diff --git a/src/nemotron/data_prep/utils/splits.py b/src/nemotron/data_prep/utils/splits.py
index 2b8cd88b8..f3ae5dfda 100644
--- a/src/nemotron/data_prep/utils/splits.py
+++ b/src/nemotron/data_prep/utils/splits.py
@@ -175,8 +175,12 @@ def realize_packed_shards_into_split_dirs(
                 logger.warning(f"Shard file not found: {parquet_path_str}")
                 continue
 
-            # Create symlink in split dir
-            link_path = split_dir / parquet_path.name
+            # Prefix symlink with dataset name to avoid collisions: multiple
+            # datasets produce identically-named shards (shard_000000.parquet, etc.),
+            # so bare filenames cause later datasets to overwrite earlier ones.
+            # Path convention: .../datasets/{dataset_name}/{plan_hash}/shard_NNNNNN.parquet
+            dataset_name = parquet_path.parent.parent.name
+            link_path = split_dir / f"{dataset_name}_{parquet_path.name}"
 
             if link_path.exists() or link_path.is_symlink():
                 # Remove existing link/file to update