diff --git a/xarray/core/duck_array_ops.py b/xarray/core/duck_array_ops.py
index 2bf9d25a1e6..6739f2118ee 100644
--- a/xarray/core/duck_array_ops.py
+++ b/xarray/core/duck_array_ops.py
@@ -253,8 +253,7 @@ def astype(data, dtype, *, xp=None, **kwargs):
     if xp is None:
         xp = get_array_namespace(data)
 
-    if xp == np:
-        # numpy currently doesn't have a astype:
+    if xp is np or not hasattr(xp, "astype"):
         return data.astype(dtype, **kwargs)
     return xp.astype(data, dtype, **kwargs)
 
diff --git a/xarray/tests/test_duck_array_ops.py b/xarray/tests/test_duck_array_ops.py
index ca03c7e67a1..8fa3641c796 100644
--- a/xarray/tests/test_duck_array_ops.py
+++ b/xarray/tests/test_duck_array_ops.py
@@ -43,6 +43,7 @@
     raise_if_dask_computes,
     requires_bottleneck,
     requires_cftime,
+    requires_cupy,
     requires_dask,
     requires_pyarrow,
 )
@@ -184,6 +185,20 @@ def test_where_extension_duck_array(self, categorical1, categorical2):
             where_res == pd.Categorical(["cat1", "cat1", "cat2", "cat3", "cat1"])
         ).all()
 
+    @requires_cupy
+    def test_where_cupy_duck_array(self):
+        import cupy as cp
+
+        arr = cp.array([[cp.nan, cp.nan], [2, 3], [4, 5]])
+        mask = ~cp.isnan(arr)
+        da = DataArray(arr, dims=("x", "y"), name="example")
+        output = da.where(mask, 0)
+
+        expected = np.array([[0, 0], [2, 3], [4, 5]])
+
+        assert isinstance(output.data, cp.ndarray)
+        assert_array_equal(output.to_numpy(), expected)
+
     def test_concatenate_extension_duck_array(self, categorical1, categorical2):
         concate_res = concatenate(
             [PandasExtensionArray(categorical1), PandasExtensionArray(categorical2)]