diff --git a/pandas/core/frame.py b/pandas/core/frame.py
index dbbe2c0751f3a..b8632a52f8988 100644
--- a/pandas/core/frame.py
+++ b/pandas/core/frame.py
@@ -4105,9 +4105,11 @@ def _set_item_mgr(self, key, value: ArrayLike) -> None:
         if len(self):
             self._check_setitem_copy()
 
-    def _iset_item(self, loc: int, value) -> None:
-        arraylike = self._sanitize_column(value)
-        self._iset_item_mgr(loc, arraylike, inplace=True)
+    def _iset_item(self, loc: int, value: Series) -> None:
+        # We are only called from _replace_columnwise which guarantees that
+        # no reindex is necessary
+        # TODO(CoW): Optimize to avoid copy here, but have ton track refs
+        self._iset_item_mgr(loc, value._values.copy(), inplace=True)
 
         # check if we are modifying a copy
         # try to set first as we want an invalid
diff --git a/pandas/tests/copy_view/test_replace.py b/pandas/tests/copy_view/test_replace.py
index c0b5498b839f5..dfb1caa4b2ffd 100644
--- a/pandas/tests/copy_view/test_replace.py
+++ b/pandas/tests/copy_view/test_replace.py
@@ -362,3 +362,14 @@ def test_replace_list_none_inplace_refs(using_copy_on_write):
         tm.assert_frame_equal(df_orig, view)
     else:
         assert np.shares_memory(arr, get_array(df, "a"))
+
+
+def test_replace_columnwise_no_op(using_copy_on_write):
+    df = DataFrame({"a": [1, 2, 3], "b": [1, 2, 3]})
+    df_orig = df.copy()
+    df2 = df.replace({"a": 10}, 100)
+    if using_copy_on_write:
+        # TODO(CoW): This should share memory
+        assert not np.shares_memory(get_array(df2, "a"), get_array(df, "a"))
+    df2.iloc[0, 0] = 100
+    tm.assert_frame_equal(df, df_orig)