diff --git a/changes/4554.misc.md b/changes/4554.misc.md
new file mode 100644
index 0000000000..a5a961b1c1
--- /dev/null
+++ b/changes/4554.misc.md
@@ -0,0 +1 @@
+Removed an unreachable scalar-conversion branch from `zarr.save_array`. Scalars are already rejected by the `NDArrayLike` check that precedes it, and mypy began reporting the branch as unreachable once numpy 2.5.4 annotated `numpy.isscalar` with `TypeIs`. A test now pins the `TypeError` that `zarr.save_array` raises for a scalar.
diff --git a/src/zarr/api/asynchronous.py b/src/zarr/api/asynchronous.py
index 476276fd82..8be7d32f7d 100644
--- a/src/zarr/api/asynchronous.py
+++ b/src/zarr/api/asynchronous.py
@@ -548,8 +548,6 @@ async def save_array(
 
     mode = kwargs.pop("mode", "a")
     store_path = await make_store_path(store, path=path, mode=mode, storage_options=storage_options)
-    if np.isscalar(arr):
-        arr = np.array(arr)
     shape = arr.shape
     chunks = getattr(arr, "chunks", None)  # for array-likes with chunks attribute
     overwrite = kwargs.pop("overwrite", None) or _infer_overwrite(mode)
diff --git a/tests/test_api.py b/tests/test_api.py
index 1d0c8f6498..e162ddfeea 100644
--- a/tests/test_api.py
+++ b/tests/test_api.py
@@ -958,6 +958,12 @@ def test_save_errors() -> None:
         zarr.save("data/example.zarr", a, mode="w")
 
 
+def test_save_array_rejects_scalar() -> None:
+    # numpy scalars lack __len__ and __setitem__, so they fail the NDArrayLike check
+    with pytest.raises(TypeError, match="arr argument must be numpy or other NDArrayLike array"):
+        save_array("data/example.zarr", np.float64(1.0))  # type: ignore[arg-type]
+
+
 def test_save_storage_options_is_not_an_array(tmp_path: Path) -> None:
     # `storage_options` is an option of `save`, not one of the arrays to save, so a
     # single array still lands as an array and not as a group holding `arr_0`.