diff --git a/pandas/core/frame.py b/pandas/core/frame.py
index 9d6af3c7b9917..29ab4d1f47170 100644
--- a/pandas/core/frame.py
+++ b/pandas/core/frame.py
@@ -8458,6 +8458,28 @@ def _arith_method_with_reindex(self, right: DataFrame, op) -> DataFrame:
         #  to avoid constructing two potentially large/sparse DataFrames
         join_columns = left.columns.join(right.columns, how="outer")
 
+        # GH#63288 Preserve ExtensionDtype (e.g. pyarrow) when reindexing
+        # introduces missing columns
+        from pandas.core.dtypes.base import ExtensionDtype
+
+        missing_cols = [c for c in join_columns if c not in result.columns]
+
+        if missing_cols:
+            for col in missing_cols:
+                src = left[col] if col in left.columns else right[col]
+
+                if isinstance(src.dtype, ExtensionDtype):
+                    # Create NA-filled Series with same ExtensionDtype
+                    fill = src.iloc[:0].reindex(result.index)
+                else:
+                    # Fallback to existing NumPy behavior
+                    fill = self._constructor_sliced(
+                        [np.nan] * len(result.index),
+                        index=result.index,
+                    )
+
+                result[col] = fill
+
         if result.columns.has_duplicates:
             # Avoid reindexing with a duplicate axis.
             # https://github.com/pandas-dev/pandas/issues/35194
diff --git a/pandas/tests/frame/test_arithmetic.py b/pandas/tests/frame/test_arithmetic.py
index 3981b0fc4e23a..030afa05367ef 100644
--- a/pandas/tests/frame/test_arithmetic.py
+++ b/pandas/tests/frame/test_arithmetic.py
@@ -2201,3 +2201,28 @@ def test_mixed_col_index_dtype(string_dtype_no_object):
     expected.columns = expected.columns.astype(string_dtype_no_object)
 
     tm.assert_frame_equal(result, expected)
+
+
+def test_arith_reindex_with_pyarrow_dtype():
+    # GH#63288 - Preserve pyarrow dtypes when reindexing introduces
+    # missing columns
+    pytest.importorskip("pyarrow")
+
+    df = DataFrame({"a": [1, 2, 3], "b": [4, 5, 6], "c": [7, 8, 9]})
+    df1 = df.iloc[:, :2].astype("int64[pyarrow]")  # columns: a, b
+    df2 = df.iloc[1:, 1:].astype("int64[pyarrow]")  # columns: b, c
+
+    result = df1 + df2
+    expected = DataFrame(
+        {
+            "a": pd.array([pd.NA, pd.NA, pd.NA], dtype="int64[pyarrow]"),
+            "b": pd.array([pd.NA, 10, 12], dtype="int64[pyarrow]"),
+            "c": pd.array([pd.NA, pd.NA, pd.NA], dtype="int64[pyarrow]"),
+        }
+    )
+    tm.assert_frame_equal(result, expected)
+
+    # Verify all columns preserved ExtensionDtype
+    assert str(result["a"].dtype) == "int64[pyarrow]"
+    assert str(result["b"].dtype) == "int64[pyarrow]"
+    assert str(result["c"].dtype) == "int64[pyarrow]"