diff --git a/doc/source/whatsnew/v1.5.0.rst b/doc/source/whatsnew/v1.5.0.rst
index 0f5b4a16d2f01..2bb9905454e77 100644
--- a/doc/source/whatsnew/v1.5.0.rst
+++ b/doc/source/whatsnew/v1.5.0.rst
@@ -222,6 +222,7 @@ Datetimelike
 - Bug in :meth:`DataFrame.quantile` with datetime-like dtypes and no rows incorrectly returning ``float64`` dtype instead of retaining datetime-like dtype (:issue:`41544`)
 - Bug in :func:`to_datetime` with sequences of ``np.str_`` objects incorrectly raising (:issue:`32264`)
 - Bug in :class:`Timestamp` construction when passing datetime components as positional arguments and ``tzinfo`` as a keyword argument incorrectly raising (:issue:`31929`)
+- Bug in :meth:`Index.astype` when casting from object dtype to ``timedelta64[ns]`` dtype incorrectly casting ``np.datetime64("NaT")`` values to ``np.timedelta64("NaT")`` instead of raising (:issue:`45722`)
 -
 
 Timedelta
diff --git a/pandas/core/dtypes/astype.py b/pandas/core/dtypes/astype.py
index 7c7b7f3ba097f..11b147e207327 100644
--- a/pandas/core/dtypes/astype.py
+++ b/pandas/core/dtypes/astype.py
@@ -15,6 +15,7 @@
 import numpy as np
 
 from pandas._libs import lib
+from pandas._libs.tslibs.timedeltas import array_to_timedelta64
 from pandas._typing import (
     ArrayLike,
     DtypeObj,
@@ -139,7 +140,7 @@ def astype_nansafe(
     elif is_object_dtype(arr.dtype):
 
         # work around NumPy brokenness, #1987
-        if np.issubdtype(dtype.type, np.integer):
+        if dtype.kind in ["i", "u"]:
             return lib.astype_intsafe(arr, dtype)
 
         # if we have a datetime/timedelta array of objects
@@ -154,9 +155,10 @@ def astype_nansafe(
                 copy=copy,
             )
         elif is_timedelta64_dtype(dtype):
-            from pandas import to_timedelta
-
-            return astype_nansafe(to_timedelta(arr)._values, dtype, copy=copy)
+            # bc we know arr.dtype == object, this is equivalent to
+            #  `np.asarray(to_timedelta(arr))`, but using a lower-level API that
+            #  does not require a circular import.
+            return array_to_timedelta64(arr).view("m8[ns]").astype(dtype, copy=False)
 
     if dtype.name in ("datetime64", "timedelta64"):
         msg = (
diff --git a/pandas/tests/indexes/object/test_astype.py b/pandas/tests/indexes/object/test_astype.py
index 9bfc0c1312200..91e266e805868 100644
--- a/pandas/tests/indexes/object/test_astype.py
+++ b/pandas/tests/indexes/object/test_astype.py
@@ -1,4 +1,9 @@
-from pandas import Index
+import pytest
+
+from pandas import (
+    Index,
+    NaT,
+)
 import pandas._testing as tm
 
 
@@ -8,3 +13,12 @@ def test_astype_str_from_bytes():
     result = idx.astype(str)
     expected = Index(["あ", "a"], dtype="object")
     tm.assert_index_equal(result, expected)
+
+
+def test_astype_invalid_nas_to_tdt64_raises():
+    # GH#45722 don't cast np.datetime64 NaTs to timedelta64 NaT
+    idx = Index([NaT.asm8] * 2, dtype=object)
+
+    msg = r"Cannot cast Index to dtype timedelta64\[ns\]"
+    with pytest.raises(TypeError, match=msg):
+        idx.astype("m8[ns]")