diff --git a/litellm/integrations/prometheus.py b/litellm/integrations/prometheus.py
index c09b91a713d..08a7bfa49b3 100644
--- a/litellm/integrations/prometheus.py
+++ b/litellm/integrations/prometheus.py
@@ -885,6 +885,30 @@ class PrometheusLogger(CustomLogger):
             print_verbose(f"Got exception on init prometheus client {e}")
             raise e
 
+        self._restore_effective_concurrency_ceiling()
+
+    def _restore_effective_concurrency_ceiling(self) -> None:
+        """Set the ceiling gauge to the value already in force.
+
+        Startup publishes the ceiling, but ``success_callback: ["prometheus"]``
+        builds this logger lazily on the first request, long after that ran with
+        no logger to publish to. The gauge would then register at its zero
+        default and stay there, reading as "no requests allowed" on a proxy
+        serving everything, which is the reading ``+Inf`` exists to prevent.
+
+        Set directly rather than through the module helper, because during
+        ``__init__`` this instance is not yet reachable by the logger lookup.
+        """
+        try:
+            from litellm.proxy.common_utils.request_pressure_metrics import effective_global_limit
+            from litellm.proxy.proxy_server import general_settings
+
+            self.set_global_max_parallel_requests_limit(
+                effective_global_limit(general_settings.get("global_max_parallel_requests"))
+            )
+        except Exception as e:  # noqa: BLE001  # a logger that cannot read proxy config must still start
+            print_verbose(f"could not publish the concurrency ceiling on init: {e}")
+
     def _parse_prometheus_config(self) -> dict[str, list[str]]:
         """Parse prometheus metrics configuration for label filtering and enabled metrics"""
         import litellm
diff --git a/tests/test_litellm/proxy/common_utils/test_request_pressure_metrics.py b/tests/test_litellm/proxy/common_utils/test_request_pressure_metrics.py
index 7d2c3d663ac..c9d2887f80b 100644
--- a/tests/test_litellm/proxy/common_utils/test_request_pressure_metrics.py
+++ b/tests/test_litellm/proxy/common_utils/test_request_pressure_metrics.py
@@ -104,3 +104,32 @@ def test_publishing_never_blocks_startup():
         ),
     ):
         publish_global_max_parallel_requests(3)
+
+def test_a_lazily_built_logger_publishes_the_ceiling_already_in_force(monkeypatch):
+    """success_callback: ["prometheus"] builds the logger on the first request,
+    long after startup published the ceiling with no logger to publish to. The
+    gauge would register at its zero default and stay there, reading as "no
+    requests allowed" on a proxy serving everything."""
+    from prometheus_client import REGISTRY
+
+    import litellm
+    import litellm.proxy.proxy_server as proxy_server
+    from litellm.integrations.prometheus import PrometheusLogger
+
+    monkeypatch.setattr(litellm, "callbacks", [])
+    monkeypatch.setattr(litellm, "success_callback", [])
+    monkeypatch.setitem(proxy_server.general_settings, "global_max_parallel_requests", 7)
+    monkeypatch.setattr(
+        "litellm.proxy.common_utils.request_pressure_metrics.is_global_limit_enforced",
+        lambda: True,
+    )
+    for collector in list(REGISTRY._collector_to_names.keys()):
+        try:
+            REGISTRY.unregister(collector)
+        except Exception:
+            pass
+
+    PrometheusLogger()
+
+    assert REGISTRY.get_sample_value("litellm_global_max_parallel_requests_limit") == 7.0
+