vao_cache_clear(batch);
-
-	if (batch->free_callback) {
-		batch->free_callback(batch, batch->callback_data);
-	}
 }
 
 void GPU_batch_discard(GPUBatch *batch)