diff --git a/python/ray/serve/_private/constants.py b/python/ray/serve/_private/constants.py
index 3b10f4dacdeb..cefa321e3d3f 100644
--- a/python/ray/serve/_private/constants.py
+++ b/python/ray/serve/_private/constants.py
@@ -437,11 +437,6 @@
     "RAY_SERVE_MIN_HANDLE_METRICS_TIMEOUT_S", 10.0
 )
 
-# Feature flag to always run a proxy on the head node even if it has no replicas.
-RAY_SERVE_ALWAYS_RUN_PROXY_ON_HEAD_NODE = get_env_bool(
-    "RAY_SERVE_ALWAYS_RUN_PROXY_ON_HEAD_NODE", "1"
-)
-
 # Default is 2GiB, the max for a signed int.
 RAY_SERVE_GRPC_MAX_MESSAGE_SIZE = get_env_int(
     "RAY_SERVE_GRPC_MAX_MESSAGE_SIZE", (2 * 1024 * 1024 * 1024) - 1
diff --git a/python/ray/serve/_private/proxy_state.py b/python/ray/serve/_private/proxy_state.py
index e0f150aa9b85..1af2c0436df7 100644
--- a/python/ray/serve/_private/proxy_state.py
+++ b/python/ray/serve/_private/proxy_state.py
@@ -21,7 +21,6 @@
     PROXY_HEALTH_CHECK_TIMEOUT_S,
     PROXY_HEALTH_CHECK_UNHEALTHY_THRESHOLD,
     PROXY_READY_CHECK_TIMEOUT_S,
-    RAY_SERVE_ALWAYS_RUN_PROXY_ON_HEAD_NODE,
     RAY_SERVE_ENABLE_TASK_EVENTS,
     REPLICA_STARTUP_SHUTDOWN_LATENCY_BUCKETS_MS,
     SERVE_LOGGER_NAME,
@@ -691,10 +690,6 @@ def update(self, proxy_nodes: Set[NodeId] = None) -> Set[str]:
         if proxy_nodes is None:
             proxy_nodes = set()
 
-        # Ensure head node always has a proxy (unless FF'd off).
-        if RAY_SERVE_ALWAYS_RUN_PROXY_ON_HEAD_NODE:
-            proxy_nodes.add(self._head_node_id)
-
         target_nodes = self._get_target_nodes(proxy_nodes)
         target_node_ids = {node_id for node_id, _, _ in target_nodes}
 
diff --git a/python/ray/serve/tests/unit/test_proxy_state.py b/python/ray/serve/tests/unit/test_proxy_state.py
index d5e76abe9146..d3f88b98968a 100644
--- a/python/ray/serve/tests/unit/test_proxy_state.py
+++ b/python/ray/serve/tests/unit/test_proxy_state.py
@@ -203,7 +203,7 @@ def test_proxy_state_manager_restarts_unhealthy_proxies(all_nodes):
 
     cluster_node_info_cache.alive_nodes = all_nodes
     # First iteration, refresh state
-    proxy_state_manager.update()
+    proxy_state_manager.update(proxy_nodes={HEAD_NODE_ID})
 
     prev_proxy_state = proxy_state_manager._proxy_states[HEAD_NODE_ID]
     # Mark existing head-node proxy UNHEALTHY
@@ -212,7 +212,7 @@ def test_proxy_state_manager_restarts_unhealthy_proxies(all_nodes):
 
     # Continuously trigger update and wait for status to be changed to HEALTHY.
     for _ in range(1):
-        proxy_state_manager.update(proxy_nodes=set(HEAD_NODE_ID))
+        proxy_state_manager.update(proxy_nodes={HEAD_NODE_ID})
         # Advance timer by 5 (to perform a health-check)
         timer.advance(5)
 
@@ -415,7 +415,7 @@ def test_proxy_manager_update_proxies_states(all_nodes, number_of_worker_nodes):
     node_ids = [node_id for node_id, _, _ in all_nodes]
 
     # No target proxy nodes
-    proxy_nodes = set()
+    proxy_nodes = {HEAD_NODE_ID}
 
     # Head node proxy should continue to be HEALTHY.
     # Worker node proxy should turn DRAINING.
@@ -533,7 +533,7 @@ def test_proxy_actor_manager_removing_proxies(all_nodes, number_of_worker_nodes)
     N = 10
     for _ in range(N):
         manager.update(
-            proxy_nodes=set(),
+            proxy_nodes={HEAD_NODE_ID},
         )
         timer.advance(5)
         # Assert that
@@ -550,7 +550,7 @@ def test_proxy_actor_manager_removing_proxies(all_nodes, number_of_worker_nodes)
     # Reconcile proxies with empty set of target nodes (worker node proxy
     # will be shutdown by now)
     manager.update(
-        proxy_nodes=set(),
+        proxy_nodes={HEAD_NODE_ID},
     )
 
     assert len(manager._proxy_states) == 1