diff --git a/pandas/core/dtypes/cast.py b/pandas/core/dtypes/cast.py
index fafa1f51c823e..35737f28da830 100644
--- a/pandas/core/dtypes/cast.py
+++ b/pandas/core/dtypes/cast.py
@@ -924,45 +924,31 @@ def maybe_infer_dtype_type(element):
 
 
 def maybe_upcast(
-    values: ArrayLike,
+    values: np.ndarray,
     fill_value: Scalar = np.nan,
-    dtype: Dtype = None,
     copy: bool = False,
-) -> Tuple[ArrayLike, Scalar]:
+) -> Tuple[np.ndarray, Scalar]:
     """
     Provide explicit type promotion and coercion.
 
     Parameters
     ----------
-    values : ndarray or ExtensionArray
-        The array that we want to maybe upcast.
+    values : np.ndarray
+        The array that we may want to upcast.
     fill_value : what we want to fill with
-    dtype : if None, then use the dtype of the values, else coerce to this type
     copy : bool, default True
         If True always make a copy even if no upcast is required.
 
     Returns
     -------
-    values: ndarray or ExtensionArray
+    values: np.ndarray
         the original array, possibly upcast
     fill_value:
         the fill value, possibly upcast
     """
-    if not is_scalar(fill_value) and not is_object_dtype(values.dtype):
-        # We allow arbitrary fill values for object dtype
-        raise ValueError("fill_value must be a scalar")
-
-    if is_extension_array_dtype(values):
-        if copy:
-            values = values.copy()
-    else:
-        if dtype is None:
-            dtype = values.dtype
-        new_dtype, fill_value = maybe_promote(dtype, fill_value)
-        if new_dtype != values.dtype:
-            values = values.astype(new_dtype)
-        elif copy:
-            values = values.copy()
+    new_dtype, fill_value = maybe_promote(values.dtype, fill_value)
+    # We get a copy in all cases _except_ (values.dtype == new_dtype and not copy)
+    values = values.astype(new_dtype, copy=copy)
 
     return values, fill_value