diff --git a/pandas/core/arrays/arrow/array.py b/pandas/core/arrays/arrow/array.py
index de1fb0ec5d4d55..7183eb502ce5e3 100644
--- a/pandas/core/arrays/arrow/array.py
+++ b/pandas/core/arrays/arrow/array.py
@@ -1322,16 +1322,12 @@ def to_numpy(
             copy = False
 
         if pa.types.is_timestamp(pa_type) or pa.types.is_duration(pa_type):
-            result = data._maybe_convert_datelike_array()
-            if (pa.types.is_timestamp(pa_type) and pa_type.tz is not None) or (
-                dtype is not None and dtype.kind == "O"
-            ):
-                dtype = object
-            else:
-                # GH 55997
-                dtype = None
-                na_value = pa_type.to_pandas_dtype().type("nat", pa_type.unit)
-            result = result.to_numpy(dtype=dtype, na_value=na_value)
+            # GH 55997
+            if dtype != object and na_value is self.dtype.na_value:
+                na_value = lib.no_default
+            result = data._maybe_convert_datelike_array().to_numpy(
+                dtype=dtype, na_value=na_value
+            )
         elif pa.types.is_time(pa_type) or pa.types.is_date(pa_type):
             # convert to list of python datetime.time objects before
             # wrapping in ndarray
diff --git a/pandas/tests/extension/test_arrow.py b/pandas/tests/extension/test_arrow.py
index 28ee5446d4b5c1..7d87c366f05668 100644
--- a/pandas/tests/extension/test_arrow.py
+++ b/pandas/tests/extension/test_arrow.py
@@ -3030,7 +3030,10 @@ def test_to_numpy_temporal(pa_type, dtype):
         value = pd.Timestamp(1, unit=pa_type.unit, tz=pa_type.tz).as_unit(pa_type.unit)
 
     if dtype == object or (pa.types.is_timestamp(pa_type) and pa_type.tz is not None):
-        na = pd.NA
+        if dtype == object:
+            na = pd.NA
+        else:
+            na = pd.NaT
         expected = np.array([value, na], dtype=object)
         assert result[0].unit == value.unit
     else:
@@ -3142,3 +3145,11 @@ def test_string_to_time_parsing_cast():
         ArrowExtensionArray(pa.array([time(11, 41, 43, 76160)], from_pandas=True))
     )
     tm.assert_series_equal(result, expected)
+
+
+def test_to_numpy_timestamp_to_int():
+    # GH 55997
+    ser = pd.Series(["2020-01-01 04:30:00"], dtype="timestamp[ns][pyarrow]")
+    result = ser.to_numpy(dtype=np.int64)
+    expected = np.array([1577853000000000000])
+    tm.assert_numpy_array_equal(result, expected)