diff --git a/pandas/core/arrays/datetimelike.py b/pandas/core/arrays/datetimelike.py
index 79ecf8620c70c..9eba4f85f23f5 100644
--- a/pandas/core/arrays/datetimelike.py
+++ b/pandas/core/arrays/datetimelike.py
@@ -601,6 +601,10 @@ def _validate_listlike(self, value, allow_object: bool = False):
         if isinstance(value, type(self)):
             return value
 
+        if isinstance(value, list) and len(value) == 0:
+            # We treat empty list as our own dtype.
+            return type(self)._from_sequence([], dtype=self.dtype)
+
         # Do type inference if necessary up front
         # e.g. we passed PeriodIndex.values and got an ndarray of Periods
         value = array(value)
diff --git a/pandas/core/internals/blocks.py b/pandas/core/internals/blocks.py
index 6500cfcd9ea5a..8752224356f61 100644
--- a/pandas/core/internals/blocks.py
+++ b/pandas/core/internals/blocks.py
@@ -1,4 +1,4 @@
-from datetime import datetime, timedelta
+from datetime import timedelta
 import inspect
 import re
 from typing import TYPE_CHECKING, Any, List, Optional, Type, Union, cast
@@ -18,7 +18,6 @@
 )
 from pandas._libs.internals import BlockPlacement
 from pandas._libs.tslibs import conversion
-from pandas._libs.tslibs.timezones import tz_compare
 from pandas._typing import ArrayLike, DtypeObj, Scalar, Shape
 from pandas.util._validators import validate_bool_kwarg
 
@@ -2190,6 +2189,15 @@ def to_native_types(self, na_rep="NaT", **kwargs):
         result = arr._format_native_types(na_rep=na_rep, **kwargs)
         return self.make_block(result)
 
+    def _can_hold_element(self, element: Any) -> bool:
+        arr = self.array_values()
+
+        try:
+            arr._validate_setitem_value(element)
+            return True
+        except (TypeError, ValueError):
+            return False
+
 
 class DatetimeBlock(DatetimeLikeBlockMixin):
     __slots__ = ()
@@ -2222,32 +2230,6 @@ def _maybe_coerce_values(self, values):
         assert isinstance(values, np.ndarray), type(values)
         return values
 
-    def _can_hold_element(self, element: Any) -> bool:
-        tipo = maybe_infer_dtype_type(element)
-        if tipo is not None:
-            if isinstance(element, list) and len(element) == 0:
-                # Following DatetimeArray._validate_setitem_value
-                #  convention, we treat this as object-dtype
-                #  (even though tipo is float64)
-                return True
-
-            elif self.is_datetimetz:
-                # require exact match, since non-nano does not exist
-                return is_dtype_equal(tipo, self.dtype) or is_valid_nat_for_dtype(
-                    element, self.dtype
-                )
-
-            # GH#27419 if we get a non-nano datetime64 object
-            return is_datetime64_dtype(tipo)
-        elif element is NaT:
-            return True
-        elif isinstance(element, datetime):
-            if self.is_datetimetz:
-                return tz_compare(element.tzinfo, self.dtype.tz)
-            return element.tzinfo is None
-
-        return is_valid_nat_for_dtype(element, self.dtype)
-
     def set_inplace(self, locs, values):
         """
         See Block.set.__doc__
diff --git a/pandas/tests/indexing/test_indexing.py b/pandas/tests/indexing/test_indexing.py
index 5cd429837a127..05d9d1a9bd74f 100644
--- a/pandas/tests/indexing/test_indexing.py
+++ b/pandas/tests/indexing/test_indexing.py
@@ -10,7 +10,7 @@
 from pandas.core.dtypes.common import is_float_dtype, is_integer_dtype
 
 import pandas as pd
-from pandas import DataFrame, Index, NaT, Series
+from pandas import DataFrame, Index, NaT, Series, date_range
 import pandas._testing as tm
 from pandas.core.indexing import maybe_numeric_slice, non_reducing_slice
 from pandas.tests.indexing.common import _mklbl
@@ -966,6 +966,56 @@ def test_none_coercion_mixed_dtypes(self):
         tm.assert_frame_equal(start_dataframe, exp)
 
 
+class TestDatetimelikeCoercion:
+    @pytest.mark.parametrize("indexer", [setitem, loc, iloc])
+    def test_setitem_dt64_string_scalar(self, tz_naive_fixture, indexer):
+        # dispatching _can_hold_element to underling DatetimeArray
+        # TODO(EA2D) use tz_naive_fixture once DatetimeBlock is backed by DTA
+        tz = tz_naive_fixture
+
+        dti = date_range("2016-01-01", periods=3, tz=tz)
+        ser = Series(dti)
+
+        values = ser._values
+
+        indexer(ser)[0] = "2018-01-01"
+
+        if tz is None:
+            # TODO(EA2D): we can make this no-copy in tz-naive case too
+            assert ser.dtype == dti.dtype
+        else:
+            assert ser._values is values
+
+    @pytest.mark.parametrize("box", [list, np.array, pd.array])
+    @pytest.mark.parametrize(
+        "key", [[0, 1], slice(0, 2), np.array([True, True, False])]
+    )
+    @pytest.mark.parametrize("indexer", [setitem, loc, iloc])
+    def test_setitem_dt64_string_values(self, tz_naive_fixture, indexer, key, box):
+        # dispatching _can_hold_element to underling DatetimeArray
+        # TODO(EA2D) use tz_naive_fixture once DatetimeBlock is backed by DTA
+        tz = tz_naive_fixture
+
+        if isinstance(key, slice) and indexer is loc:
+            key = slice(0, 1)
+
+        dti = date_range("2016-01-01", periods=3, tz=tz)
+        ser = Series(dti)
+
+        values = ser._values
+
+        newvals = box(["2019-01-01", "2010-01-02"])
+        values._validate_setitem_value(newvals)
+
+        indexer(ser)[key] = newvals
+
+        if tz is None:
+            # TODO(EA2D): we can make this no-copy in tz-naive case too
+            assert ser.dtype == dti.dtype
+        else:
+            assert ser._values is values
+
+
 def test_extension_array_cross_section():
     # A cross-section of a homogeneous EA should be an EA
     df = DataFrame(