diff --git a/pandas/core/dtypes/cast.py b/pandas/core/dtypes/cast.py
index abcc60a15c641..3ff785edf2da1 100644
--- a/pandas/core/dtypes/cast.py
+++ b/pandas/core/dtypes/cast.py
@@ -30,7 +30,6 @@
     conversion,
     iNaT,
     ints_to_pydatetime,
-    ints_to_pytimedelta,
 )
 from pandas._libs.tslibs.timezones import tz_compare
 from pandas._typing import AnyArrayLike, ArrayLike, Dtype, DtypeObj, Scalar
@@ -987,15 +986,20 @@ def astype_nansafe(
     elif not isinstance(dtype, np.dtype):
         raise ValueError("dtype must be np.dtype or ExtensionDtype")
 
+    if arr.dtype.kind in ["m", "M"] and dtype == object:
+        # make sure we wrap in Timedelta/Timestamp, not timedelta/datetime
+        from pandas.core.construction import ensure_wrapped_if_datetimelike
+
+        arr = ensure_wrapped_if_datetimelike(arr)
+        return arr.astype(dtype, copy=copy)
+
     if issubclass(dtype.type, str):
         return lib.ensure_string_array(
             arr.ravel(), skipna=skipna, convert_na_value=False
         ).reshape(arr.shape)
 
     elif is_datetime64_dtype(arr):
-        if is_object_dtype(dtype):
-            return ints_to_pydatetime(arr.view(np.int64))
-        elif dtype == np.int64:
+        if dtype == np.int64:
             if isna(arr).any():
                 raise ValueError("Cannot convert NaT values to integer")
             return arr.view(dtype)
@@ -1007,9 +1011,7 @@ def astype_nansafe(
         raise TypeError(f"cannot astype a datetimelike from [{arr.dtype}] to [{dtype}]")
 
     elif is_timedelta64_dtype(arr):
-        if is_object_dtype(dtype):
-            return ints_to_pytimedelta(arr.view(np.int64))
-        elif dtype == np.int64:
+        if dtype == np.int64:
             if isna(arr).any():
                 raise ValueError("Cannot convert NaT values to integer")
             return arr.view(dtype)
diff --git a/pandas/core/internals/blocks.py b/pandas/core/internals/blocks.py
index 2630c07814bb2..b4e84779afa60 100644
--- a/pandas/core/internals/blocks.py
+++ b/pandas/core/internals/blocks.py
@@ -677,16 +677,6 @@ def _astype(self, dtype: DtypeObj, copy: bool) -> ArrayLike:
                     #  StringDtype, this matches arr.astype(dtype), xref GH#36153
                     values = arr._format_native_types(na_rep="NaT")
 
-            elif is_object_dtype(dtype):
-                if values.dtype.kind in ["m", "M"]:
-                    # Wrap in Timedelta/Timestamp
-                    arr = pd_array(values)
-                    values = arr.astype(object)
-                else:
-                    values = values.astype(object)
-                # We still need to go through astype_nansafe for
-                #  e.g. dtype = Sparse[object, 0]
-
             values = astype_nansafe(values, dtype, copy=True)
 
         return values
diff --git a/pandas/tests/dtypes/test_common.py b/pandas/tests/dtypes/test_common.py
index 0d0601aa542b4..7e8ce19d03ccd 100644
--- a/pandas/tests/dtypes/test_common.py
+++ b/pandas/tests/dtypes/test_common.py
@@ -754,6 +754,19 @@ def test_astype_object_preserves_datetime_na(from_type):
     assert isna(result)[0]
 
 
+@pytest.mark.parametrize("from_dtype", ["datetime64[ns]", "timedelta64[ns]"])
+def test_astype_object_boxes_timestamps_timedeltas(from_dtype):
+    arr = np.array([3600], dtype=from_dtype)
+    result = astype_nansafe(arr, dtype=np.dtype("object"))
+
+    assert result.dtype == object
+
+    if from_dtype == "datetime64[ns]":
+        assert isinstance(result[0], pd.Timestamp)
+    else:
+        assert isinstance(result[0], pd.Timedelta)
+
+
 def test_validate_allhashable():
     assert com.validate_all_hashable(1, "a") is None
 
diff --git a/pandas/tests/series/test_constructors.py b/pandas/tests/series/test_constructors.py
index 5b13091470b09..8d99467e5285a 100644
--- a/pandas/tests/series/test_constructors.py
+++ b/pandas/tests/series/test_constructors.py
@@ -24,6 +24,7 @@
     Period,
     RangeIndex,
     Series,
+    Timedelta,
     Timestamp,
     date_range,
     isna,
@@ -1513,6 +1514,22 @@ def test_constructor_sparse_datetime64(self, values):
         expected = Series(arr)
         tm.assert_series_equal(result, expected)
 
+    def test_constructor_datetimelike_to_sparse_object(self):
+        arr = np.arange(3, dtype="i8").view("M8[D]").astype("M8[ns]")
+        arr2 = arr.view("m8[ns]")
+
+        ser = Series(arr, dtype="Sparse[object]")
+        assert isinstance(ser[0], Timestamp)
+
+        ser = Series(arr).astype("Sparse[object]")
+        assert isinstance(ser[0], Timestamp)
+
+        ser2 = Series(arr2, dtype="Sparse[object]")
+        assert isinstance(ser2[0], Timedelta)
+
+        ser2 = Series(arr2).astype("Sparse[object]")
+        assert isinstance(ser2[0], Timedelta)
+
     def test_construction_from_ordered_collection(self):
         # https://github.com/pandas-dev/pandas/issues/36044
         result = Series({"a": 1, "b": 2}.keys())